JP6008125B2 - Noise reduction device and noise reduction method - Google Patents

Description

  The present invention relates to a noise reduction device and a noise reduction method for reducing image noise.

  The simplest method for reducing noise such as white Gaussian noise in which the noise has equal components at all frequencies is a method using a low-pass filter. However, in the noise reduction method using the low-pass filter, high-frequency component noise is cut, while medium- and low-frequency noise remains. Even when an epsilon filter or a bilateral filter known as an edge-preserving smoothing filter other than the low-pass filter is used, noise reduction with respect to a flat area of an image results in noise reduction biased to high-frequency noise.

  The medium and low frequency noises that are difficult to reduce appear as noises with large graininess or uneven noises, and are therefore more annoying than high frequency noises in terms of human visual characteristics. Therefore, it is necessary to effectively reduce medium and low frequency noise.

  In the conventional medium / low frequency noise reduction method, the target image is band-divided into a low-frequency component and a high-frequency component, each low-frequency component and high-frequency component image is filtered, and the low-frequency component after filtering is processed. And the high-frequency component image after the filter processing are synthesized to obtain a target image after noise reduction (image after noise reduction). Thereby, it becomes possible to reduce medium and low frequency noise (for example, refer patent document 1).

JP 2008-153917 A (paragraphs 0020 to 0032, FIG. 1)

  However, in the conventional general noise reduction method as exemplified in Patent Document 1, noise removal is performed by referring to the pixels around the pixel of interest with a pattern based on the pixel arrangement, and thus noise reduction. There is a problem that an artificial pattern depending on the reference pattern is generated in a later image. Examples of the artificial pattern include uneven noise and a mottled pattern larger than the reference range of surrounding pixels.

  The present invention has been made to solve the above-described problems, and suppresses an artificial pattern depending on a reference pattern of peripheral pixels including a target pixel, which is generated when medium / low frequency noise is reduced. It is an object of the present invention to provide a noise reduction apparatus and a noise reduction method that can perform the above-described process.

  The noise reduction device according to the present invention sorts the pixel values of each pixel in the first size area including the target pixel, sets the sorting order to each pixel in the first size area, and sorts the target pixel. A rank labeling unit for labeling the rank of the pixel of interest, a weighted addition target rank is determined according to the sort rank of the pixel of interest, and the pixel value of the pixel labeled with the weighted addition rank is set to a first value including the pixel of interest. A pixel selection weighted addition unit that selects and weights and adds from a region having a second size that is larger than the size, and determines a pixel value of the pixel of interest based on the weighted addition.

  Further, the noise reduction method according to the present invention sorts the pixel values of the pixels in the first size area including the target pixel, sets the sorting order to the pixels in the first size area, and sets the target pixel. Labeling step of labeling the pixel of interest with the sort order of the pixel, the weighted addition target rank is determined according to the sort order of the pixel of interest, and the pixel value of the pixel labeled with the weighted addition target rank includes the pixel of interest A pixel selection weighted addition step of selecting from a region of a second size larger than the first size, performing weighted addition, and determining a pixel value of the pixel of interest based on the weighted addition.

  According to the present invention, the sorting order based on the pixel value of the pixel in the first size area including the target pixel is labeled to the target pixel, the weighted addition target order is determined according to the sort order of the target pixel, Since the pixel of the sort order that is the weighted addition target rank is selected from within the area of the second size larger than the size of 1 and weighted addition is performed, and the pixel value of the target pixel is determined based on the weighted addition result. The pattern can be suppressed.

It is a block diagram which shows the block diagram of the noise reduction apparatus which concerns on Embodiment 1 of this invention. It is a figure which shows the flowchart of the noise reduction method which concerns on Embodiment 1 of this invention. It is a figure which shows an example with a label of the rank labeling part which concerns on Embodiment 1 of this invention. It is a figure which shows the weighting addition object order | rank of the pixel selection weighting addition part which concerns on Embodiment 1 of this invention. It is a figure which shows the selection method of the weighting addition object order | rank of the pixel selection weighting addition part which concerns on Embodiment 1 of this invention. It is the figure which showed the principle in which the high frequency noise which concerns on Embodiment 1 of this invention remains. It is a figure which shows an example of the weighting with respect to each pixel which concerns on Embodiment 1 of this invention. It is a figure which shows the frequency characteristic of the noise reduction by the noise reduction apparatus which concerns on Embodiment 1 of this invention. It is a block diagram which shows the imaging device to which the noise reduction apparatus which concerns on Embodiment 1 of this invention is applied.

Embodiment 1 FIG.
FIG. 1 shows a configuration of a noise reduction apparatus according to Embodiment 1 of the present invention. The noise reduction device 10 according to the first embodiment of the present invention includes a rank labeling unit 1, a pixel selection weight addition unit 2, and a delay unit 3.

  The order labeling unit 1 shown in FIG. 1 sorts the pixel values in the first size area including the peripheral pixels of the target pixel of the input image In, and the pixel values of the target pixel in the first size area. The sorting order is labeled on the target pixel. The pixel selection weighted addition unit 2 is a first pixel including a target pixel that is labeled with a rank (hereinafter referred to as a weighted addition target rank) that is a target of weighted addition determined according to the rank of the target pixel. Is selected and weighted and added from within a second size region larger than the size of the pixel, and a pixel value is given to the pixel of interest based on the weighted addition result. The delay unit 3 delays the input image In until the labeling by the rank labeling unit 1 is completed, and the input timing of the input image In and the labeling result of the rank labeling unit 1 to the pixel selection weight addition unit 2 is delayed. Match.

  When processing of the entire image is considered, the input image In is input to the rank labeling unit 1 and the delay unit 3 for each pixel. The rank labeling unit 1 sequentially sets all the pixels constituting the input image In as the target pixel, sorts each pixel in the first size area including the target pixel by the pixel value, and sets the set target pixel The sorting order is labeled every time. Next, at the stage where labeling is completed for all the pixels in the second size area including the peripheral pixels of the target pixel, the pixel selection weighted addition unit 2 sets the weighted addition target rank according to the rank of the target pixel. The pixel belonging to the weighted addition target rank is selected from the second size area including the target pixel, weighted addition is performed, a pixel value is given to the target pixel based on the weighted addition result, and the output image Out Output as the value of the pixel of interest. When this is completed for all the pixels of the output image Out, the noise reduction processing for one image is completed.

  FIG. 2 shows a flowchart of the noise reduction method according to Embodiment 1 of the present invention. The rank labeling step S1 labels the rank in the first size area including the peripheral pixels of the target pixel of the input image In. The pixel selection weighted addition step S2 determines the weighted addition target rank according to the rank of the target pixel, selects a pixel belonging to the weighted addition target rank from the second size area including the peripheral pixels of the target pixel, and performs weighting. The pixel value of the target pixel is determined based on the weighted addition result, and is output as the value of the target pixel of the output image Out.

  The rank labeling unit 1 will be specifically described. The order labeling unit 1 shown in FIG. 1 sorts pixel values in a first size area including peripheral pixels of the target pixel of the input image In, for example, a 3 × 3 pixel area, and sets the pixel value of the target pixel. The sorting order in the first size area is labeled for the pixel of interest.

  FIG. 3 shows an outline of labeling in the rank labeling unit 1 shown in FIG. FIG. 3A shows a pixel value array of 3 × 3 pixels around the pixel of interest as an example of the first size region (hereinafter, referred to as a first size region 6). . FIG. 3B shows a sorting order in which the pixels in the first region 6 shown in FIG. 3A are sorted in ascending order according to the pixel values.

  As shown in FIG. 3A, the pixel value of the target pixel 5 surrounded by a thick line in the first size region 6 is 130. The sorting order of the target pixel 5 is the fifth order when it is sorted in the ascending order of the pixel values in the first size area 6, that is, in the area of 3 × 3 pixels around the target pixel 5. In this case, the value of the target pixel is labeled “5” as shown in FIG. At this time, eight pixels other than the target pixel are also related to the ranking, but only the sorting order of the target pixel is labeled. In this way, all pixels constituting the image are labeled.

  In addition, when the pixel of the same pixel value (same order) exists in the area | region 6 of 1st size, the average order shall be allocated to the said pixel. When the average rank includes a value after the decimal point, for example, rounding may be performed and the fraction may be processed to make an integer.

  Next, the pixel selection weighted addition unit 2 will be specifically described. The pixel selection weighted addition unit 2 sets the pixels labeled in the weighted addition target order determined according to the sorting order of the target pixel 5 to be larger than the first size region 6 including the peripheral pixels of the target pixel 5. A pixel is selected from a second size, for example, an area of 9 × 9 pixels (hereinafter referred to as a second size region 7), and weighted addition is performed. The average of the pixel values is calculated, and the average of the pixel values is set as the pixel value of the target pixel 5. FIG. 4 shows a method of selecting a pixel to be subjected to weighted addition in the pixel selection weighted addition unit 2. The pixel is selected based on the sort order labeled on the pixel. For example, as shown in FIG. 4, when the sorting order of the pixel of interest 5 is “5”, there is a method in which a pixel labeled “5” is subjected to addition weighting. In this case, as shown in FIG. 4, all the pixels labeled “5” in the area 7 of the second size are selected and subjected to weighted addition.

  FIG. 5 shows a method of selecting a sort order to be subjected to weighted addition, that is, a weighted addition target order when the possible values of the sort order are “1” to “9”. There are a pattern A, a pattern B, a pattern C, and a pattern D as the selection method of the weighted addition target order in FIG.

  First, a case where the weighting addition target order selection method is pattern A will be described. In order to make the explanation easy to understand, the principle for a flat image on which noise is superimposed will be described. However, it is assumed that a flat image is an image in which a change in gradation between pixels constituting an image in a local region is very small. The same applies to the patterns B to D described below. In the pattern A, as shown in FIG. 4, when the sorting order of the target pixel 5 is “5” in the second size area 7, the weight addition target rank is the second size area 7. Among the pixels, “5” is ranked. That is, the pattern A is a pattern in which pixels with the same sort order as the sort order of the pixel of interest 5 are the order of weighted addition.

  When the change of the pixel value occurs only by a random noise component, the result of labeling by the rank labeling unit 1 preserves the magnitude relationship of the change of the pixel value due to noise, and makes the rank labels of adjacent pixels different. Therefore, the result of labeling by the rank labeling unit 1 preserves the magnitude relationship of the high-frequency noise pattern.

  Subsequent pixel selection weighted addition unit 2 sets the same sorting order as the sorting order of the target pixel as the weighting addition target order, thereby subjecting the pixel group having substantially the same deviation from the distribution center of pixel values due to noise to the target of weighted addition Therefore, the weighted addition result varies depending on the rank label of the target pixel. Since the rank label stores random components of adjacent pixels, the pixel values of adjacent pixels are made different, and high-frequency noise remains.

  FIG. 6 shows the principle that high-frequency noise remains. Here, a case where the first size area 6 is 3 × 3 pixels and the possible values of the sort order are “1” to “9” will be described. FIGS. 6A, 6 </ b> B, and 6 </ b> C schematically represent noise distributions extracted from regions of a first size at different positions in the flat image on which noise is superimposed.

  First, by substituting the pixel value with the sort order in the rank labeling unit 1, it is possible to extract a normalized noise distribution that does not depend on the pixel value level or the noise amplitude. Noise is superimposed over the entire image according to the same distribution, for example, a Gaussian distribution. However, since the position and width of the distribution change depending on the pixel value level and noise amplitude, the position in the noise distribution cannot be distinguished by the pixel value itself. However, by replacing with the sort order, as shown in FIGS. 6A, 6B, and 6C, no assumptions regarding the pixel value level and the noise amplitude are required, and the center of the noise distribution is approximately the sort order “5”. ”, The lower limit of the noise distribution corresponds to the sort order“ 1 ”, and the upper limit of the noise distribution corresponds to the sort order“ 9 ”.

  Here, when the pixel selection weighted addition unit 2 performs weighted addition using the pattern A, only the pixels at the same position in the noise distribution normalized by the sort order are weighted and added. That is, when the order of the target pixel is “3”, the pixel group in which the deviation from the distribution center of the pixel value due to the noise is almost the same in the noise distribution shown in FIGS. 6A, 6B, and 6C. As described above, the randomness of the rank labeling result is preserved to make the pixel values of adjacent pixels different, and high-frequency noise remains without being reduced.

  On the other hand, for medium and low frequency noise, a pixel labeled in a weighted addition target order is selected from a region having a second size larger than the first size and subjected to weighted addition. Is significantly reduced.

  Based on the above principle, according to the pattern A in which the same rank as the sorting order of the target pixel is set as the weighted addition target order, the reference of the peripheral pixels of the target pixel 5 is determined by the pixel value, so the reference based on the pixel arrangement Artificial patterns generated by the method can be suppressed.

  In addition, it is possible to leave high frequency noise while reducing medium to low frequency noise.

  Next, the case where the weighted addition target order selection method is pattern B will be described. As shown in FIG. 5, in the case of pattern B, when the sorting order of the target pixel is “R”, pixels whose sorting orders are “R−1”, “R”, and “R + 1” are weighted addition target rankings. To do. However, when the sorting order of the pixel of interest 5 is “1” and “9”, the sorting orders “1-1” and “9 + 1” do not exist, respectively, and are therefore omitted. That is, the pattern B is a pattern in which the pixels of the sorting order adjacent to the sorting order of the target pixel 5 are set as the weighted addition target order.

  In the pattern B, a signal at a distribution position having a width is allowed and added to the deviation of the pixel value from the distribution center due to noise. The difference in pixel value for each label is reduced, and the effect of reducing high-frequency noise appears.

  In addition to the method in which the non-existing sort order is a missing number, when the target pixel sort order is “1”, “1”, “2”, “3”, and the target pixel 5 sort order is “9”. In some cases, “7”, “8”, and “9” may be set as a weighted addition target rank.

  In the pattern B, the sorting order of the pixel of interest 5 and the adjacent sorting order are weighted addition targets. However, the present invention is not limited to this. For example, when the sorting order of the pixel of interest 5 is “R”, the sorting order is “ Pixels from “Rk” to “R + k” may be set as the weighted addition target pixels. However, when there is no sort order to be referred to, a method of securing the same number of sort orders by bringing the reference rank to the center of the sort order as described above may be used.

  Based on the above principle, according to the pattern B in which the pixels in the order adjacent to the sorting order of the target pixel 5 are set as the weighted addition target order, the reference of the peripheral pixels of the target pixel 5 is determined by the pixel value. An artificial pattern generated by the reference method described above can be suppressed.

  In addition, the reduction degree of the high frequency noise can be adjusted by referring to the sorting order adjacent to the order of the pixel of interest 5 while reducing the medium to low frequency noise.

  Next, the case where the selection method of the weighted addition target rank is the pattern C will be described. In the case of the pattern C, when the sorting order of the pixel of interest 5 is smaller or larger than “5”, which is the central sorting order of all the sorting orders, the sorting from the sorting order of the pixel of interest to “5” is performed. When the rank is the target of weighted addition and the sort order of the pixel of interest 5 is “5”, the central sort order of all the sorted ranks is “5” as the target of weighted addition. That is, a pattern in which the center sorting order of all sorted sorting orders performed in the first size area 6, the sorting order of the target pixel 5, and the pixels of the sorting order between them are subjected to weighted addition. It is.

  In the pattern C, in the result of labeling by the rank labeling unit 1, the sort order close to “1”, which is the highest sort order, and “9”, which is the lowest sort order, is sorted in the region 6 of the first size. With respect to the sort order “5” at the center of all the sorted orders, pixels belonging to other than the sort order may be regarded as pixels having a larger noise amplitude as they become closer to the top or bottom of the sort order. it can. Therefore, by increasing the number of pixels to be subjected to weighted addition and selecting to include pixels with a large noise amplitude, the degree of noise smoothing is increased. As shown in FIG. 6, the sort order “5” at the center of all sort orders sorted in the area 6 of the first size can be regarded as the center of the noise distribution, that is, the predicted value when there is no noise. For this reason, the higher the sort order is, the closer the sort order is to the lowest order, and it is more reasonable to assign the sort order closer to the central sort order to the weighted addition target order.

  According to the above principle, according to the pattern C in which the center sort order of the sort orders performed in the area 6 of the first size, the rank of the target pixel 5, and the rank therebetween are weighted addition target ranks. Since the reference of the peripheral pixels of the target pixel 5 is determined by the pixel value, an artificial pattern generated by the reference method based on the pixel arrangement can be suppressed.

  In addition, it is possible to reduce noise in a well-balanced manner from high frequency noise to medium to low frequency noise while effectively suppressing noise of pixels having a large noise amplitude.

  Next, in the case of the pattern D, which describes the case where the selection method of the weighted addition target rank is the pattern D, the sort rank is almost the same as in the case of the pattern C. However, when the sort rank of the target pixel 5 is “1”, the sort rank “1” is excluded from the weighted addition target rank, and when the target pixel sort rank is “9”, the sort rank “9” is excluded from the weighted addition target rank. In other words, the middle rank of the sort order performed in the area 6 of the first size, the sort rank of the target pixel 5, and the pixel of the sort rank between them are selected, and the area 6 of the first size is selected. This is a pattern in which the sorting ranks corresponding to the top and bottom of the sorting ranks of the sorting performed in the above are excluded from the weighted addition target ranks.

  When the target image includes impulsive noise due to a pixel defect or the like, in a general noise reduction method based on a low-pass filter, the noise is often not reduced simply by spreading a point due to the impulsive noise. For such a situation, according to the pattern D, such protruding pixels with noise are labeled as “1” or “9” in the sort order, and thus the pixels with these sort orders are excluded from the weighted addition target. By doing so, the influence can be effectively removed.

  Based on the above principle, the center rank of the sort order performed in the area 6 of the first size, the sort rank of the target pixel 5, and the pixel having the sort rank in between are selected, and the first size is selected. According to the pattern D in which the sorting order corresponding to the top and bottom of the sorting order of the sorting performed in the area 6 is excluded from the weighted addition target order, the reference of the peripheral pixels of the target pixel 5 is determined by the pixel value. The artificial pattern generated by the reference method based on the pixel arrangement can be suppressed.

  In addition, while effectively reducing impulsive noise, it is possible to reduce from high frequency noise to medium to low frequency noise in a well-balanced manner.

  In this way, the pixel selection weight addition unit 2 selects and weights and adds the pixels to be subjected to the weight addition according to the sorting order of the target pixel 5 from the patterns A to D described above from the second size area 7. The pixel value of the target pixel 5 is determined based on the weighted addition result.

In the first embodiment, the first size region 6 is 3 × 3 pixels and the possible values of the sort order are “1” to “9”. Similarly, in the case where the size area 6 is N × N pixels (N is an integer) and the possible values of the sort order are “1” to “N ² ”, a method for determining the pixel to be subjected to weighted addition is similarly used. It is possible to determine.

The second size region 7 may be any size larger than the first size region 6, but when the first size is N × N pixels, the second size is N ² × N. ^It is desirable to have ^two or more pixels. By setting the second size to N ² × N ² pixels or more, pixels to be subjected to weighted addition having the same sort order as that of the target pixel 5 are found to some extent, and there is no bias in the number of weighted added pixels due to the rank label. Thereby, on average, the possibility that at least N × N pixels of the same order exist in the second size region 7 is improved.

  Next, a method for determining the pixel value of the target pixel after determining the method for selecting the weighted addition target order will be described. In the simplest configuration, after adding the pixel values of the pixels selected by the weighted addition target rank and dividing the addition result by the number of pixels of the pixels selected by the weighted addition target rank, the average value of the pixels is obtained. A method of using the average value as the pixel value of the target pixel 5 can be used. In this case, the pixels selected by the weighted addition target rank are simply added, but weighting, that is, weighting can be performed on the pixels selected by the weighted addition target rank immediately before the addition. For example, a weighting coefficient that is changed according to a spatial distance from the target pixel 5 can be used for a pixel to be subjected to weighted addition by a Gaussian filter or the like.

  FIG. 7 shows an example of weighting for each pixel in the region 7 of the second size. Numerical values in the figure indicate weighting coefficients. For example, the weighting coefficient of the target pixel 5 is “19”. In FIG. 7, the weighting is reduced as the distance from the target pixel 5 increases with the target pixel 5 as the center. That is, the weighting is changed according to the spatial distance from the target pixel 5. In this case, the pixel value of the pixel corresponding to the weighted addition target rank is multiplied by the weight coefficient shown in FIG. 7 and added, and the addition result is divided by the sum of the weight coefficients of the pixels corresponding to the weighted addition target rank. It is possible to obtain a weighted addition result of the pixel 5.

  Furthermore, such weighting may be changed in accordance with the difference between the pixel corresponding to the weighted addition target order and the pixel value of the target pixel 5 based on an idea such as an epsilon filter or a bilateral filter. As a result, it is possible to prevent edge disturbance due to irregular pixel selection across the edges.

  Next, the noise reduction effect according to the first embodiment of the present invention will be described in more detail. FIG. 8 schematically shows the frequency characteristics of noise reduction by the noise reduction apparatus according to Embodiment 1 of the present invention. WN shown in FIG. 8 represents the frequency characteristic of noise, and LN represents the frequency characteristic when the noise shown in WN is reduced by a low-pass filter.

  As shown in FIG. 8WN, when the image has noise such as white Gaussian noise, it has noise components evenly at all frequencies. As shown in FIG. 8LN, noise reduction by a general low-pass filter cannot sufficiently reduce the noise in the middle / low frequency region indicated by LM in FIG.

  According to the noise reduction apparatus according to Embodiment 1 of the present invention, when the weighted addition target order selection method is pattern A, it can be seen that high-frequency noise remains while reducing medium-low frequency noise. Further, it can be seen that the effect of reducing high-frequency noise is obtained by selecting a larger number of sorting ranks as weighted addition targets by the selection method of the weighted addition target ranks shown in the patterns B to D.

  From the above, according to the first embodiment of the present invention, the reference pattern of the peripheral pixels of the target pixel 5 is determined by the pixel value, so that an artificial pattern generated by the reference pattern based on the pixel arrangement is suppressed. In addition, by appropriately selecting the weighted addition target rank, not only the medium and low frequency noise but also the high frequency noise can be effectively reduced.

  FIG. 9 shows a configuration of an imaging apparatus 100 as an application example of the noise reduction apparatus 10 according to Embodiment 1 of the present invention. The imaging apparatus 100 includes an imaging element unit 101, an analog signal processing unit 102, an A / D conversion unit 103, and a digital signal processing unit 104.

  The imaging element unit 101 photoelectrically converts light from a subject and outputs an electric signal obtained as a result as an imaging signal S101. The analog signal processing unit 102 includes an analog noise removal circuit and a gain amplifier. The analog signal processing unit 102 performs noise removal and analog signal amplification processing on the imaging signal S101 and outputs an analog processing signal S102. The A / D conversion unit 103 converts the analog processing signal S102 into a digital signal and outputs it as a digital signal S103. The digital signal processing unit 104 performs colorization processing, gradation conversion processing, interpolation processing, filter processing, and the like according to the signal format of the digital signal S103, and processes the signal so that it can be easily used in subsequent signal processing and display devices. And output as a digital video signal S104.

  The digital signal processing unit 104 includes an area determination unit 104a and a noise reduction unit 104b. The area determination unit 104a determines a flat part of an image suitable for applying the noise reduction apparatus 10 according to the first embodiment of the present invention. The noise reduction part 104b has the noise reduction apparatus 10 of Embodiment 1 of this invention inside, and applies noise reduction to the area | region determined as the flat part by the area | region determination part 104a.

  The imaging apparatus 100 to which the noise reduction apparatus 10 according to the first embodiment of the present invention is applied reduces noise in a balanced manner over all frequencies while suppressing an artificial pattern generated by a reference pattern based on a pixel arrangement. Output high-quality images.

  As described above, according to the first embodiment of the present invention, the periphery from the region of the second size larger than the first size is determined according to the sorting order determined according to the sorting order of the target pixel. Since the pixels are weighted and added, an artificial pattern can be suppressed, and not only low and medium frequency noise can be reduced, but the degree of reduction of high frequency noise can be reduced by appropriately selecting the weighted addition target order. Can be adjusted.

  That is, it is possible to obtain a noise reduction device that can control noise reduction from a medium low frequency to a high frequency with a relatively simple processing configuration.

  Within the scope of the present invention, the present invention can be freely combined with each other, or can be appropriately changed or omitted.

DESCRIPTION OF SYMBOLS 1 Rank labeling part 2 Pixel selection weight addition part 3 Delay part 5 Target pixel 6 1st size area | region 7 2nd size area | region 10 Noise reduction apparatus In input image Out Output image

Claims (16)

The pixel values of the pixels in the first size area including the target pixel of the input image are sorted, the sorting order is assigned to each pixel in the first size area, and the sorting order of the target pixel is set as the attention order. A rank labeling section for labeling pixels;
A pixel labeled with a weighted addition target order determined according to the sort order of the target pixel is selected from a region having a second size larger than the first size and including the target pixel. A pixel selection weighted addition unit that performs weighted addition of the pixel values of the selected pixels and determines and outputs a pixel value of the target pixel based on the weighted addition;
Noise reduction device with The noise reduction apparatus according to claim 1, wherein the weighted addition target order is the same as the sorting order of the target pixel. The noise reduction apparatus according to claim 1, wherein the weighted addition target rank is a sort rank adjacent to the sort rank of the target pixel and the sort rank of the target pixel. 2. The noise reduction apparatus according to claim 1, wherein the weighted addition target rank is from the sort rank of the target pixel to the central sort rank of all sorted sort ranks. 5. The noise reduction according to claim 4, wherein the sorting order of the pixel of interest is excluded from the weighted addition target ranking when the sorting order of the pixel of interest is the lowest or highest of all sorted sorting orders. apparatus. The noise reduction apparatus according to claim 1, wherein the pixel selection weighted addition unit calculates an average value of the pixels labeled in the weighted addition target order. 6. The pixel selection weighted addition unit changes weighting for a pixel labeled in a weighted addition target rank according to a spatial distance from a target pixel. The noise reduction device described in 1. The pixel selection weighted addition unit changes weighting for a pixel labeled in a weighted addition target order according to a difference between a pixel value of the pixel and a pixel value of a target pixel. The noise reduction device according to any one of 5. The pixel values of the pixels in the first size area including the target pixel of the input image are sorted, the sorting order is assigned to each pixel in the first size area, and the sorting order of the target pixel is set as the attention order. A rank labeling step for labeling the pixels;
A pixel labeled with a weighted addition target order determined according to the sort order of the target pixel is selected from a region having a second size larger than the first size and including the target pixel. A pixel selection weighted addition step of weighting and adding pixel values of the selected pixels and determining and outputting a pixel value of the pixel of interest based on the weighted addition;
Noise reduction method with The noise reduction method according to claim 9, wherein the weighted addition target order is the same as the sorting order of the target pixel. The noise reduction method according to claim 9, wherein the weighted addition target rank is a sort rank adjacent to the sort rank of the target pixel and the sort rank of the target pixel. The noise reduction method according to claim 9, wherein the weighted addition target rank is from the sort rank of the pixel of interest to the sort rank at the center of all sorted sort ranks. The noise reduction according to claim 12, wherein when the sorting order of the pixel of interest is the lowest or highest of all the sorted sorting orders, the sorting order of the pixel of interest is excluded from the weighted addition target ranking. Method.   14. The noise reduction method according to claim 9, wherein the pixel selection weighted addition step calculates an average value of pixels labeled in the weighted addition target order. The pixel selection weighted addition step changes the weighting for the pixels labeled in the weighted addition target order according to a spatial distance from the target pixel. The noise reduction method described in 1. The pixel selection weighted addition step changes the weighting for the pixels labeled in the weighted addition target order according to a difference between a pixel value of the pixel and a pixel value of the target pixel. 14. The noise reduction method according to any one of items 13.

Images