JP2005141458A - Image processing method and apparatus, program, and recording medium - Google Patents

Abstract

When searching for a region similar to a specific template from a search target image, the search is performed with high accuracy even if the region is deformed.
A template processing means 3 creates a plurality of sub-templates having a plurality of position coordinates selected from a template as reference points. The search control means 8 places the reference point of the sub template at each position coordinate of the search target image and calculates the degree of matching between the sub template and the search target image. The score voting means 11 calculates the positional relationship between the point corresponding to the reference point of the sub template and the reference point of the corresponding template based on the assumed coordinate conversion parameter, and determines the degree of matching for the reference point of the template corresponding to each coordinate conversion parameter. Is added to the score to obtain the total value. The peak searching means 13 specifies a coordinate conversion parameter that maximizes the total value for each position coordinate of the search target image, obtains the maximum value, and searches for the position of the template reference point at which this value is maximum.
[Selection] Figure 1

Description

  The present invention relates to an image processing method, an apparatus, a program, and a recording medium for searching for a region similar to a specific minute image (template) from search target images, and more particularly to a corresponding region in the search target image. The present invention relates to a device that makes it possible to perform a search with high accuracy even when is deformed.

A technique for searching a template from an image is to calculate the degree of coincidence between the image area and the template at each position while shifting the template in the image and record it in a table. There is known a method for searching the position of (Non-patent Document 1). There are various evaluation functions for the degree of coincidence used at this time, and for example, normalized cross-correlation (hereinafter referred to as normalized correlation) is known. In addition, conventional recognition processing techniques include those disclosed in Non-Patent Document 2 and Non-Patent Document 3 below.
Makoto Nagao, “Digital Image Processing”, Modern Science, 1995, pp. 306-310 Toru Wakahara, “Detection of displacement between images using local affine transformation”, IEICE Journal D-II, The Institute of Electronics, Information and Communication Engineers, December 1989, Vol. J72-D-II, pp. 2070-2079 Wakabayashi Satoshi, Tsuruoka Shinji, Kimura Fumitaka, Miyake Koji, “High accuracy of handwritten digit recognition by increasing the number of dimensions of features”, IEICE Transactions, Institute of Electronics, Information and Communication Engineers, October 1994, Vol. J77-D-II, No. 10, pp. 2046-2053

  In the prior art, when a template is overlaid on a corresponding region in an image to be searched, it is assumed that all corresponding pixels match, and when this premise is satisfied, high search accuracy is obtained.

  However, when the corresponding region is deformed, there is a problem in that both are displaced and search accuracy is lowered. For example, a square template as shown in FIG. 41 and a search target image as shown in FIG. 42 are assumed. In the search target image shown in FIG. 42, it is assumed that the template shown in FIG. 41 is enlarged 1.2 times and further rotated 20 degrees in the right direction (negative direction). When such a deformation occurs, when a search as shown in FIG. 43 (a) is performed according to the conventional technique, as shown in FIG. 43 (b), a shift occurs even if both overlap, and the degree of coincidence is Lower. Thus, according to the prior art, there is a problem that the search accuracy is lowered when the corresponding region in the image to be searched is deformed.

  In addition, the technique disclosed in Non-Patent Document 2 is based on the premise that a specific region to be searched can be extracted from the search target image, and does not touch on a recognition method from the entire search target image. .

  Furthermore, the technique described in Non-Patent Document 3 has a problem that the search system is lowered when an extreme change in the algorithm or a strong rotation is applied.

  The present invention has been made in view of such circumstances, and even when a region similar to a specific template is searched from a search target image, the corresponding region in the search target image is deformed. The object is to provide a new image processing technique that enables high-precision search.

Therefore, the image processing method and apparatus of the present invention search for a region having a high degree of matching with a template that may exist in the search target image.
(1) Using a plurality of different positions selected from the template as reference points, creating a plurality of sub-templates with relaxed matching conditions at points other than the reference points, and (2) searching for reference points of the sub-templates Place it at each position coordinate of the target image, calculate the degree of matching between the sub-template and the search target image,
(3) obtaining a positional relationship between a point corresponding to a reference point of the sub-template and a corresponding reference point of the template when the template is deformed based on a plurality of coordinate transformation parameters assumed in advance;
(4) Obtain a total value obtained by adding the degree of matching calculated for all the sub-templates to the score of the reference point of the template corresponding to each coordinate conversion parameter,
(5) For each position coordinate of the search target image, specify a coordinate conversion parameter that maximizes the sum value, and determine the maximum value;
(6) In the search target image, by searching for the position of the reference point of the template whose maximum value is the peak (maximum),
Even when the search target image is deformed, it is possible to obtain a region having a high degree of matching with the template in the search target image.

As the image processing apparatus of the present invention,
(1) For each of a plurality of reference points selected from the template, sub-template creation means for creating a sub-template from the image composition of the neighboring area;
(2) Voting table means having a memory area for recording a score for a combination of position coordinates and a plurality of coordinate conversion parameters assumed in advance;
(3) With all sub-templates as processing targets, each sub-template is placed at each position coordinate of the search target image to calculate the degree of matching with the search target image, and for each coordinate conversion parameter, the coordinate conversion parameter and its The template position coordinate pointed to by the position coordinate of the reference point of the sub template in the arrangement is calculated, and the calculated matching degree is added to the score of the memory area pointed to by the coordinate conversion parameter and the calculated template position coordinate point. Sub-template search means;
(4) There is a template search unit that searches a search target image for a region having a high degree of matching with a template based on the score of the addition result obtained by the sub-template search unit.

  In this image processing apparatus, the template is transformed by coordinate transformation according to the coordinate transformation parameter in the position coordinate of the region having a high degree of matching as a search result, and the transformed template and the search target image at the position coordinate are transformed. You may provide the matching degree calculation means which calculates a matching degree.

  At this time, the sub-template creation unit is configured to display image information of a neighboring area in which the degree of blur increases as the distance from the reference point of the sub template increases, or a neighboring area that exhibits a uniform blur degree regardless of the distance. In some cases, processing is performed so as to create a sub-template from the dry information.

  Then, the sub-template search means creates image information in which the degree of blurring increases as the distance from the position overlapping the reference point of the sub-template increases as the search target image for which the degree of matching with the sub-template is calculated. Alternatively, image information indicating a single blur level regardless of the distance may be created, and processing may be performed to calculate the degree of matching between the image information and the sub-template.

  In addition, the sub-template search means outputs a position coordinate of a region with a high degree of matching that causes search ligation, and outputs a score and a coordinate conversion parameter at the position coordinate in addition to outputting the position coordinate. There are things to do.

  Each of the processing means described above can be realized by a computer program, and this computer program can be provided by being recorded on a recording medium such as a semiconductor memory or via a network.

  Further, in the image processing apparatus configured as described above, before entering the search process, for each of a plurality of reference points selected from the template, a plurality of sub-templates are created by creating a sub-template from the image information of the neighboring area. It is preferable to create a voting table means having a memory area for recording a score with respect to a combination of position coordinates and a plurality of coordinate transformation parameters assumed in advance, while creating a template (a sub template in a divided form).

  When entering the search process after completing this preparation, each sub-template is selected in turn, the selected sub-template is placed at each position coordinate of the search target image, and the degree of matching with the search target image is calculated, For the calculated degree of matching, for each of a plurality of coordinate conversion parameters assumed in advance, a template position coordinate indicated by a combination of the coordinate conversion parameter and the position coordinate of the sub-template reference point in the arrangement is calculated and calculated. The degree of matching is added to the score of the memory area indicated by the combination of the coordinate conversion parameter and the calculated template position coordinate.

  When the addition process is completed, a region having a high degree of matching with the template is searched from the search target image based on the score of the addition result, and the position coordinates are output or the position coordinates are output. In addition to the above, a score and coordinate conversion parameters at the position coordinates are output.

  At this time, further, the template is coordinate-transformed according to the coordinate transformation parameters in the position coordinates of the region having a high degree of matching as a search result, and the matching with the search target image in the position coordinates with the deformed template is performed. The degree may be calculated and output.

  As described above, the image processing apparatus of the present invention creates a plurality of sub-templates by dividing a template (may be divided in an overlapping form) without performing various operations to deform the template. The degree of matching between the sub-template and the search target image is calculated, and the calculation result is accumulated in the position coordinates of the template when it is assumed that the transformation is performed by a plurality of coordinate conversion parameters assumed in advance. By performing addition, the template is transformed in various ways, and processing equivalent to the processing for calculating the degree of matching is executed. Based on the addition result, the degree of matching with the template is high among the search target images. It is processed so as to search for an area.

  Therefore, according to the present invention, even when the corresponding region in the search target image is deformed, a region similar to the template can be searched with high accuracy, and the template is deformed for searching. The search can be executed with a small processing load compared to.

  When performing this search, considering the case where the corresponding area in the search target image is deformed, the sub template is created in a blurred form, and the matching target with the sub template is calculated accordingly. It is preferable to process so as to create a search contrast image to be blurred.

  In addition, when the template features become noticeable by performing the prescribed image conversion, the image information obtained by performing the image conversion on the image information of the search reference image is searched as the search target image to be processed. It is preferable to process so that it may process.

  For example, in the case where the feature of the template becomes prominent by performing differentiation, the search processing is performed using the image information obtained by performing differentiation on the image information of the search target image as the search target image to be processed. Well, if the template features become prominent by logarithmically converting the luminance value of each pixel, the image information obtained by logarithmically converting the luminance value of each pixel of the search target image is processed. It is good to process so that search processing is performed as a search target image, and the histogram is obtained by quantizing the edge direction obtained from the result of differentiation of the template features into a finite direction and weighting the edge strength obtained simultaneously. If the image becomes prominent by creating it in a template flock divided into an arbitrary number, the image information of the search target image is subjected to differentiation processing, Better to such a search process performs histogram created as a search target image to be processed in the same means as the template to the appropriate area for comparing the.

  Further, as the coordinate transformation, a geometric transformation is usually used. In addition to using one geometric transformation as the geometric transformation, in consideration of the deformation form of the corresponding region in the search target image, a plurality of geometric transformations are used. It is also possible to use a combination of geometric transformations.

  For example, it is possible to use geometric transformations such as enlargement / reduction transformation, rotation transformation and skew transformation, a combination of these geometric transformations, or geometric transformations such as affine transformation and projective transformation.

  Furthermore, in consideration of the deformation form of the corresponding region in the search target image, when the geometric transformation used at this time has a plurality of coordinate transformation parameters, some of the coordinate transformation parameters are fixed. You may process so that a search process may be performed.

Next, the image processing apparatus of the present invention
For each of a plurality of reference points selected from the template, a sub-template creation means for creating a sub-template from image information of the neighboring area and a coordinate conversion parameter to be assumed when the template is arranged at each coordinate position of the search target image Each time, the position coordinates of the reference point of each sub-template is determined, the total value of the degree of matching with the search target image for all sub-templates is calculated, and for each position coordinate of the search target image where the template is arranged A specifying means for specifying the maximum value of the sum total value and the coordinate transformation parameter that realizes the maximum value;
Based on the identification result, template search means for searching a region having a high degree of matching with the template from the search target image may be provided.

  In this image processing apparatus, the template is transformed by coordinate transformation according to the coordinate transformation parameter in the position coordinate of the region having a high degree of matching as a search result, and the transformed template and the search target image at the position coordinate are transformed. You may provide the matching degree calculation means which calculates and outputs a matching degree.

  At this time, the sub template creation means creates the sub template by blurring the image information of the neighboring area when creating the sub template from the image information in the neighboring area of the plurality of reference points selected from the template. Good to process.

  Further, the specifying means determines the position coordinates of the reference point of each sub template for each assumed coordinate conversion parameter and calculates the total value of the matching degrees with the search target images for all sub templates. Processing may be performed so as to calculate the total value of the degree of matching with the search target image subjected to blurring.

  Further, the specifying means uses the table means for recording the total maximum value in association with the position coordinates of the search target image and the coordinate conversion parameter for realizing the maximum value, and the position of the search target image where the template is arranged. For each coordinate, processing may be performed to specify the maximum sum value and the coordinate conversion parameter that realizes the maximum value.

  Here, when the template features become noticeable by performing the prescribed image conversion, the search processing is performed using the image information obtained by performing the image conversion on the image information of the search target image as the processing target image. It is preferable to process so that it may perform.

  For example, in the case where the feature of the template becomes prominent by performing differentiation, the search processing is performed using the image information obtained by performing differentiation on the image information of the search target image as the processing target image information. Preferably, the image is obtained by logarithmically converting the luminance value of each pixel of the search target image when the template feature becomes prominent by logarithmically converting the luminance value of each pixel. It is preferable to perform a search process using information as a search target image to be processed, and the edge direction obtained from the result of differentiation of the template features is quantized into a finite direction, and the edges obtained simultaneously If it becomes prominent by creating a histogram weighted with the strength of the template in the block of the template divided into an arbitrary number, the image of the search target image Distribution in performing differentiation processing, better to perform the search processing histogram created as a search target image to be processed in the same means as the template to the appropriate area for comparison with the template.

  Further, as the coordinate transformation, a geometric transformation is usually used. In addition to using one geometric transformation as the geometric transformation, in consideration of the deformation form of the corresponding region in the search target image, a plurality of geometric transformations are used. It is also possible to use a combination of geometric transformations.

  For example, it is possible to use geometric transformation such as enlargement / reduction transformation, rotation transformation or skew transformation, or a combination of these, or geometric transformation such as affine transformation or projective transformation.

  Furthermore, in consideration of the deformation form of the corresponding area in the search target image, when the geometric transformation used at this time has a plurality of coordinate transformation parameters, some of the coordinate transformation parameters are fixed. You may process so that a search process may be performed.

  Each of the processing means described above can be realized by a computer program, and this computer program can be provided by being recorded on a recording medium such as a semiconductor memory or via a network.

  In the image processing apparatus configured as described above, before entering the search process, for each of a plurality of reference points selected from the template, a plurality of sub-templates ( It is preferable to create a sub template in a divided form.

  After completing the creation of these sub-templates, when entering the search process, the templates are arranged in order at each position coordinate of the search target image, and the assumed coordinate conversion parameters are selected in order from each of these constellations. Then, based on the arranged position coordinates and the selected coordinate conversion parameter, the position coordinates of the reference point of each sub template are determined, and the sum of the matching degrees with the search target images for all the sub templates is calculated.

  Then, by specifying the maximum value of the calculated total value and specifying the coordinate conversion parameter that realizes the maximum flight, the calculated total value for each position coordinate of the search target image in which the template is arranged And the coordinate transformation parameter that realizes the maximum value.

  When the identification processing is completed, based on the identification result, an area having a high degree of matching with the template is searched from the search target image, and the position coordinates are output or added to the output of the position coordinates. Thus, the maximum sum value at the position coordinates and the coordinate conversion parameters for realizing the maximum value are output.

  In this way, the template placement position is set as the vote destination, and the total value of the degree of matching between each sub template and the search target image when the coordinate conversion parameter at the placement position is changed is set as the vote destination. If the maximum value is voted as a score and a table for recording the score is prepared, a table for recording the maximum value of the total value in association with the position coordinates of the search target image and a coordinate conversion parameter for realizing the maximum value is prepared. Since it is sufficient, it is sufficient to prepare a small table for score recording.

The image processing apparatus of the present invention is
For each of a plurality of reference points selected from the template, sub-template creation means for creating a plurality of sub-templates by blurring image information according to the reference points;
When placing the template at each position coordinate of the search target image, the position coordinates of the reference point of each sub template are determined for each assumed coordinate conversion parameter, and the degree of matching with the search target image for all the sub templates is determined. For each position coordinate of the search target image in which the template is arranged, the specifying means for specifying the maximum value of the total value and the coordinate conversion parameter that realizes the maximum value;
Based on this identification result, template search means for searching a region having a high degree of matching with the template from the search target image may be provided.

  In the present invention, the template is transformed by coordinate transformation according to the coordinate transformation parameters in the position coordinates of the region having a high degree of matching as a search result, and the degree of matching between the transformed template and the search target image at the position coordinates. There may be provided a matching degree calculating means for calculating and outputting.

  In this configuration, the specifying unit determines the position coordinates of the reference point of each sub template for each assumed coordinate conversion parameter, and calculates the total value of the degree of matching with the search target image for all sub templates. At the time of calculation, processing may be performed to calculate the total value of the degree of matching with the search target image that has been blurred.

  The specifying unit includes a table unit that records a maximum sum value and a coordinate conversion parameter that realizes the maximum value in association with the position coordinates of the search target image. Every time, processing may be performed to specify the maximum sum value and the coordinate conversion parameter that realizes the maximum value.

  In addition, a degree of matching between the sub-template and the search target image is calculated, and is stored in a memory while being associated with the sleep coordinate of the reference point of the sub-template, and at this time, the specifying unit includes the memory There is a case where processing is performed so as to calculate the total value of the matching degree using the matching degree stored in.

  Here, when the template features become prominent by performing the prescribed image conversion, the image information obtained by performing the image conversion on the image information of the search target image is used as the search target image to be processed. The search process may be performed.

  For example, in the case where the feature of the template becomes prominent by performing differentiation, the search processing is performed using the image information obtained by performing differentiation on the image information of the search target image as the search target image to be processed. If the template features become prominent by logarithmically converting the luminance value of each pixel, an image obtained by logarithmically converting the luminance value of each pixel of the search target image It is better to process the information as an image to be processed so that the search process is performed. Also, the edge direction obtained from the result of differentiation of the template features is quantized into a finite direction, and the edge strength obtained at the same time If it becomes prominent by creating a weighted histogram in a block of a template divided into an arbitrary number, a differentiation process is applied to the image information of the search target image. , It may be to perform the search processing histogram created as a search target image to be processed in the same way as the corresponding region template for comparing the template.

  In general, geometric transformation is used as the coordinate transformation. In addition to using one geometric transformation as the geometric transformation, a plurality of geometric transformations are considered in consideration of the deformation form of the corresponding region in the search target image. It is also possible to use a combination of these.

  For example, it is possible to use geometric transformation such as enlargement / reduction transformation, rotation transformation, skew transformation, or a combination of these, or geometric transformation such as affine transformation or projective transformation.

  Furthermore, in consideration of the deformation mode of the corresponding area in the search target image, when the geometric transformation used at this time has a plurality of coordinate transformation parameters, some of the coordinate transformation parameters are fixed and searched. You may process so that a process may be performed.

  Each of the processing means described above can be realized by a computer program, and this computer program can be provided by being recorded on a recording medium such as a semiconductor memory or via a network.

  In the image processing apparatus of the present invention configured as described above, before entering the search process, for each of a plurality of reference points selected from the template, the image information is blurred according to the reference points, thereby substituting a plurality of sub-points. A template (a sub template in which part of image information of the template is blurred) may be created.

  After completing the creation of these sub-templates, when entering the search process, the templates are arranged in order at the position coordinates of the search-to-produce image, and the assumed coordinate conversion parameters are selected in order at these position coordinates. Then, based on the arranged position coordinates and the selected locus conversion parameters, the position coordinates of the reference point of each sub template are determined, and the total value of the degree of matching with the search target image for all the sub templates is determined. calculate.

  Then, by specifying the maximum value of these calculated total values and specifying the coordinate transformation parameter that realizes the maximum value, the calculated total value for each position coordinate of the search target image in which the template is arranged And the coordinate transformation parameter that realizes the maximum value.

  When the identification process is completed, based on the identification result, an area having a high degree of matching with the template is searched from the search target image, and the position coordinates are output, or in addition to the output of the position coordinates, The maximum sum value at the position coordinates and a coordinate conversion parameter that realizes the maximum value are output.

  In this way, the template placement position is set as a vote destination, and the total value of the degree of matching between each sub template and the search target image when the coordinate conversion parameter at the placement position is changed is set as the vote destination. If the maximum value is voted as a score and a table for recording the score is prepared, a table for recording the maximum value of the total value in association with the position coordinates of the search target image and a coordinate conversion parameter for realizing the maximum value is prepared. Since it is sufficient, it is sufficient to prepare a small table for score recording.

  In addition, the maximum value of the total sum of the degree of matching with the search target image is obtained by using a sub template in which a part of the template is blurred. Compared to the case where a sub template configured by dividing the template is used. Thus, the maximum sum value of the degree of matching with the search target image is obtained while leaving the image information of the template, so that a more accurate search process can be realized.

  Therefore, according to the image processing method, apparatus, program, and recording medium of the present invention, when a region similar to a specific template is searched from the search target image, the corresponding region in the search target image is deformed. Even in such a case, a region similar to the template can be searched with high accuracy.

  Further, the search can be executed with a small processing load as compared with the case where the search is performed by deforming the template.

  Moreover, in realizing this, a method for realizing a high-accuracy search when the deformation of the corresponding area in the search target image is large, a search method suitable for giving priority to memory reduction, and giving priority to memory reduction. Various realization methods such as a method for realizing a search for a high sugar content can be provided.

  Embodiments of the present invention will be described with reference to the drawings.

(Example 1)
FIG. 1 is a schematic diagram of an image processing apparatus according to the first embodiment.

  The image processing apparatus according to this embodiment includes a voting coordinate information calculation unit 1, a voting coordinate information storage unit 2, a template processing unit 3, an image blurring unit 4, a template storage unit 5, and an input image processing unit 6. , Processed image storage means 7, search control means 8, matched pattern generation means 9, inter-pattern matching means 10, score voting means 11, score recording means 12, and peak search means 13. .

  The detailed operation of each of these processing means will be described.

  Here, the description of this operation can be divided into two modes: a mode for processing a template and a mode for searching for an image using the processed template. In the description of this assistance, the template shown in FIG. 41 is assumed, and the search target image (input image) shown in FIG. 42 is assumed.

(I) Example of operation in a mode for processing a template In a mode for processing a template, the template processing means 3 mainly operates. First, the template processing means 3 selects a plurality of local areas from the template. (Each area is called a floc). Any number of these blocks may be selected from the template, the size may be arbitrarily set, and the blocks may partially overlap each other.

  FIG. 2 shows an example of a block selected from the template.

  In this example, the template shown in FIG. 41 (hereinafter referred to as template 0) is divided into four to create four blocks.

  For convenience of explanation, the size of the template is represented by 4L × 4L. Therefore, the size of each block is 2L × 2L. In this example, the template has a square shape, but this shape may be any shape (rectangle, circle, ellipse, etc.).

  Next, the template processing unit 3 passes the template 0 to the image blurring unit 4. In response to this, the image blurring means 4, blurs the template 0 with a plurality of blurring degrees and returns it to the template processing means 3.

  As a specific blurring method, for example, there is a method of using a smoothing filter by convolving a two-dimensional matrix. Examples of the smoothing filter include an image analysis handbook (supervised by Takagi and Shimoda, 1991, University of Tokyo Press). 539) of local average filters and local head average filters.

  In the case of a local weighting filter, the degree of blurring can be changed by changing the degree of change in weight depending on the distance from the center.

  That is, in the case of these matrixes being enlarged, or in the case of a local weighted average filter, the degree of blurring can be increased by reducing the degree of weight reduction due to the distance from the center.

  3A to 3E show examples of blurred images in which the degree of blurring is changed in five stages.

  Thereafter, the template processing means 3 takes out each block shown in FIG. 2 one by one and performs the following processing.

  That is, taking block 1 as an example, first, the block area is divided as shown in FIG. For each region, the larger the distance from the center, the larger the distance from the center, as shown in FIG. 3G, by copying the corresponding surface element value of the blurred image having a higher degree of blurring. Create a blurred image with a greater degree of blurring. This is template 1.

  In the same manner, the blocks 2 to 4 are also processed to create templates 2 to 4 as shown in FIG.

  Next, when using a histogram feature for each direction as a template feature, the templates 1 to 4 are divided into blocks of several pixels, and within the divided blocks as shown in FIGS. Create a density histogram by edge direction. This is called templates 5-8. Templates 5 to 8 are created for each template in the number of quantized directions.

  Then, the template processing means 3 sends these templates 1 to 4 or 5 to 8 to the template storage means 5 and stores them. For the subsequent processing, when using the histogram feature for each direction, templates 5 to 8 are used instead of templates 1 to 4.

  Next, the template processing unit 3 sends the reference position of the template 0 and the reference positions of the templates 1 to 4 to the voting coordinate calculation unit 1.

  Here, as an example, as shown in FIG. 6, the reference position (reference point) of template 0 is the center point, and the reference positions (reference points) of templates 1 to 4 are the center points.

  When the voting coordinate information calculation means 1 receives the positions of the reference points of the templates 0 to 4 from the template processing means 3, when the template 1 is first placed at an arbitrary position in the search target image, the template 1 It is calculated which coordinate has a possibility that the relative position of the point corresponding to the reference point of the template 0 from the reference point exists due to the deformation of the region expressed by the assumed coordinate transformation.

  In the present embodiment, it is assumed that an example of deformation is expressed by coordinate transformation that combines enlargement / reduction and rotation.

  When no deformation occurs in the corresponding area in the search target image, the corresponding area should completely overlap with the template 0. In the coordinate system with the reference point of the template 0 at that time as the origin, The position of the point corresponding to the reference point of the template 1 is (−L, L), but when deformation occurs, if (x ′, y ′) after conversion is expressed, (x ′, y ′) is These are represented by the following formulas (1) and (2).

x ′ = − aL cos γ−a L sin γ (1)
y ′ = − aLsinγ + aLcosγ (2)
Here, a is the enlargement / reduction ratio, and γ is the rotation angle.

  On the other hand, since the reference point of template 0 is the origin of the coordinate system, even if no deformation occurs, it is (0, 0). Here, when the position of the reference point of template 0 when (x ′, y ′) is the origin is represented by (Δx, Δy), it can be represented by the following equations (3) and (4).

Δx = −x ′ = aLcosγ + aLsinγ (3)
Δy = −y ′ = aLsinγ + aLcosγ (4)
That is, when (Δx, Δy) expressed by the equations (3) and (4) is deformed by the parameter values a and γ, the point corresponding to the reference point of the template 1 is used as the origin. When the template 1 is shifted and placed at an arbitrary position in the search target image, the image of the template 1 as a pattern at that time and the region at that position in the search target image The position of the template where the degree of coincidence should be recorded.

  A plurality of (Δx, Δy) are calculated by changing a and γ within a small range within an assumed parameter value range.

  This will be described with reference to the drawings. When m ≦ a ≦ n (0 ≦ m ≦ 1, 1 ≦ n) is assumed and −β ≦ γ ≦ α, a and γ can be changed within this range. Sometimes, the miracle of the position (Δx, Δy) of the template 0 with the reference point of the process 1 as the origin is represented by the hatched portion in FIG.

  The voting coordinate information calculating means 1 performs this calculation in advance to create a table and stores it in the voting coordinate information accumulating means 2.

  In order to illustrate an example of this table, L = 50 (pixels), β = α = 20 (degrees), m = 0.7, n = 1.3 are shown as numerical examples of the values shown so far. Is used. As values for changing a and γ, a is (0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3), and γ is (−20 degrees, −10 degrees, 0 degrees, 10 degrees, 20 degrees).

  FIG. 8 shows a table in which values calculated by the equations (3) and (4) using these numerical values are recorded.

  In the same way, FIG. 9 shows the result of creating a table for template 2, FIG. 10 shows the result of creating a table for template 3, and FIG. 11 shows the result of creating a table for template 4.

  When the voting coordinate information calculating unit 1 creates these tables (hereinafter sometimes referred to as a voting coordinate position table), the voting coordinate information calculating unit 1 accumulates them in the voting chair information storage unit 2.

(Ii) Example of operation in mode for searching image In the description of this operation, the template shown in FIG. 41 is assumed, and the search target image shown in FIG. 42 is assumed. In addition, 4L = 200 (image search) is assumed. In the search target image shown in FIG. 42, an image obtained by enlarging the template 1 shown in FIG. 41 by 1.2 times and rotating it by −20 degrees (that is, 20 degrees to the right) has a center position (X = 3000). (Pixel, Y = 1000 pixels).

  Therefore, the template (template 0) shown in FIG. 41 is enlarged 1.2 times and rotated by −20 degrees, and the reference position is overlapped with this center position (X = 3000 pixels, Y = 1000 image search). Sometimes the degree of matching between patterns should have a peak.

  Here, for template 1, the X coordinate of the reference point is 2964 pixels (the value obtained by adding X = 3000 pixels to x ′ in equation (1)), and the Y coordinate is 1077 pixels (y ′ in y ′ in equation (2)). = The value obtained by adding 1000 strokes), the degree of coincidence should be a peak. Similarly, for template 2, the X coordinate of the reference point is 3077 pixels, and the Y coordinate is 1036. The degree of coincidence should peak when overlaid on the pixel area. The template 3 and the template 4 should have a peak degree of coincidence when they are superimposed on the positions shown in FIG.

  In the image search mode, first, the image blurring unit 4 is used to create a plurality of blurred images having different degrees of blurring with respect to the search target image.

  This blurred image is created by using the same method as described in the above “template processing mode”. FIGS. 13A to 13E show the creation results.

  Next, the processing moves to the search control means 8.

  The search control means 8 first takes out the image of the template 1 ((g) in FIG. 3) from the template storage means 5. Then, the search control means 8 calculates the degree of coincidence at each position while shifting the extracted image of the template 1 to each position of the blurred search target image, and sends the result to the score voting means 11. In response, the score voting means 11 records it in the score recording means 12.

  Here, a procedure for calculating the degree of coincidence when the search control unit 8 shifts the image of the template 1 to one position of the search target image will be described with reference to FIG.

  As an example, as will be described with reference to (f) of FIG. 13, the operation when the position of the reference point of the template 1 is placed so as to be the most coincident coordinates (2964 pixels, 1077 pixels) will be described.

  In this case, the search control means 8 first processes the coordinates of the reference point of the template 1 (2964 pixels, 1077 pixels), the size of the template 1 (100 pixels × 100 pixels in this description), and the input image processing means 6. The plurality of blurred images (original images are search target images) having different degrees of blurring shown in FIGS. 13A to 13E are sent to the matched pattern generation means 9.

  Here, the input image processing unit 6 uses the image blurring unit 4 to convert the search target image into a plurality of blurred images having different blurring degrees, and the processed image storage unit 7 stores the blurred image. Become.

  In response to this, the collation pattern generation means 9 uses the same method as that described in the above-mentioned “mode for processing a template”, and blurs as the distance from the center is the same size as the template 1. Is created and returned to the search control means 8 (an example of the processing result is shown in FIG. 13 (h)).

  Then, the search control unit 8 sends the processing result of the blurred image to the inter-pattern matching unit 10.

  In response to this, the inter-pattern matching unit 10 collates the passed image with the image of the template 1 to calculate the degree of matching between the patterns, and returns the result to the search control unit 8.

  As the degree of matching between patterns, for example, the normalized correlation described in the description of the prior art can be used. Here, for example, “0.8” is calculated.

  Subsequently, the search control means 8 identifies the value returned from the inter-pattern matching means 10 to the score voting means 11 (hereinafter referred to as score; here, “0.8” above) and the template 1. The code and the coordinates of the reference point of the template 1 (in this case (2964 pixels, 1077 pixels)) are sent. Here, as the identification codes of templates 1 to 4, if “1” to “4” are used as an example, “1” is sent in this case. Next, the processing moves to the score voting means 11.

  Prior to the description, a configuration example of the score recording unit 12 is illustrated. This figure shows the initial state before the search is performed.

  As shown in FIG. 14, the score recording means 12 has an array (hereinafter referred to as a score recording table) for recording a score at each coordinate of each position coordinate of the search target image. In this score recording table, comprehensive combinations of parameter values for deformation are registered in advance, and the initial value (in the example of FIG. 14) is displayed in the score column corresponding to each combination in the initial state. 0) is entered.

  Next, the operation of the score voting means 11 will be described.

  The score voting unit 11 identifies the identification code “1” of the template 1 sent from the search control unit 8 and stores the voting coordinate position table (shown in FIG. 8) of the corresponding template 1 in the voting coordinate information storage unit 2. Take out from.

  Then, the coordinates of the reference point of the template 1 sent from the search control means 8 (in this case (2964 exposure element, 1077 pixels)) are added to the numerical values of Δx and Δy recorded in each row of the vote coordinate position table. By adding each, the X coordinate of the throwing destination and the Y coordinate of the voting destination in that row are output (FIG. 15A).

  Here, in FIG. 15A, the voting destination X coordinate and the voting destination Y coordinate are calculated for the row “rotation angle = −20 degrees, enlargement ratio = 1.2” of the voting coordinate position table of the template 1. Although an example is shown, such processing is performed for each row of the voting coordinate position table.

  Next, for each row of the voting coordinate position table, the score voting means 11 has a portion corresponding to the rotation angle and enlargement ratio of the row of the score recording table corresponding to the vote destination X coordinate and the vote destination Y coordinate of the row. The score sent from the search example control means 8 (0.8 in this case) is added to the score column (FIG. 15B).

  Here, in FIG. 15B, “voting destination X coordinate = 3000 pixels, voting destination Y coordinate = 1000 pixels” associated with the row of “rotation angle = −20 degrees, enlargement ratio = 1.2”. Although an example is shown in which a score of 0.8 is added to the corresponding column of the score recording table (the column of “rotation angle = −20 degrees, magnification factor = 1.2”), such processing is performed in the voting coordinate position. This is done for each row in the table.

  Regarding the processing when the template 1 described above is placed at one position of the search target image (inter-pattern matching processing with the template 1 and addition processing to the score recording table as a result), all the positions of the search target image Will be done about.

  The search control means 8 and the score voting means 11 perform the same processing for the remaining templates 2 to 4 after completing the above processing for the template 1.

  FIG. 16 shows the operation of the score voting means 11 in the operation when the template 2 is placed so that the reference position is the closest coordinate (3077 pixels, 1036 lines) as shown in FIG. It is a figure explaining an example.

  For “voting destination X seat image = 3000 pixels, voting destination Y coordinate = 1000 pixels”, the score is also voted here. This is the same for the template 3 and the template 4.

  That is, with respect to template 0, although the size and inclination of the template 0 are significantly different from the corresponding region in the search target image, the template 0 is divided into templates 1 to 4 and collated, so that the reference of template 0 can be obtained. A large score is accumulated at the point (3000 pixels, 1000 both primes).

  Finally, the process proceeds to the peak search means 13.

  FIG. 17 shows an explanatory diagram of the processing of the peak searching means 13.

  As shown in this figure, the peak search means 13 acquires the maximum value of the score from each score recording table (FIG. 17A) in the score recording means 12, and uses the score value as a score map (FIG. 17). (B)), and the deformation parameter values at that time are plotted on the parameter value map (FIG. 17C).

  Then, the peak searching means 13 extracts a peak search technique (for example, a score having a specified threshold value or more from the score map, performs grouping so that adjacent items belong to the same group, Find the peak location by identifying the one with the highest score among them), find the peak location (s), output the coordinates of the location and the score value, and The parameter value is obtained from the corresponding spot in the value map and is also output.

  18 and 19 show the processing flow of the image search mode described above.

  That is, when the image processing apparatus according to the present invention enters the image search mode, first, as shown in this processing flow, first, in step 1, the initial value 1 is set to the variable i which becomes the identification number of the template. In this embodiment, i = 1 to 4 can be taken.

  Subsequently, in step 2, it is determined whether or not the value of the variable i is larger than 4, and when it is determined that the value is not larger, the process proceeds to step 3 where the template storage means 5 stores the template i. In step 4, the image is extracted, and the voting coordinate position table (shown in FIG. 8 from FIG. 8) of the template i is extracted from the voting coordinate information accumulating means 2.

  As will be described below, when it is determined that all positions of the template i are not processed while shifting one pixel at a time in the X direction and the Y direction, the process proceeds to step 7 where the template i is selected from the unprocessed positions. One processing position (the position of the reference point of template i) is selected. For example, the selection is made by shifting one stroke in the X direction and the Y direction.

  Subsequently, in step 8, a blurred image that is a collation target of the template i at the selected processing position is created from the search target image (created by synthesizing a search target image that has been blurred in multiple stages in advance). In the subsequent step 9, the degree of coincidence between the template i and the created blurred image is calculated.

In accordance with the calculation of the degree of coincidence, the process described below is performed for each row of the voting coordinate position table. Subsequently, in step 10, whether all the lines of the voting coordinate position table have been processed. When it is determined that the processing has been completed for all the rows in the voting coordinate position table, the process returns to step 5 in order to calculate the degree of coincidence at the next processing position of the search target image.
On the other hand, when it is determined in step 10 that the processing has not been completed for all the rows of the voting coordinate position table, the process proceeds to step 11 and from the voting coordinate position table (extracted in step 4) of template i, Select one unprocessed row. For example, when selecting rows (entry areas) one by one from the top of the voting coordinate position table (shown in FIGS. 8 to 11), the row below the previously processed row is selected. is there.

  Subsequently, at step 12, the vote destination X coordinate / voting destination Y coordinate is calculated by adding Δx / Δy set to the selected row and the processing position. Subsequently, in step 13, the score recording table indicated by the calculated vote destination X coordinate / voting destination Y coordinate is selected, and the rotation angle set in the selected row among the entries of the selected score recording table. / After adding the calculated degree of coincidence (score) to the entry pointed by the enlargement ratio (recording the score), the process goes to step 10 in order to perform processing for the next row in the vote-destination coordinate position table. Return.

  In this manner, when it is determined that the value of the variable i is larger than 4 in step 2 when the processing of steps 2 to 14 is repeated, the processing is completed for all templates. When this is determined, the process proceeds to step 14, and a score having a value equal to or greater than a prescribed threshold value included in the score recording table is extracted.

  Subsequently, in step 15, the extracted scores are grouped so that adjacent ones belong to the same group, and in step 16, the score indicating the maximum value in the same group is selected as a search result.

  Subsequently, in step 17, by referring to the score recording means 12 (score recording table), the position coordinate / rotation angle / magnification rate pointed to by the score of the search ligation is acquired. Subsequently, in step 18, the template 0 is deformed according to the acquired rotation angle / magnification ratio, and the degree of coincidence with the search target image indicated by the acquired position coordinates is calculated.

  Subsequently, in step 19, the score of the search result selected in step 16, the position coordinates / rotation angle / magnification rate acquired in step 17, and the degree of coincidence calculated in step 18 are output, and the process is terminated. To do.

  In this manner, the image processing apparatus according to the present embodiment divides the template 0 to create, for example, four templates 1 to 4, and between those templates 1 to 4 and the search target image. The degree of matching is calculated, and the result of the calculation is cumulatively added to the position coordinates of template 0 when it is assumed that it has been deformed by a plurality of coordinate transformation parameters assumed in advance, thereby deforming the template in various ways. Then, a process equivalent to the process of calculating the degree of matching is executed, and based on the addition result, processing is performed so as to search for an area having a high degree of matching with the template 0 from the search target image.

  In the embodiment described above, the templates 1 to 4 are configured with a blurred image in which the degree of blur increases as the distance from the center increases, and the search target image to be compared with the templates 1 to 4 As for the image, the blur image increases in the degree of blur as the distance from the center overlaps, but only one of the images may be composed of the blur image. The method may also be configured with a blurred image showing a single degree of blur regardless of distance.

Embodiment 2
FIG. 20 is a schematic diagram of an image processing apparatus according to the second embodiment.

  The image processing apparatus according to the present embodiment includes a voting coordinate information calculation unit 1a, a voting coordinate information storage unit 2a, a template processing unit 3a, an image blurring unit 4a, a template storage unit 5a, and an input image processing unit 6a. Processed image storage means 7a, search control means 8a, matched pattern generation means 9a, inter-pattern verification means 10a, score voting means 11a, score recording means 12a, peak search means 13a, and verification result storage Means 14a.

  The voting coordinate information calculation means 1a performs a process of creating a voting coordinate table. In the first embodiment, the voting coordinate position table shown in FIGS. 8 to 11 according to the equations (3) and (4). Will be created.

  As will be described later, when the embodiment is followed, the reference points of the templates 1 to 4 (positions shown in FIG. 6) are based on the positions of the reference points of the template 0 (positions shown in FIG. 6). Therefore, the voting coordinate information calculating unit 1a calculates the voting coordinate position table according to the above equations (1) and (2).

  From this, the voting coordinate information accumulating means 2a accumulates a voting coordinate position table as shown in FIG. 21 in which the signs of the table data shown in FIGS. 8 to 11 are inverted.

  The template processing unit 3a uses the image blurring unit 4a that performs the same processing as the image blurring unit 4 of the first embodiment, and executes the same processing as the template processing unit 3 of the first embodiment, thereby performing FIG. And templates 1 to 4 shown in FIG. 4 are created.

  Thus, the templates 1 to 4 shown in FIGS. 3 and 4 are stored in the template storage unit 5a.

  The input image processing unit 6a, the processed image storage unit 7a, the collated pattern generation unit 9a, and the inter-pattern verification unit 10a are respectively the input image processing unit 6, the processed image storage unit 7, and the collated pattern generation unit 9 of the first embodiment. In addition, in the case of following the second embodiment, the matching degree (score) between patterns is not sent to the score voting unit 11a each time. Instead, it is stored in the collation result storage unit 14a via the search control unit 8a.

  FIG. 22 illustrates the data structure of the score stored in the collation result storage unit 14a. As shown in this figure, the collation result accumulating means 14a associates each of the templates 1 to 4 with one of the search target images on which the reference points of the templates 1 to 4 are placed, and blurs them at that position. The degree of coincidence (score) with the searched image is stored.

  As shown in FIG. 23, the score recording unit 12a is associated with the position of the search target image where the reference point of the template 0 is placed, and the maximum value of the sum of the scores of the templates 1 to 4 at that position, It is generated by a score recording table that manages deformation parameters (dog expansion / reduction rate a, rotation angle γ) that realize the maximum score total value.

  The score voting means 12a executes data registration in the score recording means 12a, and arranges the reference point of the template 0 at each position of the search target image, and creates a voting coordinate position table (shown in FIG. 21). By referencing, the position of the reference point of the templates 1 to 4 when the deformation parameter is changed at each arrangement position is obtained, and the collation result detecting means 14a (shown in FIG. 22) is obtained using that position as a key. By referencing, the scores of templates 1 to 4 are obtained, and the sum value thereof is calculated.

  Then, when the score voting means 11a calculates the score total value in the form as shown in FIG. 24 in this way, the score total value is larger than the maximum score total value recorded in the score recording table. Is calculated, the data recorded in the score recording table is updated accordingly, thereby registering data in the score recording means 12a.

  That is, the score voting means 11a executes data registration in the score recording means 12a (score recording table) by executing the processing shown in FIG.

  Thus, in the case of following this embodiment, the score recording table constituting the score recording means 12a corresponds to the position of the search target image where the reference point of the template 0 is placed, as shown in FIG. In addition, the maximum value of the sum of the scores of the templates 1 to 4 at the position and the deformation parameter that realizes the maximum score sum value are recorded.

  Therefore, the peak searching means 13a can immediately execute the plot on the score map and the parameter value map as shown in FIG. 17 by referring to the score recording table.

  From now on, the peak search means 13a extracts the peak search technique (for example, the maximum score sum value having a specified threshold value or more from the score map, as in the peak search means 13 of the first embodiment, The location of the peak (several things) by grooving in the form of belonging to the same group, and finding the peak location by identifying the one with the maximum total score value indicating the maximum value in each group) And the coordinates of the position and the maximum score total value are output, and the deformation parameter value is acquired from the corresponding coordinates of the parameter value map, and this is also output.

  In the case of following the first embodiment, since the scores are added by moving the templates 1 to 4 independently, even if the template image portion included in the search target image is complicatedly deformed, it can be detected. However, as shown in FIG. 14, a large memory-based score recording table prepared for each position of the search target image must be used.

  On the other hand, in the case of following the second embodiment, as can be seen from the above description, the templates 1 to 4 are moved in an integrated manner on the basis of the template 0 to obtain the total value of the scores, Therefore, it is sufficient to prepare a score recording table having a small memory procedure as shown in FIG.

  Next, processing in the image search mode executed by the image processing apparatus of the present invention when following the present embodiment will be described according to the processing flow shown in FIGS.

  When entering the image search mode, the image processing apparatus according to the present embodiment first sets an initial value 1 to a variable i serving as a template identification number in step 20, as shown in this processing flow. In this embodiment, i = 1 to 4 can be taken.

  Subsequently, in step 21, it is determined whether or not the value of the variable i is larger than 4, and when it is determined that the value is not larger, the process proceeds to step 22 where the template accumulating means 5a receives the template. Take out the image of i.

  As will be described below, the template i is placed on the search target image while shifting one image at a time in the X direction and the Y direction, and predetermined processing is executed. When it is determined whether or not the process described below for the position has been completed and it is determined that the process has been completed for all positions, the process proceeds to step 24 to set the value of the variable i to process the next template i. After one increment, return to step 21.

  On the other hand, when it is determined in step 23 that all positions of the template i have not been processed, the process proceeds to step 25, where the processing position for placing the reference point of the template i from the unprocessed positions of the search target map is set to 1. Are selected (selected while shifting one pixel at a time in the X and Y directions).

  Subsequently, in step 26, a blurred image that is a collation target of the template i at the selected processing position is created from the search target image (created by synthesizing a search target image that has been blurred in multiple stages in advance). .

  Subsequently, in step 27, the degree of coincidence between the template i and the created blurred image is calculated and registered in the collation ligation accumulation means 14a, and then the degree of coincidence at the next position of the search target image is calculated. Therefore, the process returns to step 23.

  In this way, by repeating the processing of step 22 to step 27, data registration in the collation result storage means 14a having the data structure as shown in FIG. If it is determined that the value of is greater than 4, the process proceeds to step 28 to determine whether or not the processing has been completed for all positions of the search target image.

  That is, as described below, template 0 (division source template) is placed on the search target image while shifting one pixel in the X direction and the Y direction, and a predetermined process is executed. It is determined whether or not all the processes have been completed.

  When it is determined by the determination process in step 28 that the process has not been completed for all positions of the search target image, the process proceeds to step 29, in which the search target places the reference point of the template 0 from the unprocessed positions. One position is selected (selected while shifting one stroke in the X and Y directions).

  Subsequently, in step 30, one unprocessed deformation parameter is selected by selecting one unprocessed line from the vote coordinate position table. For example, when selecting rows one by one from the top of the voting coordinate position table (shown in FIG. 21) (rows common to the four tables) one by one, By selecting, one unprocessed deformation parameter is selected.

  Subsequently, in step 31, by adding Δx / Δy registered in the selected row and the processing position, the position of the reference point of the templates 1 to 4 on the search target image is specified, and the specification is performed. The matching result (score) between the templates 1 to 4 registered in the matching result accumulation unit 14a and the search target image is obtained by referring to the matching result accumulation unit 14a using the determined position as a key. Calculate the sum.

  Subsequently, at step 32, it is determined whether or not the calculated total value is larger than the maximum score total value indicated by the processing position of the selected search target image recorded in the score recording table (shown in FIG. 23). If it is determined that the value is larger, the process proceeds to step 33, and the data recorded in the score recording table is updated (if not registered) in accordance with the calculated total value and the deformation parameter that realizes the sum. New registration).

  On the other hand, when it is determined in step 32 that the calculated total value is smaller than the maximum score total value indicated by the processing position of the selected search target image recorded in the score recording table, the processing in step 33 is omitted. To do.

  Subsequently, in step 34, it is determined whether or not the processing has been completed for all the rows of the voting coordinate position table, and when it is determined that the processing has not been completed for all the rows, In order to perform the process, the process returns to step 30. On the other hand, when it is determined that the process has been completed for all rows, the process returns to step 28 in order to perform the process for the processing position of the next search target image.

  In this way, by repeating the processing from step 29 to step 34, data registration to the score recording table is performed in the form as shown in FIG. 25, and by completing the data registration, in step 28, When it is determined that the processing has been completed for all positions of the search target image, the process proceeds to step 35, and the maximum score sum value having a value equal to or greater than a prescribed threshold value included in the score recording table is extracted.

  Subsequently, in step 36, the extracted maximum score sum values are grouped so that adjacent ones belong to the same groove, and in step 37, the maximum score sum value indicating the maximum value in the same group is searched. Choose as.

  Subsequently, by referring to the score recording table (score recording means 12a) in step 38, the position coordinate / deformation parameter indicated by the score total value of the search result is acquired. Subsequently, in step 39, the template 0 is deformed according to the acquired deformation parameter, and the degree of coincidence with the search target image indicated by the acquired position coordinate is calculated.

  Subsequently, in step 40, the total score of the search target image selected in step 37, the position coordinates / deformation parameter acquired in step 38, and the degree of coincidence calculated in step 39 are output, and the process is terminated. To do.

  FIG. 29 shows an overview of the processing flows of FIGS. 26 to 28, and FIG. 30 shows an overview of the processing flows of FIGS. 18 and 19 executed by the first embodiment.

  As shown in FIG. 30, in the case of following the first embodiment, the templates 1 to 4 are moved independently and the scores are added, so that the template image portion included in the search target image is complicated. Although it becomes possible to search even if it is deformed, as shown in FIG. 14, it is necessary to use a score recording table having a large memory capacity prepared for each position of the search target image.

  On the other hand, when the second embodiment is followed, as shown in FIG. 29, the templates 1 to 4 are moved in an integrated manner with the template 0 as a reference to obtain the total value of the scores. Therefore, it is sufficient to prepare a score recording table having a small memory capacity as shown in FIG.

  Here, in the second embodiment shown in FIG. 20, the configuration in which the collation result storage unit 14a is prepared is adopted. However, without preparing the collation result storage unit 14a, the templates 1 to 4 and the search target image are prepared each time. The degree of coincidence may be calculated to obtain the sum.

  When this configuration is used, the collation result accumulating unit 14a can be omitted, and thus the memory area can be further saved.

  In the second embodiment described above, the templates 1 to 4 are composed of blurred images in which the degree of blur increases as the distance from the center increases, and the search target images to be collated with the templates 1 to 4 Even in this case, the degree of blur increases as the distance from the position overlapping the center increases in accordance with the process of the pattern generation unit 9a that executes the same process as the pattern generation unit 9 of the first embodiment. Although it is configured with a blurred image, it may be configured with an image with only one of them blurred, and the blurring method is also configured with a blurred image showing a single blur level regardless of the distance. You may do it.

(Embodiment 3)
FIG. 31 is a schematic configuration diagram of an image processing apparatus according to the third embodiment.

  The image processing apparatus according to the present embodiment includes a voting coordinate information calculation unit 1b, a voting coordinate information storage unit 2b, a template processing unit 3b, an image blurring unit 4b, a template storage unit 5b, and an input image processing unit 6b. Processed image storage means 7b, search control means 8b, matched pattern generation means 9b, inter-pattern matching means 10b, score voting means 11b, score recording means 12b, peak search means 13b, and verification result storage Means 14b.

  In the first and second embodiments, the template 0 is divided to generate templates 1 to 4 as shown in FIGS. 3 and 4, for example, and the template image portion included in the search target image is used. In contrast to the configuration of searching, in this embodiment, as shown in FIG. 32, the image of template 0 is displayed according to a plurality of reference points (excluding the reference point of template 0) included in template 0. For example, four templates 9 to 12 are generated by blurring, and a template image portion included in the search target image is searched for using the templates.

  In order to generate such templates 9 to 12, the template processing unit 3b selects a plurality of reference points when the template 0 is given. Next, the template 0 is transferred to the image blurring unit 4b. Upon receiving this, the image blurring unit 4b blurs the template 0 with a plurality of blurring degrees and returns it to the template processing unit 3b.

As a specific blurring method, for example, there is a method using a smoothing filter by convolving a two-dimensional matrix. Examples of the smoothing filter include an image analysis handbook (supervised by Takagi and Shimoda, 1991, University of Tokyo Press). ) On page 539, there is a local average filter and a local weighted average filter,
In the case of the size of these matrices, or in the case of a local weighted average filter, the degree of blurring can be changed by changing the degree of change in weight depending on the distance from the center. That is, in the case of increasing these matrices or in the case of a local weighted average filter, a blurred image having a higher degree of blurring is generated as the distance from the center is larger.

  In this way, the template processing unit 3b generates templates 9 to 12 as shown in FIG. 32 by blurring the image of the template 0 according to a plurality of reference points included in the template 0.

  From this, templates 9 to 12 as shown in FIG. 32 are stored in the template storage lateral means 5b.

  The voting coordinate information calculation means 1b performs processing for creating a voting coordinate position table. In the first embodiment, the voting coordinate position table shown in FIGS. 8 to 11 according to the equations (3) and (4). Is created in accordance with the equations (1) and (2).

  As will be described later, in the case of following the third embodiment, the processing is performed so as to specify the positions of the reference points of the templates 9 to 12 on the basis of the position of the reference point of the template 0. The information calculation means 1b calculates the voting coordinate position table according to the equations (1) and (2).

  From this, the voting coordinate information accumulating means 2b accumulates a voting coordinate position table as shown in FIG. 33 in which the signs of the table data shown in FIGS.

  The input image processing means 6b, the processed image storage means 7b, the to-be-matched pattern generation means 9b, and the inter-pattern matching means 10b each execute the processing of the input image processing means 6 of the first embodiment. When the embodiment is followed, the degree of matching (score) between the patterns of the templates 9 to 12 calculated by the inter-pattern matching unit 10b and the search target image is not sent to the score voting unit 11b each time. The result is accumulated in the collation result accumulating unit 14b via the search control unit 8b.

  FIG. 34 illustrates the data structure of the score stored in the collation frame storage unit 14b. As shown in the figure, the collation result accumulating unit 14b associates each of the templates 9 to 12 with the position of the search target image where the reference point (the position shown in FIG. 32) of each template 5 to 8 is placed. The degree of coincidence (score) with the blurred search target image at that position is accumulated.

  As shown in FIG. 35, the score recording unit 12b associates the reference point of the template 0 (the position shown in FIG. 32) with the position of the search target image and the score of the templates 9 to 12 at that position. The score recording table manages the maximum value of the total sum and the deformation parameters (enlargement / reduction ratio a, rotation angle γ) that realize the maximum score total value.

  The score voting unit 11b executes data registration in the score recording unit 12b. The score voting unit 11b arranges the reference point of the template 0 at each position of the search target image, and a voting coordinate position table (shown in FIG. 33). By referencing, the positions of the reference points of the templates 9 to 12 when the deformation parameters are changed at the respective arrangement positions are obtained, and the collation result accumulating means 14b (shown in FIG. 34) is referred to using the positions as keys. As a result, the scores of the templates 9 to 12 are obtained, and the total value thereof is calculated.

  Then, when the score voting means 11b calculates the score total value in the form as shown in FIG. 36 as described above, the score voting means 11b compares the maximum score total value recorded in the score recording table with a larger score total value. Is calculated, data registration in the score recording unit 12b is executed by updating the data recorded in the score recording table accordingly.

  That is, the score voting unit 11b executes data registration in the score recording unit 12b (score recording table) by executing the process shown in FIG.

  Thus, in the case of following this embodiment, the score recording table constituting the score recording unit 12b corresponds to the position of the search target image where the reference point of the template 0 is placed, as shown in FIG. In addition, the maximum value of the sum of the scores of the templates 9 to 12 at that position and the deformation parameter that realizes the maximum score sum value are recorded.

  Therefore, the peak searching means 13b can immediately execute plotting of the score map and parameter values on the map as shown in FIG. 17 by referring to the score recording table.

  From now on, the peak search means 13b, like the peak search means 13 of the first embodiment, extracts a peak search method (for example, a maximum score sum having a specified threshold value or more from a score map, Perform grouping in such a way as to belong to the same group, and identify the peak location (a method that finds the peak location by specifying the one with the maximum total score value indicating the maximum value in each group) And the coordinates of the position and the maximum score total value are output, and the deformation parameters are obtained from the corresponding coordinates in the parameter value map and are also output.

  In the case of following the first embodiment, since the scores are added by moving the templates 1 to 4 independently, even if the template image portion included in the search target image is complicatedly deformed, it can be detected. However, as shown in FIG. 14, it is necessary to use a score recording table having a large memory capacity prepared for each position of the search target image.

  On the other hand, in the case of following this embodiment, as can be seen from the above description, the templates 9 to 12 are moved in an integrated manner with reference to the template 0 to obtain the total value of the scores, and this is used as the template. Since voting is performed using the 0 position as a voting destination, it is sufficient to prepare a score recording table having a small memory capacity as shown in FIG.

  In addition, according to the second embodiment, the templates 1 to 4 generated in the form of dividing the template 0 are used, so that the search processing is performed in a form in which some image information of the template 0 is removed. On the other hand, in the case of following the present embodiment example, since the templates 9 to 12 that are generated in the form of blurring part of the image information of the template 0 are used, such image information is not removed and a high-accuracy search is performed. Processing can be realized.

  Next, according to the processing flow shown in FIGS. 38 to 40, the processing in the image search mode executed by the image information processing apparatus of the present invention when following the present embodiment will be described.

  Here, in this processing flow, the collation result accumulating unit 14b is not prepared, and the degree of coincidence between the templates 9 to 12 and the search target image is calculated and the sum is obtained each time.

  When entering the image search mode, the image processing apparatus according to the present embodiment first determines in step 50 whether or not the processing has been completed for all positions of the search target image, as shown in this processing flow.

  That is, as described below, template 0 (the template from which templates 9 to 12 are generated) is placed on the search target image while shifting one pixel at a time in the X direction and the Y direction, and predetermined processing is executed. Therefore, it is determined whether or not all the processes have been completed.

  When it is determined by the determination processing in step 50 that the processing has not been completed for all positions of the search target image, the process proceeds to step 51 to set the reference point of template 0 from the unprocessed positions of the search target image. One processing position is selected (an arrest is made while searching one screen in the X and Y directions).

  Subsequently, in step 52, one unprocessed deformation parameter is selected by selecting one unprocessed line from the vote coordinate position table. For example, when selecting rows one by one from the top of the voting coordinate position table (shown in FIG. 33) (rows that are common to the four tables) one by one, By selecting, one unprocessed deformation parameter is selected.

  Subsequently, in step 53, an initial value 9 is set to a variable i which becomes a template identification number. In this embodiment, i = 9 to 12 can be taken.

  Subsequently, at step 54, it is determined whether or not the value of the variable i has become larger than 12, and when it is determined that the value has not increased, the routine proceeds to step 55 where the template i is stored in the template storage means 5b. Take out the image.

  Subsequently, in step 56, Δx / Δy pointed to by the template i registered in the row selected in step 52 and the processing position selected in step 51 are added, so that the reference of the template i on the search target image is added. Specify the position of the point.

  Subsequently, in step 57, a blurred image that is a target for collation of the template i at the selected processing position is created from the search target image (the search target image blurred in advance in a plurality of stages has been described in the first embodiment. In step 58, the degree of coincidence between the template i and the created blurred image is calculated.

  Subsequently, in step 59, the value of the variable i is incremented by one to process the next template i, and then the process returns to step 54.

  In this way, by repeating the processing of step 55 to step 59, in step 54, the processing position (reference point position) of the selected template 0 is specified by the deformation parameter registered in the selected row. When it is determined that the degree of coincidence calculation has been completed at the position of the reference point of the templates 9 to 12, the process proceeds to step 60 to calculate the sum of the calculated degrees of coincidence (the degree of coincidence between the templates 9 to 12 and the search target image) To do.

  Subsequently, in step 61, it is determined whether or not the calculated total value is larger than the maximum score indicated by the processing position of the selected search target image recorded in the score recording table (shown in FIG. 35). When it is determined that the value is large, the process proceeds to step 62, and the data recorded in the score recording table is updated according to the calculated total value and the deformation parameter that realizes the sum (if it is not registered, it is newly registered). Become).

  On the other hand, when it is determined in step 61 that the calculated total value is smaller than the maximum score total value indicated by the processing position of the selected search target image recorded in the score recording table, the processing in step 62 is omitted. To do.

  Subsequently, in step 63, it is determined whether or not the processing has been completed for all the rows of the voting coordinate position table, and when it is determined that the processing has not been completed for all the rows, the processing is performed for the next row. If it is determined that the process has been completed for all the rows, the process returns to step 50 to perform the process for the processing position of the next search target image.

  In this way, by repeating the processing of step 50 to step 63, data registration to the score recording table is performed as shown in FIG. 37, and the data registration is completed. When it is determined that the processing has been completed for all positions of the search target image, the process proceeds to step 64, and the maximum score total value having a value equal to or greater than a prescribed threshold included in the score recording table is extracted.

  Subsequently, in step 65, the sum of the extracted maximum scores is grouped in such a way that adjacent ones belong to the same group, and in step 66, the maximum score sum indicating the maximum value in the same group is searched. Select as a result.

  Subsequently, in step 67, by referring to the score recording table (score recording means 12b), the position coordinate / deformation parameter indicated by the total score value of the search result is acquired. Subsequently, in step 68, the template 0 is deformed according to the acquired deformation parameter, and a coincidence point with the search target image indicated by the acquired position coordinate is calculated.

  Subsequently, in step 69, the total number of search results selected in step 66, the position coordinates / deformation parameters acquired in step 67, and the degree of coincidence calculated in step 68 are output, and the process is terminated.

  In this way, even in the case of following this embodiment, the templates 9 to 12 are moved in an integrated manner with reference to the template 0 as in FIG. 29 showing the outline of the processing of the embodiment 2. Since the total value of the scores is obtained and voted with the position of template 0 as the vote destination, it is sufficient to prepare a score recording table having a small memory capacity as shown in FIG.

  In addition, in the case of following the present embodiment example, the total value of the score is obtained using the templates 5 to 8 obtained by blurring a part of the template 0, and it is voted with the position of the template 0 as the vote destination. Compared with the case where templates 1 to 4 configured by dividing 0 are used, the voting is performed while leaving the image information of template 0. Therefore, the search process is more accurate than the case according to the second embodiment. Can be realized.

  Here, in the processing flow of FIGS. 38 to 39, the matching result between the templates 9 to 12 and the search target image is calculated and the sum is obtained each time without preparing the collation result accumulating unit 14b. However, a method may be used in which the degree of coincidence between the templates 9 to 12 and the search target image is calculated in advance and registered in the collation result accumulation unit 14b.

  When this configuration is used, processing similar to the processing flow of FIGS. 26 to 28 described in the second embodiment is executed.

  By using this configuration of preparing the collation result accumulating unit 14b, it is not necessary to calculate the degree of coincidence between the templates 9 to 12 and the search target image, and the calculation time can be shortened.

  That is, when the processing position of the template 0 is selected, each time the degree of coincidence between the templates 9 to 12 and the search target image is calculated, the calculation is performed before when another processing position is selected. However, when the matching result accumulating unit 14b is used, it is possible to reduce the calculation time by eliminating the need for such overlapping calculation processing. It becomes like this.

  In Embodiment 3 described above, the templates 9 to 12 are configured with a blurred image in which the degree of blurring increases as the distance from the reference point increases, and the search target to be collated with the templates 9 to 12 As for the image, the process of the collation pattern generation unit 9b that executes the same process as the collation pattern generation unit 9 of the first embodiment. Therefore, the degree of blurring increases as the distance from the position overlapping the reference point increases. Although it is configured with a blurred image that becomes larger, only one of them may be configured with a blurred image, and the blurring method is also configured with a blurred image that shows a single blur level regardless of the distance. You may make it do.

  Although the image processing method of the present invention has been described according to the above embodiment, the present invention is not limited to this. For example, the numerical values shown in the embodiment are merely for convenience of description. However, the present invention is not limited to this.

  Further, in the image processing method described in the embodiment, it is needless to say that the processing steps shown in FIGS. 1 to 43 are configured by a computer program, and this program can be executed by the computer. A recording medium that can be read by the computer, such as a flexible disk, MO, ROM, memory card, CD, DVD, removable disk, etc. It can be recorded and stored or distributed. It is also possible to provide this program via a network such as the Internet or electronic mail.

  Then, the present invention can be implemented by installing the program from these recording media into a computer, or by downloading the program from a network and installing the program into the computer. However, installation on a computer is a computer unit, and when there are multiple computers to be installed due to multiple devices and systems, it is natural that the program is installed for each necessary processing part. is there. In this case, the program may be recorded on a recording medium corresponding to a computer, or downloaded via a network.

1 is a schematic configuration diagram of an image processing apparatus according to Embodiment 1. FIG. Explanatory drawing of the division | segmentation process of the template in Embodiment 1. FIG. Explanatory drawing of the creation process of the template in Example 1 of Embodiment. Explanatory drawing of the template used in Embodiment Example 1. FIG. Explanatory drawing of the template used in Embodiment Example 1. FIG. Explanatory drawing of the reference position of a template. Explanatory drawing of the vote destination X coordinate / vote destination Y coordinate in Example 1 of Embodiment. Explanatory drawing of the voting coordinate position table used in Embodiment Example 1. FIG. Explanatory drawing of the voting coordinate position table used in Embodiment Example 1. FIG. Explanatory drawing of the voting coordinate position table used in Embodiment Example 1. FIG. Explanatory drawing of the voting coordinate position table used in Embodiment Example 1. FIG. Explanatory drawing of the template position where the degree of matching becomes a peak. Explanatory drawing of the blurred image produced | generated from a search object image. Explanatory drawing of the formation of the score recording means in Example 1 of Embodiment. Explanatory drawing of the voting process in Example 1 of Embodiment. Explanatory drawing of the voting process in Example 1 of Embodiment. Explanatory drawing of the peak search process in Embodiment 1. FIG. 5 is a process flowchart of an image search mode in the first embodiment. 5 is a process flowchart of an image search mode in the first embodiment. FIG. 6 is a schematic configuration diagram of an image processing apparatus according to Embodiment 2; It is explanatory drawing of the vote coordinate position table used in Example 2 of an embodiment. Explanatory drawing of the data structure of the score accumulate | stored in the collation result accumulation | storage means in Example 2 of an embodiment. Explanatory drawing of the structure of the score recording means in Example 2 of an embodiment. Explanatory drawing of the voting process in Example 2 of Embodiment. Explanatory drawing of the voting process in Example 2 of Embodiment. 10 is a processing flowchart of an image search mode in the second embodiment. 10 is a processing flowchart of an image search mode in the second embodiment. 10 is a processing flowchart of an image search mode in the second embodiment. Explanatory drawing which roughly demonstrated the process of the image search mode of Embodiment 2. FIG. Explanatory drawing which roughly demonstrated the process of the image search mode of Embodiment 1. FIG. FIG. 10 is a schematic configuration diagram of an image processing apparatus according to Embodiment 3; Explanatory drawing of the template used in Embodiment Example 3. FIG. Explanatory drawing of the voting coordinate position table used in Example 3 of an embodiment. Explanatory drawing of the data structure of the score accumulate | stored in the collation result accumulation means in Embodiment 3. FIG. Explanatory drawing of the structure of the score recording means in Embodiment Example 3. FIG. Explanatory drawing of the voting process in Example 3 of an embodiment. Explanatory drawing of the voting process in Example 3 of an embodiment. 14 is a processing flowchart of an image search mode in the third embodiment. 14 is a processing flowchart of an image search mode in the third embodiment. 14 is a processing flowchart of an image search mode in the third embodiment. Explanatory drawing of the template image used for description of a prior art and this invention. Explanatory drawing of the template image used for description of a prior art and this invention. Explanatory drawing of a prior art.

Explanation of symbols

1, 2a, 2b ... Voting coordinate information calculation means 2, 2a, 2b ... Voting coordinate information storage means 3, 3a, 3b ... Template processing means 4, 4a, 4b ... Image blurring means 5, 5a, 5b ... Template storage means 6 , 6a, 6b ... input image processing means 7, 7a, 7b ... processed image storage means 8, 8a, 8b ... search control means 9, 9a, 9b ... matched pattern generation means 10, 10a, 10b ... inter-pattern matching means 11 , 11a, 11b ... score voting means 12, 12a, 12b ... score recording means 13, 13a, 13b ... peak search means 14a, 14b ... collation result storage means

Claims (28)

An image processing apparatus that searches for a region having a high degree of matching with a template that may exist in a search target image,
Means for creating a plurality of sub-templates each having a plurality of position coordinates selected from the template as reference points;
Means for disposing a reference point of the sub-template at each position coordinate of the search target image and calculating a degree of matching between the sub-template and the search target image;
Calculating a positional relationship between a point corresponding to a reference point of the sub-template and a corresponding reference point of the template when the template is deformed based on a plurality of coordinate transformation parameters assumed in advance; Means for calculating a sum total value obtained by adding the degree of matching calculated with respect to the score of the reference point of the template corresponding to each coordinate transformation parameter;
For each position coordinate of the search target image, specifying a coordinate conversion parameter that maximizes the total value and calculating the maximum value;
An image processing apparatus comprising: means for searching for a position of a reference point of the template where the maximum value is maximized. An image processing apparatus that searches for a region having a high degree of matching with a template that may exist in a search target image,
For each of a plurality of reference points selected from the template, means for creating a sub-template from the image information of the neighboring area;
Means for holding a memory area for recording a score for a combination of position coordinates and a plurality of coordinate transformation parameters assumed in advance;
With all sub-templates as processing targets, each sub-template is placed at each position coordinate of the search target image to calculate the degree of matching with the search target image, and for each of the coordinate conversion parameters, the coordinate conversion parameter and the placement Means for calculating a template position coordinate pointed to by a position coordinate of a reference point of the sub-template, and adding the degree of adjustment to a score of the memory area pointed to by the coordinate conversion parameter and the template position coordinate;
An image processing apparatus comprising: means for searching a region having a high degree of matching with a template from a search target image based on the score of the addition result. The means for creating the sub-template is:
As the distance from the reference point of the sub-template increases, the sub-template is created from the image information of the neighboring area in which the degree of blur increases, and as a search target image for calculating the degree of matching with the sub-template, Create image information in which the degree of blur increases as the distance from the position overlapping the reference point of the sub-template increases, or image information of the neighboring region that indicates a single blur degree regardless of the distance Create a sub template from
The image information indicating a single blurring degree is created as a search target image for which the degree of matching with the sub-template is to be calculated, and the degree of matching between the sub-template and the sub-template is calculated. 2. Image processing apparatus according to item 2. An image processing apparatus that searches for a region having a high degree of matching with a template that may exist in a search target image,
For each of a plurality of reference points selected from the template, means for creating a sub-template from the image information of the neighboring area;
When placing the template at each position of the search target image, for each assumed coordinate conversion parameter, determine the position coordinates of the reference point of each sub template, and the total degree of matching with the search target image for all sub templates A means for calculating a value and specifying a maximum value of these total values and a coordinate conversion parameter for realizing the maximum value for each position coordinate of the search target image in which the template is arranged;
An image processing apparatus comprising: means for searching a region having a high degree of matching with a template from a search target image based on the identification result. The image processing apparatus according to claim 4, wherein the means for creating the sub-template blurs image information of the neighboring area. An image processing apparatus that searches for a region having a high degree of matching with a template that may exist in a search target image,
For each of a plurality of reference points selected from the template, means for creating a plurality of sub-templates by blurring image information according to the reference points;
When placing the template at each position coordinate of the search target image, the position coordinates of the reference point of each sub template are determined for each assumed coordinate conversion parameter, and the degree of matching with the search target image for all the sub templates is determined. A means for calculating a sum value and specifying a maximum value of these sum values and a coordinate conversion parameter for realizing the maximum value for each position coordinate of the search target image in which the template is arranged;
An image processing apparatus comprising: means for searching a region having a high degree of matching with a template from a search target image based on the identification result. The image processing apparatus according to claim 4, wherein the specifying unit calculates a total value of a degree of matching with a search target image subjected to blurring. The specifying means uses a table means for recording a maximum value of the total value and a coordinate conversion parameter for realizing the maximum value in association with the position coordinates of the search target image, and the position of the search target image where the template is arranged. The image processing apparatus according to claim 4, wherein a maximum value of the total value and a coordinate conversion parameter that realizes the maximum value are specified for each coordinate. A means for calculating the degree of matching between the sub-template and the search target image and storing it in a memory while taking correspondence with the position coordinates of the reference point of the sub-template, the specifying means being stored in the memory The image processing apparatus according to claim 4, wherein a total value of the matching degrees is calculated using a matching degree. The searching means includes sub template, template and image information of search target image as prescribed image conversion,
Image transformation for differential processing, image transformation for logarithmic transformation, image information is divided into multiple blocks, the direction components obtained from the differential processing results in this block are quantized into a finite number of direction components, and similarly differentiated Image conversion for calculating a histogram for each direction weighted by the direction intensity obtained from the processing result, image conversion for weighting the histogram for each direction of the adjacent block when calculating the histogram for each direction, and for calculating the histogram for each direction. Performing one or more image conversions of image conversion for calculating a weighted value of the histogram as a histogram of the block, image conversion for performing geometric conversion,
10. The image processing apparatus according to claim 2, wherein a region having a high degree of matching between the search target image and the sub-template or the search target image and the template is searched. 11. When the geometric transformation has a plurality of coordinate transformation parameters, an area having a high degree of matching with the template is searched in such a manner that some of the coordinate transformation parameters are fixed. The image processing apparatus described. The searching means outputs position coordinates of a region with a high degree of matching as a search result, or outputs the total value and / or the coordinate conversion parameter in the position coordinates in addition to outputting the position coordinates. The image processing apparatus according to claim 2, wherein the image processing apparatus is characterized. The template is transformed by coordinate transformation according to the coordinate transformation parameter in the position coordinates of the region with a high degree of matching to be searched, and the degree of matching between the transformed template and the search target image at the position coordinates is calculated. The image processing apparatus according to claim 2, further comprising an output unit. An image processing method for searching for a region having a high degree of matching with a template that may exist as a search target image,
Sub template creation means, sub template search means, coordinate parameter identification means, template search means,
A step of creating a plurality of sub-templates each having a plurality of position coordinates selected from the template as reference points,
The sub-template search means arranges the reference point of the sub-template at each position coordinate of the search target image and calculates the degree of matching between the sub template and the search target image, and a plurality of coordinate conversion parameters assumed in advance. Calculating a positional relationship between a point corresponding to the reference point of the sub-template and a corresponding reference point of the template when the template is deformed, and calculating the degree of matching calculated for all the sub-templates A step of calculating a total value added to the reference point score of the template corresponding to each conversion parameter, and
A coordinate parameter specifying unit for each position coordinate of the search target image, specifying a coordinate transformation parameter that maximizes the total value and calculating the maximum value;
And a template searching means for searching for a position of a reference point of the template where the maximum value is maximized. An image processing method for searching a region having a high degree of matching with a template that may exist in a search target image,
A sub-template creation means, a table means, a sub-template search means, and a template search means,
A process of creating a sub-template from image information in the vicinity of each of a plurality of reference points selected from the template by the sub-template creating means;
A process in which the table means records a score for a combination of position coordinates and a plurality of coordinate transformation parameters assumed in advance;
The sub template search means sets all sub templates as processing targets, arranges each sub template at each position coordinate of the search target image, calculates the degree of matching with the search target image, and for each of the coordinate transformation parameters, A template position coordinate indicated by the conversion parameter and the position coordinate of the reference point of the sub-template in the arrangement is calculated, and the degree of adjustment is added to the score of the memory area indicated by the coordinate conversion parameter and the template position coordinate The process of
A template image processing method comprising: a step of searching for a region having a high degree of matching with a template from a search target image based on a score of the addition result. In the process of creating a sub template by the sub template creating means,
As the distance from the reference point of the sub-template increases, the sub-template is created from the image information of the neighboring area in which the degree of blur increases, and as a search target image for calculating the degree of matching with the sub-template, Create image information in which the degree of blur increases as the distance from the position overlapping the reference point of the sub-template increases, or image information of the neighboring region that indicates a single blur degree regardless of the distance The process of creating sub-templates from
A process of creating image information indicating a single degree of blur as a search target image that is a target for calculating a degree of matching with the sub-template and calculating a degree of matching with the sub-template. 16. The image processing method according to claim 15, further comprising: An image processing method for searching a region having a high degree of matching with a template that may exist in a search target image,
Sub template creation means, coordinate parameter identification means, and template search means,
A process of creating a sub-template from image information in the vicinity of each of a plurality of reference points selected from the template by the sub-template creating means;
When the coordinate conversion parameter specifying means places the template at each position of the search target image, the position coordinates of the reference point of each sub template are determined for each assumed coordinate conversion parameter, and the search target for all sub templates is determined. Calculating the total value of the degree of matching with the image, and for each position coordinate of the search target image in which the template is arranged, specifying a maximum value of these total values and a coordinate conversion parameter that realizes the maximum value;
An image processing method comprising: a template searching unit searching a region having a high degree of matching with a template from a search target image based on the identification result. 18. The image processing method according to claim 17, wherein the step of creating the sub-template by the sub-template creation unit includes a step of blurring image information of the neighboring area. An image processing method for searching a region having a high degree of matching with a template that may exist in a search target image,
Sub template creation means, coordinate transformation parameter identification means, and template search means,
A process of creating a plurality of sub-templates by blurring image information in accordance with the reference points for each of a plurality of reference points selected from the template,
When the coordinate conversion parameter specifying unit arranges the template at each position coordinate of the search target image, the position coordinate of the reference point of each sub template is determined for each assumed coordinate conversion parameter, and search for all sub templates is performed. Calculating the total value of the degree of matching with the target image, and for each position coordinate of the search target image in which the template is arranged, identifying a maximum value of these total values and a coordinate conversion parameter that realizes the maximum value;
An image processing method, comprising: a step of searching for a region having a high degree of matching with the template from the search target image based on the identification result. In the process of specifying the coordinate transformation parameter specifying means,
20. The image processing method according to claim 17, further comprising a step of calculating a total value of matching degrees with the search target image subjected to blurring. In the process of specifying the coordinate transformation parameter specifying means,
For each position coordinate of the search target image in which the template is arranged, using the table process for recording the maximum value of the total value and the coordinate conversion parameter that realizes the maximum value in association with the position coordinates of the search target image 21. The image processing method according to claim 17, further comprising a step of specifying a maximum value and a coordinate transformation parameter that realizes the maximum value. Calculating the degree of matching between the sub-template and the search target image, and storing it in the memory means while taking correspondence with the position coordinates of the reference point of the sub-template,
In the process of specifying the coordinate transformation parameter specifying means,
The image processing method according to any one of claims 17 to 21, further comprising a step of calculating a total value of the matching degrees using a matching degree stored in the memory means. The template search means divides the image information into a plurality of blocks, image conversion for performing differential processing, image conversion for performing logarithmic conversion, and image information as prescribed image conversion for the image information of the sub-template, template and search target image, Quantize the direction component obtained from the differential processing result in the image into a finite number of direction components, and simultaneously calculate the direction-specific histogram weighted by the direction intensity obtained from the differential processing result, adjacent when calculating the direction-specific histogram Any one of image conversion for weighting a histogram for each direction of a block to be performed, image conversion for calculating a weighted value for a histogram in the adjacent direction as the histogram for the block at the time of calculating the histogram for each direction, and image conversion for performing geometric conversion One or more image conversions and search target images and sub-tempts Over preparative or image processing method according to any one of claims 15 to 22 characterized by having a process of searching for a region with a high matching degree between the search target image and the template. In the process of searching by the template search means,
In the case where the geometric transformation has a plurality of coordinate transformation parameters, a part of the coordinate transformation parameters is fixed, and a process of searching for a region having a high degree of matching with the template is included. The image processing method for starvation according to claim 23. In the process of searching by the template search means,
The process of outputting the position coordinates of the region with a high degree of matching as a search result, or
The image processing method according to any one of claims 15 to 24, further comprising a step of outputting the total value and / or the coordinate conversion parameter at the position coordinates in addition to outputting the position coordinates. In the process of searching by the template search means,
The template is transformed by coordinate transformation according to the coordinate transformation parameter in the position coordinates of the region with a high degree of matching to be searched, and the degree of matching between the transformed template and the search target image at the position coordinates is calculated. 26. The image processing method according to claim 15, further comprising an output process. An image processing program for causing a computer to execute the image processing method according to any one of claims 14 to 26. 27. A computer-readable recording medium on which an image processing program for causing a computer to execute the image processing method according to any one of claims 14 to 26 is recorded.

Images