JPH08212342A - Method for extracting texture area of image - Google Patents

Abstract

PURPOSE: To provide an extracting method which can extract the texture area of the image by accurately separating the texture area and a non-texture area. CONSTITUTION: After a high-frequency part of the original image is extracted and the directivity of pixels is detected by a spatial filter (operator) (steps S101 and S102), pixels judged to have directivity as to respective directions are counted, the counted value is compared with a threshold value th1 (step 103). When the number of the pixels judged to have directivity is larger than the threshold value th1, variance on the plane is calculated (step 105), and when the sum of the variance is larger than a threshold value th2, it is judged that a point is in the texture area (step 108). After smoothing and binarization, the region of a large are consisting of pixels having luminance larger than a specific value is judged as the texture area (steps 109-111).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for extracting a texture area in an image, and more particularly to a method for extracting only the texture area from a grayscale image including the texture area.

[0002]

2. Description of the Related Art A texture (raw eye) area of an image is a redundant area that is visually recognized as a uniform appearance image even if it is not strictly uniform. Processing separately from the area is useful in image compression.

Therefore, conventionally, the texture region of the image is extracted. When extracting the texture region, since the homogeneous region of the repeated pattern is not clear in the texture in the natural image, feature extraction is performed to clarify the texture structure. As a method for extracting the texture region, a method using an average value of density, a method using variance, a method using spectrum information of orthogonal transformation, etc. have been conventionally known.

[0004]

However, in the conventional texture region extraction method described above, the former average value of the density,
In the method using dispersion, in the case of a natural image, even in the visually same texture, the part where homogeneity is lost is
The average value and the dispersion value of the density cannot be relied on, and the texture area cannot be extracted. Also, depending on the type of texture, the average value of the non-texture area and density in the image,
The variances are so close that it is not possible to extract texture regions.

On the other hand, in the latter conventional method using orthogonal transform spectrum information, the high-frequency spectrum is judged to be a texture based on the spectrum calculated for each cell, so that the contour of the non-texture area is determined by the cell. In the case of crossing the inside, the high band of the non-texture region and the high band of the texture region are mixed in the same cell, and the classification of the texture region and the non-texture region is very complicated or unrealistic in reality. Often possible.

The present invention has been made in view of the above points,
An object of the present invention is to provide a method for extracting a texture region in an image, which can accurately separate the texture region and the non-texture region to extract the texture region.

[0007]

In order to achieve the above-mentioned object, in the invention described in claim 1, the directionality of the line of each target pixel of the image including the texture is set to the neighborhood of each target pixel. A spatial filter is used for the pixels to detect each of a plurality of predetermined directions, and the number of pixels having the same directionality among the pixels for which the directionality is detected is calculated for each direction. Areas in which the number of pixels with the same directionality is larger than a fixed value and the pixels with the same directionality are more dispersed than a certain degree on the plane are determined and extracted as texture areas. .

According to the third aspect of the present invention, among the neighboring pixels whose directionality is detected, the number of pixels having the same specific directionality as the arrangement direction of those pixels is calculated for each direction, and the calculation is performed. The number of pixels having a specific directionality is equal to or less than a certain value, and the pixels that form a specific directionality are determined and extracted as texture areas if they are not arranged in an array that has a specific directionality. It is a thing.

[0009]

In the present invention, attention is paid not to the average value of the densities and the spectrum of the orthogonal transformation, but to what kind of directivity each pixel in the vicinity has. In other words, it is determined in what direction the target pixel is on the top of the roof or the bottom of the valley that is continuous.

Next, in the invention according to claim 1, assuming that each pixel on the screen has this directionality information, a certain number of pixels having the directionality for each directionality in a certain study area. It is determined whether or not the above exist and they are sufficiently dispersed on the screen (plane). If it is sufficiently dispersed, this area is determined and extracted as a texture area.

Therefore, in the invention according to claim 1, the texture region is not "a region having a certain density distribution" or "a region having a certain high-frequency spectrum pattern", but "a region having various directions". Alternatively, it is defined as "a region that has a certain direction and is well dispersed".

Therefore, the area of the pattern in which each pixel of the target image as shown in FIG. 2 has a plurality of types of directionality and is sufficiently dispersed is determined as a texture area. Further, as shown in FIG. 3, even if each pixel of the target image has one type of directionality (vertical direction in this case), a pattern area that is sufficiently dispersed on the plane is also determined as a texture area. It However, as shown in FIG. 4, an area of a pattern in which each pixel of the target image has one directionality and is concentrated on a plane is determined as a non-texture area. As a result, it becomes possible to discriminate between the non-texture contour portion and the texture, which was difficult with the orthogonal transform spectrum information.

Further, according to the present invention, since the area in which the number of pixels having the luminance of the predetermined value or more in the area determined as the texture area is larger than the set value is extracted as the final texture area, As shown in FIG. 5, the number of pixels having directionality (that is, even if it is a region of the pattern in which each pixel of the target image has a plurality of directions and is considered to be sufficiently dispersed) When the number of pixels having a brightness equal to or higher than a predetermined value) is too small, it is determined as a non-texture area.

According to the third aspect of the present invention, if each pixel on the screen is provided with the information on the directionality, the information on the contour of the high-frequency part, that is, the texture region or the non-texture part is used. Paying attention to being in a position,
To exclude contours of non-texture areas, determine whether there are a certain number of pixels with a certain directionality for a certain directionality, and whether they are arranged with that particular directionality. Then, the portion determined in this way is excluded from the above-mentioned portion having many high frequencies, and the remaining area is determined to be the texture area.

Accordingly, in the invention according to the third aspect, the texture region is defined as "a region having a large number of high frequencies such that a certain number of pixels having a specific directionality are not arranged in the specific direction". Attached.

Therefore, in the present invention, for example, seven in the vertical direction,
It is assumed that the respective directivities in 49 pixels of seven so-called 7 × 7 cells in the horizontal direction are detected as shown in FIG. 6 (A) or FIG. 7 (A). Here, "the directionality is detected" is synonymous with including a large number of high frequencies, and since both have directionality for all pixels, both are images that are texture region candidates.

However, considering the directivity of each pixel in a total of four types of directions, that is, the horizontal direction, the vertical direction, the diagonally upward leftward direction, and the diagonally upward rightward direction, in the case of the pixel of the cell shown in FIG. Shows that the directivity of each pixel in the above four kinds of directions shown in (B), (C), (D) and (E) of FIG. I think.

On the other hand, in the case of the pixel of the cell shown in FIG. 7A, FIG. 7B, FIG. 7C, FIG.
The directionality of each pixel in the horizontal direction, the vertical direction, the diagonally upward leftward direction, and the diagonally upward rightward direction is appropriately dispersed for each pixel in the vertical direction, diagonally upward leftward direction, and diagonally upward rightward direction. However, the directionality of each pixel in the horizontal direction is shown in FIG.
As shown in, each pixel having a horizontal directionality is arranged.

In such a case, it is natural to think that there is a non-textured contour that crosses the center of the cell. Therefore, in the present invention, the cell shown in FIG. 7A is determined to be a non-texture area. It goes without saying that a region including a pattern in which the examination direction and the pixel directionality all match as in this example is determined to be a non-texture region, and in the present invention, most (depending on the case) pixel directionality is the array direction. If it matches with, it is determined as a non-texture area.

[0020]

Next, an embodiment of the present invention will be described. FIG. 1 shows a flow chart for explaining the operation of the first embodiment of the present invention. In the present embodiment, the pixel located at the center of a cell of 7 × 7 = 49 pixels is positioned as a pixel for texture detection, the entire cell is considered to be in the vicinity, and when the cell can be regarded as a texture region, The target pixel at the center is regarded as a part of the texture.

In this embodiment, the natural image of FIG. 8 is used as the original image, and the planting in the lower center and the forest in the lower right half are the texture regions to be extracted this time. Here, the purpose is to make the texture region into a white image.

First, in order to make the characteristics of the texture stand out, the high frequency part of the original image is extracted by using a 3 × 3 spatial filter (operator), and the high frequency part emphasized against the background of 50% gray is extracted (step). 101). As a result, the high frequency extraction image shown in FIG. 9 is obtained.

Next, referring to FIG.
Directionality is detected by eight types of 7 × 7 spatial filters (operators) shown in (A) to (H) (step 1).
02). 16 (A) and 16 (B) are horizontal line operators, FIGS. 16 (C) and 16 (D) are vertical line operators, and FIGS. 16 (E) and 16 (F) are upper left diagonal line detection operators. G) and (H) show operators for detecting the upper right diagonal line. There are two types for each of the four directions because they correspond to two types of thickness (resolution).

The use of the above spatial filter means, for example, when the horizontal line operator shown in FIG. 16 (A) is used, the value of the target pixel (luminance value) and the value of the center position of the horizontal line operator " 1 ”, and the values of neighboring pixels around it are also multiplied by the respective numerical values at the positions corresponding to the horizontal line operator shown in FIG. 16A, and the absolute value of the sum of the multiplication results is set to the cell. That is, the arithmetic operation is performed with the value of the target pixel at the center position of.

When the maximum absolute value of the operation results obtained by the eight types in these four directions for the target pixel at the center of the cell thus obtained exceeds a certain threshold value (not shown), the central pixel of the cell Is determined to be directional in that direction. However, it is classified into four directions regardless of the thickness. Therefore, the directionality of a directional pixel is only one of the above four directions.

With respect to this directivity, when the pixel brightness is represented by, for example, 256 gradations of 8 bits, the brightness of the central pixel with directionality in the horizontal direction is "160", and in the vertical direction is "190". FIG. 10 shows an image in which “220” is set in the line direction and “250” is set in the upper right diagonal line direction.
This is an image of 5 gradations combined with the brightness "0" having no directionality. As for this directional information, it is sufficient that it is originally data, and it is not necessary to indicate the type classification by brightness, but it was imaged for the convenience of processing.

Next, the number of pixels determined to be directional in each direction is counted while moving a 7 × 7 cell by one pixel, and the number of pixels is determined by the threshold value th1.
It is determined whether or not it is larger (step 103).
Specifically, in each cell in FIG.
The number of pixels of "0" corresponds to the number of pixels in the horizontal direction, and the like is counted in four directions.

As a result of the above comparison judgment, when the number of directional pixels is less than or equal to the threshold value th1, since the absolute number is small and it is difficult to distinguish from noise, the spatial variance of pixels is set to 0 (step 104). ), Its direction is not a texture. On the other hand, when the number of directional pixels is larger than the threshold value th1, the variance on that plane is calculated (step 105). There are various methods for calculating the variance on the plane, but in the present embodiment, the center of gravity of pixels with directionality was calculated, and the average distance from that was used as the variance. In addition, evaluation based on the average distance between omnidirectional pixels can be considered.

Next, the above step 104 or 10
Following the processing of 5, it is determined whether the sum of variances in all directions is larger than the threshold value th2 (step 106). In this embodiment, since the division is between the texture area and the non-texture area, the simple sum is used. When the sum of the variances is less than or equal to the threshold value th2, the brightness is "0" as a point in the non-texture area (step 107), and when it is larger than th2, the brightness is "255" as a point in the texture area (step 108). ). As a result, a texture area candidate image as shown in FIG. 11 is obtained.

Depending on the type of texture, a texture having a certain fixed directionality may be extracted by weighting, and in the area determined to be the texture area, pixels in the vicinity of the target pixel have the same direction. Regions may be divided into a plurality of types of texture regions based on the value of the number of pixels having the property and the degree of dispersion in each direction.

Here, the whitened portion of the texture region candidate image obtained by the above processing includes the texture region, and most of the non-textures including high regions such as windows of buildings are avoided. , The part where the non-texture contour is complicated is detected as the texture. In addition, there is a minute portion that is not textured even in the texture.

Therefore, next, these minute portions are smoothed by 7 × 7 cells (step 109). Specifically, this smoothing is an arithmetic operation in which the average value of the respective values of 49 pixels of a 7 × 7 cell is used as the value of the central pixel of that cell. As a result, the image shown in FIG. 12 is obtained.

Then, the smoothed image is binarized with a predetermined threshold value (step 110). As a result, a binarized image as shown in FIG. 13 is obtained. It can be seen that this binarized image has considerably less micro error than the texture region candidate image shown in FIG.

Then, in the white area (pixel area of luminance "255") of the binarized image shown in FIG. 13, a large area portion having a number of pixels larger than the threshold th3 is left (step 111). By this area determination step 111,
A minute area which is a detection error in the binarized image of FIG. 13 is removed, and a desired texture area image is obtained as shown in FIG.

It can be said that the whitened texture region in the image shown in FIG. 14 extracts a considerable part of the region clearly judged to be the texture by human vision in the original image shown in FIG. An image obtained by cutting out the original image from the image shown in FIG. 14 is as shown in FIG. In the image of FIG. 15, the non-texture portion is slightly cut out, but it can be seen that the texture-like portion of the leaves is extracted.

Next, a second embodiment of the present invention will be described. FIG. 17 shows a flowchart for explaining the operation of the second embodiment of the present invention. Also in this embodiment, a pixel located at the center of a cell of 7 × 7 49 pixels is positioned as a pixel for texture detection, and the entire cell is considered to be in the vicinity thereof. The target pixel is regarded as a part of the texture, and in FIG.
The same reference numerals are given to the same processing steps as.

In the present embodiment as well, the natural image shown in FIG. 8 in which the density difference between the texture region and the non-texture region around the artificial region and the difference in the power spectrum in the orthogonal transformation are not sufficient as the original image is used as the original image, and The planting in the lower part and the forest in the lower right half are the texture areas to be extracted this time, and the purpose is to make this texture area a blank image.

In this embodiment, as in the first embodiment, high frequency extraction is performed on the image information of the original image (step 101), and line (directional) detection by a spatial filter (step 102).
After performing the above, for each of a total of four types of directions, that is, the horizontal direction, the vertical direction, the diagonally upward leftward direction, and the diagonally upward rightward direction, each pixel having the directivity located in that direction (that is, the luminance “160”). , "190", "220", and "250" pixels), and a histogram is created (step 201).

Subsequently, among the pixels arranged in the count direction, the number of pixels whose directionality is the same as that of the count direction (ie, when the count direction is the horizontal direction, the pixels are arranged in the horizontal direction in the cell). It is determined whether the number of pixels having horizontal directionality among the seven pixels) is larger than a predetermined threshold value thr (step 202).

When the number of pixels in any one of the four types of counting directions is larger than thr,
For the reason described above, the target pixel has a luminance of “0” as a point in the non-texture area (step 203).

On the other hand, when the number of pixels is less than or equal to thr in any of the four types of counting directions, the directionality is appropriately dispersed in each direction, so it is determined that this cell is within the texture area. Then, a value obtained by multiplying the sum of all pixels having omnidirectionality by a constant is set as the luminance level of the central pixel of the cell (step 204). With the above operation, a texture region candidate image as shown in FIG. 18 is obtained.

The following processing is the same as that of the first embodiment, and the texture area candidate image is smoothed by 7 × 7 cells (step 109), whereby the minute portion which is not textured as shown in FIG. After obtaining the smoothed image, the binarized image is binarized with a predetermined threshold value (step 110). As a result, as shown in FIG. 20, a binarized image in which the minute error is considerably reduced compared to the texture region candidate image shown in FIG. 18 is obtained.

Then, in the blank area (pixel area of luminance "255") of the binarized image shown in FIG. 20, a large area portion having the number of pixels larger than the threshold th3 is left (step 111). As a result, a minute area which is a detection error in the binarized image of FIG. 20 is removed, and a desired texture area image is obtained as shown in FIG.

It can be said that the whitened texture region in the image shown in FIG. 21 extracts a considerable part of the region which is clearly judged to be a texture by human vision in the original image shown in FIG. An image obtained by cutting out the original image from the image shown in FIG. 21 is as shown in FIG. In the image of FIG. 22, a slightly non-textured portion is cut out, but it can be seen that the texture-like portion of the leaves is extracted.

[0045]

As described above, according to the present invention,
The texture area is not "the area with a certain density distribution" or "the area with a certain high frequency spectrum pattern",
Whether to define the texture region based on the pixel directionality and the degree of dispersion by defining it as a "region in which various directions are intricate" or "a region that is sufficiently dispersed even if it has a certain directionality" Or, define the directionality of pixels and the way of arranging the directionality by defining it as "a region with many high frequencies such that a certain number of pixels with a specific directionality are not arranged in the specific direction or more". Since the texture area is determined based on the original, it is possible to distinguish between the non-texture contour part and the texture, which was difficult with the orthogonal transform spectrum information. Texture areas can be extracted.

Further, according to the present invention, by extracting an area in which the number of pixels having a luminance equal to or higher than a predetermined value of the area determined as the texture area is larger than the set value as the final texture area, the directional If there are too few pixels with, it is determined as a non-texture area.
It is possible to prevent the noise portion from being extracted as a texture area.

[Brief description of drawings]

FIG. 1 is a flow chart for explaining the operation of the first embodiment of the present invention.

FIG. 2 is a diagram showing an example of a texture region extracted by the invention according to claim 1;

FIG. 3 is a diagram showing another example of a texture region extracted by the invention according to claim 1;

FIG. 4 is a diagram showing an example of a non-texture region detected by the invention according to claim 1;

FIG. 5 is a diagram showing another example of a non-texture area detected by the invention according to claim 1;

FIG. 6 is an explanatory diagram of an example of a texture region extracted by the invention according to claim 3;

FIG. 7 is an explanatory diagram of an example of a non-texture area detected by the invention according to claim 3;

FIG. 8 is a diagram showing an example of an original image.

FIG. 9 is a diagram of an example of a high frequency extraction image.

FIG. 10 is a diagram showing an example of a directional display 5-gradation image.

FIG. 11 is a diagram showing an example of a texture region candidate image according to the first embodiment.

FIG. 12 is a diagram showing an example of a smoothed image according to the first embodiment.

FIG. 13 is a diagram of an example of a binarized image according to the first embodiment.

FIG. 14 is a diagram showing an example of a texture region image according to the first embodiment.

FIG. 15 is a diagram showing an example of an image obtained by cutting out an original image from a texture region image according to the first embodiment.

FIG. 16 is an explanatory diagram of a spatial filter (line detection operator).

FIG. 17 is a flowchart for explaining the operation of the second embodiment of the present invention.

FIG. 18 is a diagram showing an example of a texture region candidate image according to the second embodiment.

FIG. 19 is a diagram showing an example of a smoothed image according to the second example.

FIG. 20 is a diagram showing an example of a binarized image according to the second embodiment.

FIG. 21 is a diagram showing an example of a texture region image according to the second embodiment.

FIG. 22 is a diagram showing an example of an image obtained by cutting out an original image from a texture region image according to the second embodiment.

[Explanation of symbols]

101 High-frequency extraction step 102 Line detection step 103, 106, 202 Large / small judgment step 105 Variance calculation step 107, 108, 203, 204 Pixel brightness setting step 109 Smoothing step 110 Binarization step 111 Region determination step 201 Pixel count Step

Claims (7)

[Claims] 1. The directionality of a line of each target pixel of an image including a texture is determined for each of a plurality of predetermined directions by using a spatial filter for pixels in the vicinity of each target pixel. The number of pixels having the same directivity among the pixels detected and having the same directivity is calculated for each direction, and the calculated number of pixels having the same directivity is larger than a certain value, and A method of extracting a texture region in an image, comprising determining and extracting a region in which pixels having the same directionality are dispersed more than a certain degree on a plane as a texture region. 2. When the center pixel of a cell made up of a plurality of pixels is the target pixel and the pixels in the remaining cells are the neighboring pixels, the same directivity calculated in each of the directions is provided. When the number of pixels in the neighborhood is larger than a fixed value, the variance of the neighboring pixels having the same directionality is calculated, and when the sum of all the variances in each direction is larger than a set value, the The method for extracting a texture region in an image according to claim 1, wherein the cell is determined to be in the texture region. 3. The directionality of a line of each target pixel of an image including a texture is determined for each of a plurality of predetermined directions by using a spatial filter for pixels in the vicinity of each target pixel. Detecting, the number of pixels showing the same specific directionality as the arrangement direction of those pixels among the neighboring pixels in which the directionality is detected is calculated for each direction,
The calculated number of pixels having a specific directionality is less than or equal to a certain value, and the pixels that form the specific directionality are determined and extracted as a texture area in an area in which the pixels that do not have the specific directionality are arranged. A method for extracting a texture region in an image characterized by the above. 4. When the center pixel of a cell composed of a plurality of pixels is the target pixel and the pixels in the remaining cells are the neighboring pixels, the number of neighboring pixels having the specific directionality is When the value is equal to or less than the constant value in all of the respective directions, it is determined that the cell is in the texture area,
4. The texture area extracting method according to claim 3, wherein a value obtained by multiplying a sum of directional pixels in the cell by a constant is used as a brightness level of a central pixel of the cell. 5. A texture area is obtained by performing smoothing for setting each pixel of the area determined to be the texture area to an average value of the values of neighboring pixels. A method for extracting a texture region in an image according to item 1. 6. The final texture area is extracted as an area in which the number of pixels having a brightness equal to or higher than a predetermined value in the area determined to be the texture area is larger than a set value. A method of extracting a texture region in an image according to any one of the above. 7. In the area determined as the texture area, a plurality of pixels in the vicinity of the target pixel having the same directionality and a degree of dispersion are calculated based on values calculated in each direction. The method for extracting a texture area in an image according to claim 1, wherein the texture area is divided into the following types of texture areas.