JPH05167889A - Image signal processor - Google Patents

Abstract

PURPOSE:To provide an image processing circuit by which uneven illumination can be eliminated or suppressed without emphasizing unnaturally its edge, even in a place in which is luminance is varied stepwise. CONSTITUTION:A processor is provided with a filtering circuit 9c for performing filtering to an image signal subjected to logarithmic compression, a filtering characteristic setting circuit 9b for setting adaptively a band pass characteristic of this filtering circuit 9c at every position of an image in accordance with a feature of the image signal, and a dynamic range/gain controller 10 for executing control of a dynamic range and gain with respect to the signal to which filtering is performed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image signal processing device for performing image processing for suppressing uneven illumination.

[0002]

2. Description of the Related Art In a general CRT monitor, the dynamic range that can be displayed is about 46 dB. Therefore, even if an image having a wider dynamic range than this, that is, an image having a large difference in brightness is displayed, it is not possible to display the bright part and the dark part of the image at the same time.

For example, in an image by an electronic endoscope (hereinafter referred to as an electronic scope) that displays an electric image signal photoelectrically converted by using a solid-state image pickup device such as a CCD on a CRT monitor, the illumination serves as a point light source. Due to the closeness, the difference in distance causes a large unevenness of illumination, and a near subject strongly causes halation in one screen, and a far subject is dark and cannot see anything.

As a means for solving the above phenomenon,
An image signal correction circuit using a two-dimensional filter as disclosed in Japanese Patent Laid-Open No. 62-132479 has been proposed.

[0005]

In the image signal correction circuit described above, the luminance signal of the image signal is logarithmically compressed so that an image having a wide dynamic range can be displayed, and a two-dimensional filter is used to lower the spatial frequency range. The effect of illumination unevenness is reduced through a series of operations of suppressing and then converting to exponential characteristics. The low-pass suppression by this two-dimensional filter is very effective when the amount of illumination light changes gently like an endoscope.

However, in general images other than endoscopes, there are many so-called edge portions such as when the illumination light is stepwise or near the outline of the subject. When the above processing is performed,
The edges are unnaturally emphasized and become unsightly.

The image signal processing apparatus of the present invention has been made in view of such a problem, and its purpose is to make the edges of the image unnatural even in the place where the luminance changes stepwise. An object of the present invention is to provide an image signal processing device capable of removing or suppressing uneven illumination without emphasizing.

[0008]

In order to solve the above-mentioned problems, the image signal processing apparatus of the present invention comprises a logarithmic compression means for logarithmically compressing an image signal including a color signal, and filtering for the logarithmically compressed image signal. And a filtering characteristic setting means for adaptively setting a bandpass characteristic of the filtering means according to the characteristics of the image signal, and a dynamic range / gain control for controlling the dynamic range and the gain of the filtered signal. And gain control means.

[0009]

That is, according to the present invention, the input image signal is logarithmically compressed and then filtered to be converted into an exponential characteristic. During this filtering, the band pass characteristic is adaptively controlled according to the characteristics of the image signal. As a result, where the brightness changes stepwise, higher frequencies are suppressed by filtering, and not only low frequency illumination unevenness but also stepwise illumination unevenness is suppressed without being unnatural. .. Therefore, an image with wide latitude and good contrast is displayed on the monitor.

[0010]

FIG. 1 is a block diagram of a first embodiment in which the image signal processing apparatus of the present invention is applied to an electronic camera capable of picking up an object having a wide dynamic range by using cumulative addition. Reference numeral 1 is an optical lens, and 2 is a solid-state image sensor for converting an image formed by the optical lens into an electric signal.
The image signal read from the solid-state image sensor 2 is amplified to a predetermined level by the preamplifier 3 and then digitally encoded by the A / D converter 4.

The digitally encoded image signal is added by an adder 5
Are sequentially stored in the frame memory 6 for each frame via. The adder 5 adds the image signal stored and stored in the frame memory 6 and the image signal of the next one frame sequentially read out from the solid-state image pickup device 2 and re-writes it in the frame memory 6. By doing so, the sequentially read image signals are cumulatively added to the frame memory 6. By repeating such cumulative addition of the image signal a predetermined number of times, the dynamic range of the image signal can be expanded. Regarding the method of expanding the dynamic range in this imaging, Japanese Patent Application No. 1-3450
Detailed description is given in No. 8, so it is omitted here.

If cumulative addition is used as described above, for example, 5
Using the solid-state image sensor 2 having a dynamic range of 0 dB, a subject having a dynamic range of 80 dB can be imaged.

In order to display the 80 dB image signal thus obtained on an image monitor such as a TV receiver, the dynamic range (for example, 46 dB) of the image monitor is used.
The dynamic range of the image signal must be compressed in accordance with the above.

For this purpose, the video processor 7 divides the image signal into a luminance signal Y and color signals R, G, B, and D /
After the analog signals are converted by the A converters 17 _y , 17 _r , 17 _g , and 17 _b , the luminance signal Y is logarithmically converted by the log amplifier 8 (log Y), and the uneven illumination is removed by the adaptive filter 9 ( log Y ′), the dynamic range / gain controller (DGC) 10 adjusts the dynamic range of the image according to the dynamic range (46 dB) of the image monitor with the coefficient α, and the average value is subtracted with the coefficient log β to lower the level. The gain that amplifies the image signal level is adjusted, and the output, log βY ′ ^α, is obtained. This signal, log βY ′
^α is used in the subtracter 12 for the filter 9 and the DGC.
The signal log Y which has passed through the delay circuit 11 for the same time as that processed in 10 is subtracted. From this, a signal having a compression coefficient for multiplying the color signal as shown in Expression 1 is created.

[0015]

[Equation 1] The adaptive filter 9 comprises an edge detection circuit 9a, a characteristic selection circuit 9b ₁ and a filter 9c ₁ . Adaptive filter 9
FIG. 2 shows a specific example of the above. The operation of the adaptive filter 9 will be described with reference to FIG.

In FIG. 2, the output from the log amp 8, log Y
(= Y (x), where x is the horizontal position coordinate of the pixel) is delayed by four pixels via the delay circuits 910 to 913 for one pixel. Let this signal be y (x ₀ ). In the adder 920, the signals of a total of 9 pixels, which are four pixels before and after the pixel y (x ₀ ) output to the delay circuit 913, are added together, and multiplied by 1/9 in the multiplier 924. That is, it is a local average value of signals of 9 pixels.

Similarly, a pixel y (x ₀ ) and three pixels before and after the pixel y (x ₀ ) in the adder 921 and the multiplier 925, a local average value of signals of a total of 7 pixels, and a pixel y in the adder 922 and the multiplier 926.
(X ₀ ) and 2 pixels each before and after it, a total local average value of signals of 5 pixels in total, and an adder 923 and a multiplier 927 obtain local average values of signals of pixel y (x ₀ ) and 3 pixels before and after it. Which are input to the selector 929.

On the other hand, the edge detection circuit 9 surrounded by a broken line
At a ₁ , the central pixel y output from the delay circuit 913
(X ₀ ) is doubled by the multiplier 902, and the pixels before and after that are
Multipliers 903 and 901 respectively multiply by -1 to adder 90
Add with 4. The output of the adder 904 is the absolute value unit 90
An absolute value is taken at 5 and this is used as an edge signal e (x) representing the size of the edge. The operation of this edge detection circuit is to perform a convolution operation on the luminance signal y (x) using the coefficients (-1, 2, -1) and take the absolute value thereof. If this is expressed by a formula, it becomes like Formula 2.

[0019]

[Equation 2] Here, when e (x) is large, there is a large edge, and when it is small, there is a small change in luminance. e (x) is input to the characteristic selection circuit 9b ₁ . The characteristic selection circuit 9b ₁ sends a signal to the selector 929 according to the size of e (x). The selector 929 can select four kinds of average values of local average widths 3, 5, 7, and 9. A signal is sent from the characteristic selection circuit 9b ₁ to the selector 929 such that the local average width is narrowed when e (x) is large and the local average width is widened when e (x) is small. The output of the selector 929 is multiplied by a in the multiplier 928 (where 0 ≦ a ≦ 1, where a = 0.5), and then in the subtractor 930, the output y (x ₀ ) from the delay circuit 913 is used. Subtracted, logY ′ = y ′ (x)
Is sent to the DGC 10. When the local average width is represented by w pixels, the operation of the above-described series of filters can be expressed by Equation 3.

[0020]

[Equation 3] That is, an average value subtraction type filter that subtracts the average value of the pixels up to the local average width before and after the pixel value y (x) is configured.

The coefficient a by which the local mean value is multiplied determines the strength of the effect of the filter. The following Equation 4 can be obtained by expressing the Equation 1 by a convolution operation with the rect function.

[Equation 4] FIG. 3 shows δ (x)-(a / w) rect (x / w).

Further, when the equation 4 is subjected to the Fourier transform, the following equation 5 is obtained.

[0023]

[Equation 5] FIG. 4 shows 1-a · sinc (wu). Where u is the spatial frequency.

FIG. 4 shows the characteristics of this filter.
It can be seen that it has the characteristic of suppressing the low frequency range where uneven illumination exists. The larger the coefficient a, the stronger the suppression.

Further, by suppressing the local average width w to be smaller such that w '= w / 2, the frequency to be suppressed can be expanded to the high frequency range.

FIG. 5 shows changes in the characteristics of the filter with changes in w. In FIG. 5, a = 1 and the horizontal axis is normalized with the Nyquist frequency being 1.0.

According to FIG. 5, as w becomes smaller, the mid range is suppressed at the normalized spatial frequency.
By the way, the frequency components forming the image feature are often distributed in the vicinity of the middle range of the normalized spatial frequency. In addition, uneven illuminance due to illumination of the entire subject, so-called illumination unevenness, is often distributed in the low range of the spatial frequency. Therefore, in order to reduce the uneven illumination, it is effective to set w to be large and suppress only low frequencies.

However, when an edge is present in the image, suppressing only the low frequency causes the edge to be overemphasized, resulting in an unnatural image.

FIG. 10 shows an image signal before filtering,
The image signals after filtering are superimposed and shown. In FIG. 10, the solid line is the signal before filtering, and there is a bright → dark edge from left to right. The broken line is the waveform after filtering with the local average width w = 9, and overshoot and undershoot are observed from the portion where there is no brightness change to the vicinity of the edge. The alternate long and short dash line shows the waveform after filtering when w = 3. Compared with w = 9, the overshoot and undershoot are smaller, and it can be seen that filtering is performed without disturbing the original waveform.

As described above, w is set to a large value in a place where there is no edge, the frequency components of the image are stored, only the low band due to uneven illumination is suppressed, and w is set to be small in the vicinity of the edge. It is possible to effectively suppress the uneven illumination without causing excessive emphasis of.

As described above, due to the functions of the edge detection circuit 9a ₁ , the characteristic selection circuit 9b ₁ , and the selector circuit 929, when there is a large edge, the filter with a small local average width is selected to suppress the high frequency and the Reduce the adverse effects of. When there are no edges, a filter having a large local average width is selected to suppress only low frequencies. In this embodiment, the maximum value of the local average width w of the filter is set to 9, and the variable value of w is set to 4 steps. However, the maximum value of w may be further increased and the variable step may be increased.

The above equation 3 can be rewritten as the following equation 6.

[0033]

[Equation 6] This k (i) is shown in FIG. In addition, this k (i) is shown in FIG.
When a spline function such as 4 or a sinc function (not shown) is used, ripple as seen in the frequency characteristic of the filter of FIG. 5 is suppressed.

As can be seen from FIG. 4, the zero frequency of the image signal filtered as described above has changed, and the gain is adjusted by the DGC 10 so that the gain is good on the monitor display.

On the other hand, the color signals R, G, B separated by the video processor 7 in FIG. 1 and converted into analog signals by the D / A converters 17 _r , 17 _g , 17 _b are respectively log amplifiers 13 _r , The delay circuit 1 is logarithmically converted by 13 _g and 13 _b.
4 _r, 14 _g, 14 _b in is delayed by a time until the compression factor is made, the adder 15 _r, 15 _g, 15 _b
In, the compression coefficient is added. This is R
Taking a signal as an example, it is the same as calculating the following Equation 7.

[0036]

[Equation 7] Since this is further inversely logarithmically converted by the inverse log amplifiers 16 _r , 16 _g , and 16 _b , R · βY ′ ^α / Y (in the case of R signal), and the compression coefficients βY ′ ^{α are} added to the color signals R, G, and B. /
You have multiplied Y.

In this way, the dynamic range is compressed and displayed on the monitor.

As described above, by the series of operations of the adaptive filter 9, a filter for suppressing higher frequencies is selected according to the size of the edge portion of the image signal, so that unnatural edge enhancement does not occur. It is possible to eliminate illumination unevenness, reduce the compression ratio of the dynamic range by the DGC 10, and display a good image with contrast.

Next, a second embodiment in which the configuration of the edge detection circuit 9a in the adaptive filter 9 of the first embodiment is changed will be described with reference to FIG.

In this embodiment, multipliers 901 and 9 are used.
02, 903, adder 904, absolute value unit 905,
An edge signal e (x) is obtained as in the first embodiment. This signal represents the size of the difference from the adjacent pixel, that is, the size of the edge. The edge signal e (x) is delayed by the delay circuit 9
31 to 938, each pixel is delayed by one pixel. Edge signal e output from the delay circuit 934
E (x ₀ -4) ... e (x ₀ ) ... e (x ₀ +4) corresponding to four pixels before and after (x ₀ ) respectively.
Multipliers 941 to 948 for the nine e (x) of
The center load sum is obtained by the adder 949, and the center load sum is output to the characteristic selection circuit 9b as the edge information signal E (x). When the edge information detection width is S, E (x) is expressed by Eq.

[0041]

[Equation 8] However, C (i) is a function as shown in FIG. 7, and has a load coefficient that decreases as the distance from the center increases. In this embodiment, when a large edge exists at the position x ₁ within the edge information detection width S with respect to the position x of the pixel S = 9, the edge signal e (x ₁ ) becomes a large value. According to Equation 3, when x is close to x ₁ , the weighting factor becomes large, so E (x) becomes large, and x
When is far from x ₁ within the detection width S range, the weighting factor is set small, so E (x) becomes small. That is, the edge information signal E (x) also includes information on the near and far edges.

In the characteristic selection circuit 9b ₂ , the edge information signal E
According to the size of (x), a signal is sent to the selector 929 in the filter 9c ₂ to select the optimum local average width.
FIG. 15 shows the relationship between the magnitude of E (x) and the local average width.
That is, when E (x) is large, the local average width is small, and
When (x) is small, it operates to increase the local average width.

By the series of operations described above, it is possible to suppress the illumination unevenness of the entire image without unnaturally enhancing the vicinity of the edge, and a good image can be obtained.

According to this embodiment, the local average width of the filter becomes continuously smaller as the edge approaches,
A more natural-looking image can be obtained.

Although the edge detection width S is 9 in this embodiment, it may be wider or narrower. Further, although the coefficient of the load sum is configured to be linear, the coefficient is not limited to this and may be changed appropriately.

The edge information signal E (x) may be normalized by the average value of the image or the like. Next, a third embodiment in which the filter 9c in FIG. 1 has a simpler configuration will be described with reference to FIG.

Reference numeral 9a ₂ is an edge detecting circuit, which has the same structure as that described in the second embodiment. 9c ₃ surrounded by broken line
1 corresponds to the filter 9c in FIG. From the input signal y (x), the edge detection circuit 9a ₂ outputs the edge signal e (x) as described in the second embodiment.

On the other hand, in the filter 9c ₃ , each of the delay circuits 910 to 917 is delayed by one pixel. Letting the signal output from the delay circuit 913 be the central pixel y (x ₀ ), the delay circuits 912 and 9 before and after the center pixel y (x ₀ ).
The signals for three pixels including y (x ₀ −1) and y (x ₀ +1) output from 14 are added by the adder 954 and output to the adder 958. The adders 953 add the y (x ₀ -2) and y (x ₀ +2) output from the delay circuits 911 and 915, and the multiplier 957 adds the characteristic selection circuit 9 b.
After being multiplied by the value output from ₃ , it is sent to the adder 958. Y (x output from the delay circuits 910 and 916
₀₋₃ ) and y (x ₀ +3) are added up by the adder 952, multiplied by the value output from the characteristic selection circuit 9b ₃ by the multiplier 956, and then sent to the adder 958. The y (x ₀ -4) input to the delay circuit 910 and the output y (x ₀ +4) of the delay circuit 917 are added by the adder 951 and output from the characteristic selection circuit 9b ₃ by the multiplier 957. It is sent to the adder 958 after being multiplied by the value. In the adder 958, the above-mentioned four inputs are added and the multiplier 959 is added.
Then, the multiplier 959 multiplies the value output from the characteristic selection circuit 9b ₃ and outputs the product to the subtractor 930. Subtractor 9
At 30, the value of the output of the multiplier 959, which is the local average value, is subtracted from the output y (x ₀ ) from the delay circuit 913.

The characteristic selection circuit 9b ₃ outputs as shown in FIG. 16 according to the magnitude of the edge information signal E (x) from the edge detection circuit 9a ₂ . In FIG. 16, a
Is a coefficient of 0 ≦ a ≦ 1, whereby the effect of the filter can be adjusted.

As described above, the combination of the edge detection circuit 9a, the characteristic selection circuit 9b ₃ and the filter 9c ₃ provides a simpler structure.
It is possible to effectively remove the entire illumination unevenness without causing a bad influence in the vicinity of the edge and obtain a good image.

Next, a fourth embodiment in which the edge information signal E (x) and the edge direction signal ed (x) representing the direction of the edge (bright → dark or dark → bright) are input to the characteristic selection circuit State.

FIG. 9 shows the configuration of the fourth embodiment of the present invention. The filter 9c ₃ has the same structure as the filter 9c ₃ of the third embodiment shown in FIG. The edge detection circuit 9a ₄ is the edge detection circuit 9 of the second embodiment shown in FIG.
The output of the adder 904 of a ₂ is the edge direction signal ed (x)
It is what ed (x) is positive when the change in luminance at the edge is light → dark and is negative when dark → bright. ed (x) is delayed by 4 pixels in the delay circuit 960 to transmit information of the same pixel as E (x), and then
It is supplied to the characteristic selection circuit 9b ₄ together with E (x).

The characteristic selection circuit 9b ₄ changes the value of a between 0.3 and 0.7 so that it is large when the edge direction signal ed (x) is positive and small when it is negative, as shown in FIG. With respect to the input edge information signal E (x), the values shown in FIG. 16 are output to determine the characteristics of the filter.

FIG. 11 shows waveforms before and after filtering when w = 3. The solid line shows the waveform before filtering, and the broken line shows the waveform after filtering. The plot on the line is 1
Represents a pixel. Even if w = 3 is selected near the edge by the edge information signal E (x), large overshoot and undershoot appear in the filtered waveform when the edge becomes steep as shown in the figure. At the time of filtering as in this embodiment, the edge direction signal ed (x)
FIG. 1 shows a case in which the coefficient a of the filter is adaptively changed using
2 shows. In the figure, the edge direction signal e
d (x) has a large positive value, and a = 0.65 is selected. Further, in the pixel of Δ, ed (x) becomes negative, and a
= 0.4 is selected, overshoot,
Filtering can be performed while suppressing undershoot.

Next, the fifth embodiment of the present invention constituted by a digital circuit
An example will be described.

FIG. 18 shows a block diagram of this embodiment. In this figure, the configuration up to the video processor is the same as in FIG. The luminance signal Y output from the video processor 7 is subjected to illumination unevenness removal and luminance compression by a logarithmic circuit 8, an adaptive filter circuit 9, and a dynamic range gain controller 10. Reference numeral 20 is a subtractor, and 21 is an antilogarithmic circuit, which outputs coefficients to be multiplied by the R, G, and B signals. 22r, 22g, and 22b are multipliers, which are multiplied by a coefficient for compression.

Further, the adaptive filter circuit 9 includes a band setting circuit 23 for detecting illumination unevenness and a band variable LPF (Low
Pass Filtering circuit 24, a root mean square value detection unit 25 for generating a coefficient n to be multiplied by this output, a coefficient setting circuit 26, a multiplier 27, and a subtractor 28 for subtracting the detected illumination unevenness. Composed of.

Further, the band setting circuit 23 and the band variable LPF
The circuit 24 has the configuration shown in FIG. That is, the signal lo
An edge detection circuit 9a for detecting the edge signal of g Y
Further, the edge detection circuit 9a is composed of a characteristic selection circuit 9b, and the edge detection circuit 9a performs a two-dimensional Laplacian circuit, an absolute value detection circuit 30 for obtaining an absolute value, and an edge signal output from the absolute value detection circuit 30. e (x)
Is composed of a low-pass filter circuit 31 which performs two-dimensional low-pass filtering and outputs an edge information signal E (x). Further, the characteristic selection circuit 9b outputs, from the edge information signal E (x), W, which is a coefficient that determines the band in the band variable LPF, based on the characteristics shown in FIG. The band variable LPF 24 is composed of a Gaussian coefficient selection circuit 32 and a convolution circuit 33. The Gaussian coefficient selection circuit selects the coefficient of the following equation 9 according to W output from the characteristic selection circuit (see FIG. 21).

[0059]

[Equation 9] In this equation, c is an appropriate coefficient and A is obtained as the following equation 10.

[0060]

[Equation 10] It should be noted that although this formula is represented by one dimension of x, it is actually two dimensions. This A is a coefficient for adjusting the change in the brightness level due to the difference in W, and is set so that the sum of the coefficient values of the Gaussian coefficient becomes 1.

Since the band setting circuit selects a small W when there are many edges or near edges, the low pass effect in the band variable LPF is small. In addition, since the large W is selected when the edge is few or far, the low pass effect in the band variable LPF becomes large. The mean square value detection unit 2
5 and the coefficient setting circuit 26 obtain a coefficient n (n ≦ 1) by which the signal low-pass filtered by the band variable LPF is multiplied. For this coefficient n, the mean square value of the output log Y ′ of the adaptive filter circuit 9 is the coefficient setting circuit 2
The value R _power input to 6 is determined.

By multiplying this coefficient, the mean square value of log Y ', that is, the power becomes constant. That is, for a subject with large illumination unevenness, n is large and suppression of illumination unevenness is large, and for a subject with small illumination unevenness, n is small and suppression of illumination unevenness is weakened. When this adaptive filter is expressed by an equation, the following equation 11 is obtained, and the band characteristic is as shown in FIG.

[0063]

[Equation 11] That is, when the edge amount is small, W becomes large and only the low frequency component corresponding to the uneven illumination is suppressed. Also, when the edge amount is large, W becomes small, and it is suppressed to the middle range. In addition, n is changed according to the absolute magnitude of the unevenness of illumination, and adaptive filtering is performed so that the power of log Y'is constant. Here, since the Gaussian filter is used as the band variable filter, the change in the filter characteristic is a monotonous change, there is no ripple as shown in FIG. 5, and there is no concern that an artifact or the like will occur in the filtered image.

Further, the dynamic range controller 1
0 is the average value ave from the average value detection unit 34 and the set value R
A coefficient setting circuit 35 for setting the addition coefficient log β from _ave , a standard deviation detection unit 36, and a coefficient setting circuit 37 for setting the multiplication coefficient α from the standard deviation SD and the set value R _SD.
And a multiplier 38 and an adder 39. The coefficient setting circuits 35 and 37 generate coefficients such that the average value and the standard deviation of the output signal of the DGC 10 are R _ave and R _SD , respectively.

Next, the operation of this embodiment will be described. The digital signal cumulatively added and input is separated by the video processor 7 into luminance signals Y and R, G, and B signals. Here, it is assumed that the dynamic range of the luminance signal Y is about 75 dB. The adaptive filter circuit 9 detects the edge amount as described above, performs adaptive filtering according to this, and outputs log Y '. This log Y'is R so that the dynamic range is suppressed to about 50-60 dB.
_{The power} value is set. And log Y'is DGC
The signal is input to the circuit 10, and the dynamic range and the gain are adjusted by the already set R _ave and R _{SD so} that the dynamic range is compressed to 45 to 50 dB which is the dynamic range of the monitor. To be done.

[0066]

[Equation 12] The subtractor 20 and the antilogarithm circuit 21 allow the multiplier 2
Coefficients multiplied by 2r, 22g, 22b (= Y ′ ^α · β
/ Y) is output. Then, taking the R signal as an example, the following calculated value of Expression 13 is output.

[0067]

[Equation 13] The signal thus compressed in the dynamic range is displayed on the monitor.

As described above, the unevenness of illumination is suppressed by the adaptive filter 9, the dynamic range is compressed by the DGC 10, and the original image having a wide dynamic range can be displayed on the monitor as a good image with contrast.

Further, since the degree of suppression of illumination unevenness can be adjusted by R _power and the compression rate of DGC can be adjusted by R _ave and R _SD , these three parameters R can be adjusted as shown in FIG.
_By using the controller 40 that externally controls the _{powe r} , R _ave , and R _SD , it is possible to display a favorite image for each individual.

Further, as shown in FIG. 24, by using the pattern detector 41 that adaptively controls the R _{powe r} , R _ave , and R _SD according to the pattern and type of the object, any object can be detected. Good display can be performed. This pattern detection is performed by intelligent processing using, for example, a neural network.

Further, in this embodiment, since a Gaussian filter is used for the band-variable LPF, it is possible to carry out filtering having a filter characteristic in which ripples do not occur in the frequency plane and obtain a good image.

Further, in order to select the characteristic value W of the Gaussian coefficient from the edge information signal E (x), the characteristic shown in FIG. 20 is used in the present embodiment, but the characteristic is changed by connecting to the histogram of the edge information signal. May be.

A sixth embodiment of the present invention having a simpler circuit configuration will be described below. FIG. 25 shows a block diagram of this embodiment. The luminance signal Y output from the video processor 7 is subjected to illumination unevenness removal and luminance compression by a log amplifier 8, an adaptive filter circuit 9, and a dynamic range gain controller 10. Reference numeral 20 is a subtracter, and 21 is an antilogarithm circuit, which outputs coefficients to be multiplied by the R, G, and B signals. Reference numerals 22r, 22g and 22b denote multipliers, which are multiplied by a coefficient for dynamic range compression.

The adaptive filter circuit 9 detects the average level of the values of the pixels in the vicinity of the pixel of interest.
And HPF (High Pass Filter) for extracting high frequency components of images
er) 43 and a coefficient setting circuit 42 that sets a coefficient by which the output of the HPF 43 is multiplied by the output of the level detection circuit 40,
It comprises a multiplier 44 for multiplying the output of the coefficient setting circuit 42 by the output of the HPF 43, and an adder 45 for adding the output of the multiplier 44 to the output logY of the logarithmic circuit 8.

The level detection circuit 40 is a convolution filter, and performs a convolution operation on the output logY from the log amp 8 using a 3 × 3 coefficient as shown in FIG.

The HPF 43 can be realized by a convolution filter using 3 × 3 coefficients as shown in FIG.

LogY output from the log amp 8 is
The level detection circuit 40 detects the average level in the vicinity of the target pixel, and from the output thereof, the coefficient setting circuit 42
The coefficient for that pixel is determined. Coefficient setting circuit 42
Then, a large value for pixels with a large average level,
A small value is output for a pixel with a small average level.

On the other hand, logY obtained by HPF43
Is multiplied by the coefficient obtained by the coefficient setting circuit 42, and the adder 45 adds the output logY of the log amp 8 to the output of the multiplier 44 and outputs the result.

The output is the coefficient α,
The dynamic range is controlled by logβ. At this time, with respect to the luminance signal Y to control the dynamic range by beta · Y ^alpha becomes input-output characteristics. FIG. 28 shows high-frequency input / output characteristics when the dynamic range is compressed. When compressing the dynamic range,
As shown in FIG. 28, the high brightness level has a larger compression ratio than the low brightness level, and the contrast at the high brightness level is lowered. However, in advance, the adaptive filter circuit 9
Since the high frequency component of the high brightness level is emphasized, a good image is displayed on the monitor without a decrease in contrast when the dynamic range is compressed.
Further, since the window width of the convolution filter used is as small as 3 × 3, the influence of overshoot and undershoot near the edge is also small.

Further, in order to further simplify the circuit configuration, the level detection circuit 40 in FIG. 25 is omitted, and the output of the log amplifier 8 is directly input to the coefficient setting circuit 42.
A configuration such as 9 may be used.

According to the present embodiment, by emphasizing the high frequency component of the high luminance level before the dynamic range compression, the same effect as the illumination unevenness suppression can be achieved without causing the contrast reduction of the image after the dynamic range compression. Even if the adaptive filter circuit 9 is a two-dimensional filter, it can be realized with a simple configuration, and the device can be manufactured at low cost.

Further, a seventh embodiment of the present invention for more adaptively setting a coefficient for emphasizing a high frequency component will be described.

FIG. 30 shows a block diagram of this embodiment. The adaptive filter 9 of this embodiment comprises a level detection circuit 40, an adaptive coefficient setting circuit 46, an HPF 43, a multiplier 44, and an adder 45. Of these, the components other than the adaptive coefficient setting circuit 46 are shown in FIG.
Since it has the same function as that of the above embodiment, its explanation is omitted here.

The adaptive coefficient setting circuit 46 converts the average level output from the level detection circuit 40 into a high-frequency weighting coefficient. At this time, the conversion table is switched using the coefficients α and logβ obtained by the DGC 10.

FIG. 31 shows the input / output characteristics when α changes.

As shown in FIG. 31, when α is smaller than 1, the high frequency component of the input signal having a high average level is compressed at a higher compression rate than the high frequency component having a low average level. A conversion table that outputs a large coefficient value when the output of the level detection circuit 40 is large and a small coefficient value when the output of the level detection circuit 40 is small is selected.

However, when α takes a value close to 1, there is no difference between where the compression rate is high and where the average level is low. In that case, the same value is output regardless of the magnitude of the output of the level detection circuit 40.

As such a coefficient, for example, f
It is possible to use k / f '(Y), which is obtained by multiplying the reciprocal 1 / f' (Y) of the differential coefficient of (Y) = β · Y ^{α by} an appropriate constant k.

According to this embodiment, even if the coefficient at the time of dynamic range compression varies depending on the image, the high frequency is always emphasized to the same extent from the low luminance region to the high luminance region, so that a good image with contrast can be obtained. Be done.

In the above embodiment, the filtering characteristic is changed according to the position of each pixel in one image according to the characteristics of the image signal in one image. However, for image signals of a plurality of different images, The filtering characteristics may be changed according to the characteristics of each image.

In each embodiment, the filter and the edge detection circuit have a one-dimensional structure, but of course, a two-dimensional structure may be used. Although the configuration of the two-dimensional filter becomes very complicated, it can be relatively easily realized by using a DSP for image signal processing.

[0092]

As described above, according to the present invention,
An image with improved latitude can be obtained by effectively reducing the influence of uneven illumination without causing unnatural enhancement in the edge portion of the image.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a first embodiment in which the present invention is applied to an electronic camera.

FIG. 2 is a diagram showing a specific configuration example of the adaptive filter of FIG.

FIG. 3 is a diagram showing a relationship between horizontal position coordinates of pixels and a luminance signal.

FIG. 4 is a diagram showing characteristics of a filter.

FIG. 5 is a diagram showing a change in filter characteristics due to a change in local average value.

FIG. 6 is a second embodiment of the present invention in which the configuration of the edge detection circuit is changed.
FIG.

FIG. 7 is a diagram showing an example of weighting coefficients.

FIG. 8 is a third embodiment of the present invention in which the filter has a simpler configuration.
FIG.

FIG. 9 is a diagram showing a fourth embodiment of the present invention.

FIG. 10 is a diagram showing an image signal before filtering and an image signal after filtering in an overlapping manner.

FIG. 11 is a diagram showing waveforms before and after filtering when w = 3.

FIG. 12 is a waveform diagram when the filter coefficient is adaptively changed using an edge direction signal.

FIG. 13 is a diagram showing an example of a rect function.

FIG. 14 is a diagram showing an example of a spline function.

FIG. 15 is a diagram showing the relationship between the magnitude of the edge information signal and the local average width.

FIG. 16 is a diagram showing a relationship between an edge information signal and an output of each multiplier.

FIG. 17 is a diagram showing the relationship between edge direction signals and coefficients.

FIG. 18 is a configuration diagram showing a fifth embodiment of the present invention.

FIG. 19 is a diagram showing configurations of a band setting circuit and a band variable LPF circuit.

FIG. 20 is a diagram showing characteristics output from a characteristic selection circuit.

FIG. 21 is a diagram showing coefficients k (x) selected by a Gaussian coefficient setting circuit.

FIG. 22 is a diagram showing band characteristics of an adaptive filter.

FIG. 23: Three parameters, R _power , R _ave and R _SD
The figure which shows the controller which controls from outside.

FIG. 24 is a view showing a pattern detector.

FIG. 25 is a diagram showing a circuit configuration of a sixth embodiment of the present invention.

FIG. 26 is a diagram showing an example of coefficients used for convolution calculation.

FIG. 27 is a diagram showing an example of coefficients used for another convolution calculation.

FIG. 28 is a diagram showing high-frequency input / output characteristics when the dynamic range is compressed.

FIG. 29 is a diagram showing a circuit configuration in which the circuit configuration of FIG. 25 is further simplified.

FIG. 30 is a diagram showing a circuit configuration of a seventh embodiment of the present invention.

FIG. 31 is a diagram showing input / output characteristics when the coefficient α changes.

[Explanation of symbols]

1 ... Optical lens, 2 ... Solid-state image sensor, 3 ... Preamplifier,
4 ... A / D converter, 5 ... adder, 6 ... frame memory,
7 ... Video processor, 8 ... Log amplifier, 9 ... Adaptive filter, 9a ... Edge detection circuit, 9b ... Characteristic selection circuit, 9
c ... Filter, 10 ... Dynamic range / gain controller (DGC).

Claims (1)

[Claims] 1. A logarithmic compression means for logarithmically compressing an image signal including a color signal, a filtering means for filtering the logarithmically compressed image signal, and a bandpass characteristic of the filtering means adaptively according to the characteristics of the image signal. 2. An image signal processing device, comprising: a filtering characteristic setting means for setting the dynamic range and gain of the filtered signal; and a dynamic range / gain control means for controlling the dynamic range and gain of the filtered signal.