JP5365878B2 - Filter device and filtering method - Google Patents

Description

  The present invention relates to a filter device that reduces noise in a digital image in an image processing device, and in particular, even when a pixel of interest includes a noise component, an edge portion is retained and a flat portion is sufficiently removed with noise. The present invention relates to a filter device and a filtering method that can perform the above.

  Conventionally, Gaussian filters and median filters have been widely used for the purpose of reducing noise in digital images taken by an imaging device such as a digital camera. The Gaussian filter calculates, for example, an average value based on pixel values weighted according to the distance from the target pixel in a 5 × 5 pixel block region centered on the target pixel in the digital image, and the value of the target pixel. By doing so, mainly random noise can be removed.

  The median filter, for example, rearranges all the pixels in the 3 × 3 pixel block area centering on the pixel of interest in the digital image in descending order, and sets the median value (median value) to the pixel of interest. By setting this value, it is possible to mainly remove impulsive noise.

  In recent years, in addition to these filters, as described in Patent Document 1, an ε (epsilon) filter has been used. The ε filter calculates, for example, the difference in pixel value from the target pixel for all peripheral pixels included in the 5 × 5 pixel block region centered on the target pixel in the digital image, and the absolute value is preset. In this filter, the pixel value of the pixel of interest is a value obtained by averaging the pixel values of pixels smaller than the determined ε value. The ε filter can remove random noise while maintaining the edge components in the digital image more than the Gaussian filter.

JP 2008-124976 A

  The ε filter can remove noise with less image degradation than a Gaussian filter or a median filter. However, image noise has a property that it is more conspicuous in a flat part such as a wider blue sky than an edge part. If an ε value is set so that noise can be removed well in a flat part of the image, the edge part If the ε value is set such that the edge portion is blurred and the edge portion is blurred, and the edge portion is not blurred, there is a problem that noise is not sufficiently removed in the flat portion and noise becomes conspicuous.

  In Patent Document 1, it is noted that the influence of noise increases as the pixel value increases, that is, the brighter part, and by setting the ε value to increase as the pixel value increases, It is described that noise is well removed in a bright part and overcorrection is prevented in a dark part.

  However, in the technique described in Patent Document 1, since the edge portion and the flat portion are uniformly smoothed if the brightness is the same, noise removal of the dark flat portion is not sufficient, or the bright edge portion is blurred. May occur.

  In addition, in the conventional filter device using the ε determination value including the technique described in Patent Document 1, if the target pixel itself contains a noise component, an inappropriate ε value is set. There is also a problem.

  Accordingly, an object of the present invention is to provide a filter device and a filtering method capable of retaining an edge portion and performing a sufficient noise removal on a flat portion even when a pixel of interest includes a noise component. To do.

  In order to solve the above-described problem, the filter device according to the first aspect of the present invention includes a pixel value of a target pixel and a pixel value of each peripheral pixel in a pixel block including the target pixel and a plurality of peripheral pixels. And a filter device that uses a value obtained based on a pixel value of a surrounding pixel whose absolute value is smaller than a determination value as a pixel value of the target pixel, and includes a plurality of pixels in the pixel block. Edge evaluation value obtained based on the mutual relationship between the subarea average calculation unit that sets the subarea and calculates the average value of the pixel values of each subarea, and the calculated average value for each subarea An edge evaluation value calculation unit that outputs a flat part noise evaluation value that outputs a flat part noise evaluation value obtained based on an average deviation of pixel values of each pixel in the pixel block, and the edge evaluation value Preliminary using said flat portion noise evaluation value characterized by comprising a setting unit that sets the determination value.

  Here, the edge evaluation value calculation unit converts the average value of each sub-area into an index value based on a predetermined edge evaluation formula, and when the index value exceeds a predetermined threshold, The edge evaluation value is output, and when the predetermined threshold value is not exceeded, 0 can be output as the edge evaluation value.

  In addition, the sub-area average calculating unit includes a first sub-area that is a central region including the target pixel in the pixel block, and an entire region excluding the first sub-area from the pixel block or the pixel block The second sub-area as a whole is set, and the edge evaluation value calculation unit can use the absolute value of the difference between the average values of both sub-areas as the index value.

  The setting unit may set the determination value to be small according to the edge evaluation value and to be large according to the flat portion noise evaluation value.

  In order to solve the above-described problem, the filtering method according to the second aspect of the present invention calculates, for a plurality of pixels constituting a pixel block including a target pixel, a difference in pixel value from the target pixel, A filtering method in a filter device in which a value obtained based on a pixel value of a pixel whose absolute value is smaller than a determination value is the pixel value of the target pixel, wherein a plurality of subareas are set in the pixel block. An edge evaluation value calculation that calculates an edge evaluation value based on a sub-area average calculation step for calculating an average value of pixel values of pixels included in each sub-area and an average value for each calculated sub-area A flat part noise evaluation value calculating step for calculating a flat part noise evaluation value based on an average deviation of pixel values of each pixel in the pixel block; and And having a setting step of setting the determination value by using the di-evaluation value and the flat portion noise evaluation value.

  According to the present invention, it is possible to provide a filtering device and a filtering method capable of retaining an edge portion and performing sufficient noise removal on a flat portion even when a pixel of interest includes a noise component.

It is a block diagram which shows the structure of the nonlinear filter apparatus which concerns on this embodiment. It is a block diagram which shows the structure of an epsilon filter part. It is a block diagram which shows the structure of an epsilon value calculating part. It is a figure which shows the example of a subarea setting pattern. It is a figure which shows the example of the evaluation type | formula defined for every subarea setting pattern. It is a flowchart explaining the process sequence of an edge evaluation part. It is a flowchart explaining the process sequence of a flat part noise evaluation part. It is a flowchart explaining the process sequence of a correction | amendment part. It is a figure explaining the specific effect of this embodiment, and is a figure showing an original picture. It is a figure explaining the specific effect of this embodiment, and is a figure showing the image processing result using a moving average filter. It is a figure explaining the specific effect of this embodiment, and is a figure which shows the image processing result which fixed the (epsilon) value and used the (epsilon) filter. It is a figure explaining the specific effect of this embodiment, and is a figure showing the image processing result using the ε filter of this embodiment. It is a figure explaining the specific effect of this embodiment, and is a comparison figure which shows PSNR (Peak S / N Ratio), noise removal performance, and the preservability of an edge.

  Embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the nonlinear filter device according to this embodiment. The nonlinear filter device is used in, for example, an image processing device mounted on an imaging device such as a digital camera, an information processing device that functions as an image processing device, and the like.

  As shown in the drawing, the nonlinear filter device 10 includes an ε value calculation unit 100 and an ε filter unit 200. These functional units may be realized by software or hardware. The non-linear filter device 10 is a device that inputs an m × n pixel block including a target pixel, corrects the target pixel by an ε filter, and outputs the corrected pixel. In the digital image to be processed, a pixel block obtained by moving the target pixel is sequentially input to the nonlinear filter device 10, and the target pixel is replaced with the output value, thereby applying an ε filter to the digital image to be processed. Can do.

  The ε value calculation unit 100 receives an m × n pixel block including the pixel of interest and a preset standard ε value, and calculates and outputs an ε value applied to the pixel of interest. In order to perform this process, the ε value calculation unit 100 includes an edge evaluation unit 110, a flat part noise evaluation unit 120, and a correction unit 130.

  In the present embodiment, the edge evaluation unit 110 determines whether or not the pixel of interest in the m × n pixel block corresponds to the edge portion. When it is determined that the target pixel corresponds to the edge portion, the correction unit 130 corrects the standard ε value so that the ε value becomes small. As a result, blurring can be prevented and the edge portion can be held.

  The flat part noise evaluation unit 120 evaluates the noise level of the m × n pixel block. Then, the correction unit 130 corrects the standard ε value so that the ε value increases according to the evaluation of the noise level. Thus, sufficient noise removal can be performed particularly in a flat portion where noise is conspicuous.

  The ε filter unit 200 receives the m × n pixel block including the target pixel and the ε value calculated by the ε value calculation unit 100, and performs ε filter processing on the target pixel. Then, the target pixel on which the ε filter processing has been performed is output.

FIG. 2 is a block diagram showing a configuration of the ε filter unit 200. Note that the m × n pixel block includes the target pixel X (x, y), the peripheral pixel X ₁ (x ₁ , y ₁ ), the peripheral pixel X ₂ (x ₂ , y ₂ ), and the peripheral pixel X _k (x _k , y _k ). _However, an _{x 1 = x-m / 2} , x 2 = x-m / 2 + 1 ... x k = x + m / 2-1, y 1 = y-n / 2, y 2 = y-n / 2 + 1 ... y _k = y + n / 2-1.

  As shown in the figure, the ε filter unit 200 includes an ε determination unit 210, an addition unit 220, a buffer 230, a counter 240, and a division unit 250.

The ε determination unit 210 compares the absolute value of the difference between each peripheral pixel (x ₁ , x ₂ ... x _k ) constituting the m × n pixel block and the pixel of interest with the ε value, and the absolute value of the difference is ε A pixel value of a peripheral pixel smaller than the value is output as valid. In the present embodiment, m × n−1 ε determination units 210 are provided, and ε determination processing is performed in parallel for all the peripheral pixels.

  The addition unit 220 adds the pixel values output as valid by the ε determination unit 210, and the buffer 230 stores the addition result of the addition unit 220 at a time. The counter 240 counts the number of pixel values output as valid by the ε determination unit 210.

  The division unit 250 divides the counter value of the counter 240 from the addition result stored in the buffer 230, thereby obtaining an average value of the pixel values of the peripheral pixels whose absolute value of the difference from the pixel value of the target pixel is smaller than the ε value. Calculate and output as ε filter result.

  By inputting an m × n pixel block including the target pixel to the ε filter unit 200, an ε filter result based on the determination value ε for the target pixel can be obtained. The configuration of the ε filter unit 200 is not limited to this example, and a known ε filter technique can be applied.

  FIG. 3 is a block diagram illustrating a configuration of the ε value calculation unit 100. As shown in the figure, the ε value calculation unit 100 includes an edge evaluation unit 110, a flat part noise evaluation unit 120, and a correction unit 130.

  In general, in the ε filter, if the ε value is set to an appropriate value that is sufficiently smaller than the amplitude of the edge portion of the original image and sufficiently larger than the amplitude fluctuation between the target pixel and the surrounding pixels due to random noise, the original value is theoretically determined. Random noise can be reduced without causing blur at the edge of the image. In this embodiment, the edge evaluation unit 110 evaluates whether the pixel of interest corresponds to an edge, the flat part noise evaluation unit 120 evaluates the noise level of the pixel block, and the ε value is determined according to each evaluation. Correct to an appropriate value.

  As shown in the figure, the edge evaluation unit 110 includes a sub-area average calculation unit 111 and an edge evaluation value calculation unit 112.

  The sub-area average calculation unit 111 sets a plurality of sub-areas in the m × n pixel block, and calculates the average value of the pixel values included in each sub-area.

  FIG. 4 is a diagram illustrating an example of a subarea setting pattern. In this figure, three patterns are shown for an example in which a plurality of subareas are set in a 5 × 5 pixel block. The subarea average calculation unit 111 calculates the average value of the pixel values of the pixels included in each subarea based on any of the subarea setting patterns. However, the subarea setting pattern shown in this figure is an example, and other subarea setting patterns may be adopted.

  FIG. 4A shows a subarea setting pattern 1. As shown in this figure, in the subarea setting pattern 1, a subarea 1 composed of three pixels in the central region including the target pixel and a subarea 2 composed of pixels in the peripheral region are set. The sub area 2 may include all pixels of the 5 × 5 pixel block including the pixels of the sub area 1.

  FIG. 4B shows a subarea setting pattern 2. As shown in this figure, in the sub-area setting pattern 2, each sub-area is a 3 × 3 rectangular area, sub-area 1 is at the upper left, sub-area 2 is at the upper right, sub-area 3 is at the lower left, and sub-area 4 Is located at the bottom right.

  FIG. 4C shows the sub area setting pattern 3. As shown in the figure, in the sub-area setting pattern 3, the sub-area 1 is a central region including the pixel of interest, and the sub-areas 2 to 5 are each formed in a triangular shape including six pixels. Area 2 is arranged at the upper left, sub area 3 at the upper right, sub area 4 at the lower left, and sub area 5 at the lower right.

  In this way, by setting multiple sub-areas in the pixel block and calculating the average value of each, it is possible to reduce the effect of noise on the pixel of interest itself, and set an appropriate ε value. Will be able to.

  The edge evaluation value calculation unit 112 calculates and outputs an edge evaluation value based on the average value for each subarea calculated by the subarea average calculation unit 111. The edge evaluation value is a value obtained by using a predetermined evaluation formula for the mutual relationship between the average values for each sub-area. The evaluation formula is a formula for calculating an index value that is a degree that the target pixel can be regarded as corresponding to the edge portion, and is determined for each sub-area setting pattern.

  FIG. 5 is a diagram illustrating an example of an evaluation formula defined for each subarea setting pattern. FIG. 5A shows an evaluation formula corresponding to the sub-area setting pattern 1. In the evaluation formula corresponding to the subarea setting pattern 1, the absolute value of the difference between the average value of the subarea 1 and the average value of the subarea 2 is used as the index value.

  FIG. 5B shows an evaluation formula corresponding to the subarea setting pattern 2. In the evaluation formula corresponding to the sub-area setting pattern 2, different calculations are performed for the horizontal direction, the vertical direction, the right 45 degree direction, and the left 45 degree direction, and the largest value is used as the index value. That is, in the evaluation formula corresponding to the subarea setting pattern 2, the edge direction can also be determined.

  FIG. 5C shows an evaluation formula corresponding to the sub-area setting pattern 3. In the evaluation formula corresponding to the sub-area setting pattern 3, different calculations are performed for the horizontal direction, the vertical direction, the right 45 degree direction, the left 45 degree direction, and the impulse, and the largest value is used as the index value. In other words, the evaluation formula corresponding to the sub-area setting pattern 3 can also determine the direction of the edge including the impulsive edge.

  Further, the edge evaluation value calculation unit 112 determines whether or not the index value calculated by the evaluation formula exceeds a predetermined edge level threshold, and if it exceeds, the target pixel corresponds to the edge part. Assume that an index value is output as an edge evaluation value. On the other hand, if the index value does not exceed the edge level threshold value, the target pixel is not an edge portion, and 0 is output as the edge evaluation value.

  Returning to the description of FIG. 3, the flat part noise evaluation unit 120 includes an average deviation calculation unit 121 and a flat part noise evaluation value calculation unit 122.

The average deviation calculation unit 121 calculates an average deviation of pixel values of pixels included in the m × n pixel block. Here, the average deviation is an average of absolute values of deviations from the average value of each pixel value, and can be calculated by the equation shown in [Equation 1]. Here, x is a pixel value, x _Avg is a pixel average value of the pixel block, and m × n is the number of pixels in the pixel block.
The flat part noise evaluation value calculation unit 122 determines whether or not the average deviation calculated by the average deviation calculation unit 121 exceeds a predetermined noise level threshold value. The average deviation is output as the flat part noise evaluation value. On the other hand, if the average deviation does not exceed the flat part noise level threshold, it is assumed that there is not much noise in the pixel block, and 0 is output as the flat part noise evaluation value.

  As illustrated in FIG. 3, the correction unit 130 includes a coefficient multiplication unit 131, an ε value correction unit 132, and an ε limiter 133.

  The coefficient multiplier 131 calculates an edge correction value by multiplying the edge evaluation value by a predetermined edge correction coefficient, and also multiplies the flat part noise evaluation value by a predetermined flat part correction coefficient. Thus, the flat part noise correction value is calculated, and each correction value is output.

  In this embodiment, the edge portion correction coefficient is a value between 0 and 1.0, and is based on the noise characteristics of the image sensor and an image processing policy such as whether importance is placed on noise removal or edge retention. Can be determined. For example, if you want to set a small ε value in the edge part to hold the edge, set a value close to 1.0. If you want to reduce the ε value fluctuation in the edge part, set a value close to 0. To do.

  In this embodiment, the flat part correction coefficient is a value between 0 and 1.0, and image processing such as whether noise characteristics of the image sensor and noise removal of the flat part are important or texture is important. Can be determined based on policy. For example, if sufficient noise removal is desired in the flat part, a value close to 1.0 is set to obtain a larger ε value, and if it is desired to preserve the texture even in the flat part, a smaller ε value is set. Therefore, a value close to 0 is set.

  The flat part noise evaluation by the flat part noise evaluation unit 120 may be performed only for pixel blocks whose edge evaluation value output by the edge evaluation unit 110 is 0 and not determined as an edge part. In the present embodiment, the flat part noise evaluation by the flat part noise evaluation unit 120 is performed regardless of the edge evaluation value. This prevents noise removal from becoming insufficient due to erroneous detection of the edge portion.

  The ε correction unit 132 sets the ε value by subtracting the edge correction value from the predetermined standard ε value and adding the flat portion noise correction value. That is, if the edge portion is inconspicuous, the correction is made so that the ε value becomes small in order to reduce the filtering effect and retain the edge portion. Further, if there is a lot of noise, correction is made so that the ε value becomes large so that noise removal at the flat portion can be sufficiently performed.

  Here, the standard ε value can be determined based on the noise characteristics of the image sensor, the gain at the time of imaging, and the like. In this embodiment, the standard ε value is set by an external device of the nonlinear filter device 10 such as an imaging device. .

  However, the correction of the standard ε value is not limited to addition and subtraction, and may be performed by, for example, multiplying by a coefficient corresponding to the edge evaluation value and the flat part noise evaluation value. Also in this case, the coefficient is set so that the ε value decreases when the edge evaluation value is high, and the ε value increases when the flat part noise evaluation value is high.

  The ε limiter 133 prevents the ε value from exceeding the maximum value when the ε value exceeds a predetermined maximum value or falls below a predetermined minimum value as a result of correction of the ε value by the ε correction unit 132. Alternatively, the ε value is limited so that it does not fall below the minimum value. However, the ε limiter 133 may not be provided.

  Next, a processing procedure in the ε value calculation unit 100 of the present embodiment will be described. First, the processing procedure of the edge evaluation unit 110 will be described with reference to the flowchart of FIG.

  When the pixel block including the target pixel is input (S101), the edge evaluation unit 110 sets a plurality of subareas in the image block according to a predetermined subarea setting pattern (S102).

  Then, for each sub-area, the pixel average value of the pixels included in the sub-area is calculated (S103), and the calculation according to the evaluation formula corresponding to the sub-area setting pattern is performed using the calculated average value (S104). ).

  When one value is obtained by the evaluation formula as in the subarea setting pattern 1, the value is used as an index value, and a plurality of evaluation formulas are used as in the subarea setting pattern 2 and the subarea setting pattern 3. When the value is obtained, the largest value is set as the index value (S105).

  Next, it is determined whether the index value is larger than the edge level threshold (S106). If the index value is larger than the edge level threshold (S106: Yes), the index value is output as an edge evaluation value (S107). On the other hand, when the index value is not larger than the edge level threshold value (S106: No), 0 is output as the edge evaluation value (S108).

  Next, the processing procedure of the flat part noise evaluation part 120 is demonstrated with reference to the flowchart of FIG.

  When the flat part noise evaluation part 120 receives a pixel block including the target pixel (S201), the flat part noise evaluation part 120 calculates an average value of the pixel values in the pixel block (S202). Then, the average deviation of each pixel value is calculated by calculating the average of the absolute values of deviations from the average value (S203).

  Next, it is determined whether the average deviation is larger than the noise level threshold (S204). If the average deviation is larger than the noise level threshold (S204: Yes), the average deviation is output as a flat part noise evaluation value (S205). . On the other hand, when the average deviation is not larger than the noise level threshold (S204: No), 0 is output as the flat portion noise evaluation value (S206).

  Next, the processing procedure of the correction unit 130 will be described with reference to the flowchart of FIG. The correction unit 130 multiplies the edge evaluation value output from the edge evaluation unit 110 by a preset edge correction coefficient to calculate an edge correction value (S301), and the flat portion output from the flat portion noise evaluation unit 120. The flat part noise correction value is calculated by multiplying the part noise evaluation value by a preset flat part correction coefficient (S302).

  Then, the ε value is corrected by subtracting the edge correction value from the externally input standard ε value and adding the flat portion noise correction value (S303).

  If the corrected ε value is within the predetermined minimum ε value to maximum ε value (S304: No), the corrected ε value is output as it is (S306). On the other hand, when the corrected ε value is smaller than the minimum ε value or larger than the maximum ε value (S304: Yes), the ε value falls within the range from the minimum ε value to the maximum ε value by the limiter process. The ε value is output (S 306) by limiting to be within the range (S 305).

  Finally, specific effects of the present embodiment will be described with reference to FIGS. 9 to 12 and FIG. 9 to 12, for simplicity, the image block is one-dimensional, each pixel is associated with the horizontal axis, and the pixel value is illustrated on the vertical axis. Each pixel is given a pixel number.

  FIG. 9 shows the pixel value distribution of the original image. The broken line in the figure shows the true value for each pixel not including the noise component, that is, the luminance distribution of the subject, and the broken line indicates the pixel of the original image including the noise component. The value is shown. As indicated by the broken line, the object has pixel numbers 6 and 7 as edge portions, and pixel numbers 1 to 6 and pixel numbers 7 to 19 are flat portions. In contrast, in the original image, the noise component increases as the pixel number increases (as it goes to the right). The PSNR (Peak S / N Ratio) of the original picture is 34.8 dB as shown in FIG.

  FIG. 10 is a diagram illustrating a result of removing noise by applying a 5-pixel moving average filter to an original image. In this case, as shown in this figure and FIG. 13, it can be seen that the PSNR is improved to 48.0 dB and the noise removal effect is high. However, the edge portions of pixel number 6 and pixel number 7 are smoothly connected, and edge blurring occurs.

  FIG. 11 is a diagram illustrating a result of removing noise by applying an ε filter with a fixed ε value to an original image. The luminance difference at the edge portion of the original image is 5, but the amplitude of the noise in the flat portion is equal to or greater. Therefore, the ε value is set to 8 in order to hold the edge portion.

  The PSNR in this case was 37.7 dB. If the ε value is further increased, the S / N is improved, but it approaches the moving average filter and blur occurs at the edge portion. Further, it can be seen that with this ε value, noise cannot be removed in a region where the amount of noise after pixel number 14 is large.

  FIG. 12 is a diagram showing a result of removing noise by adaptively changing the ε value according to the present embodiment. In this example, the sub-area setting pattern 1 is adopted, the standard ε value is set to 10, the edge level threshold is set to 2, the edge portion correction coefficient is set to 0.8, the noise level threshold is set to 5, and the flat portion correction coefficient is set to 0.8. did.

  As a result, the PSNR was the highest at 48.3 dB, and an image in which the edge portions of the pixel numbers 6 and 7 were retained could be obtained. Thus, in the nonlinear filter device 10 of the present embodiment, the ε value of the ε filter can be adaptively changed, and the noise removal performance can be improved while holding the edge portion.

DESCRIPTION OF SYMBOLS 10 ... Nonlinear filter apparatus, 100 ... Epsilon value calculation part, 110 ... Edge evaluation part, 111 ... Subarea average calculation part, 112 ... Edge evaluation value calculation part, 120 ... Flat part noise evaluation part, 121 ... Average deviation calculation part, 122: flat part noise evaluation value calculation unit, 130 ... correction unit, 131 ... coefficient multiplication unit, 132 ... ε value correction unit, 132 ... correction unit, 133 ... ε limiter, 200 ... ε filter unit, 210 ... ε determination unit, 220 ... adder, 230 ... buffer, 240 ... counter, 250 ... divider

Claims (4)

For a pixel block composed of a pixel of interest and a plurality of peripheral pixels, the difference between the pixel value of the pixel of interest and the pixel value of each peripheral pixel is calculated, and the pixel value of the peripheral pixel whose absolute value is smaller than the determination value A value obtained on the basis of the pixel value of the target pixel,
A plurality of subareas in the pixel block, and a subarea average calculating unit that calculates an average value of pixel values of pixels included in each subarea;
An edge evaluation value calculation unit that outputs an edge evaluation value obtained based on the correlation between the calculated average values for each sub-area;
A flat part noise evaluation value calculating unit that outputs a flat part noise evaluation value obtained based on an average deviation of pixel values of each pixel in the pixel block;
A setting unit configured to set the determination value such that the determination value decreases according to the edge evaluation value and the determination value increases according to the flat part noise evaluation value. apparatus. The filter device according to claim 1,
The edge evaluation value calculation unit converts an interrelation between average values for each subarea into an index value based on a predetermined edge evaluation formula, and when the index value exceeds a predetermined threshold, the edge evaluation value is converted into the edge evaluation value. A filter device that outputs as a value and outputs 0 as the edge evaluation value when the predetermined threshold value is not exceeded. The filter device according to claim 2,
The sub-area average calculating unit includes, in the pixel block, a first sub-area that is a central region including the target pixel, and an entire region excluding the first sub-area from the pixel block or the entire pixel block. Set a second sub-area,
The edge evaluation value calculation unit uses the absolute value of the difference between the average values of both subareas as the index value.   For a plurality of pixels constituting a pixel block including the pixel of interest, a difference in pixel value from the pixel of interest is calculated, and a value obtained based on a pixel value of a pixel whose absolute value is smaller than the determination value is calculated. A filtering method in a filter device for setting a pixel value of the target pixel,
  A sub-area average calculating step of setting a plurality of sub-areas in the pixel block and calculating an average value of pixel values of pixels included in each sub-area;
  An edge evaluation value calculating step for calculating an edge evaluation value based on the correlation between the calculated average values for each sub-area;
  A flat part noise evaluation value calculating step for calculating a flat part noise evaluation value based on an average deviation of pixel values of each pixel in the pixel block;
  And a setting step for setting the determination value such that the determination value decreases according to the edge evaluation value and the determination value increases according to the flat part noise evaluation value. .