JP2015076676A - Image processing apparatus, image processing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent a noise reduction method from varying in noise reduction effect in each area of an image.SOLUTION: An image processing apparatus comprises: estimation means which generates the histogram of similarity between a target pixel subjected to noise reduction processing and plural reference pixels corresponding to the target pixel and estimates the characteristics of the neighboring area of the target pixel on the basis of the generated similarity histogram; and noise reduction processing means which determines a weight relating to the reference pixels on the basis of the characteristics of the neighboring area of the target pixel estimated by the estimation means and performs the noise reduction processing on the target pixel.

Description

  The present invention relates to an image processing technique for reducing noise included in image data.

  Digital imaging devices such as digital still cameras and digital video cameras are widely spread and are generally used. These digital imaging devices generate digital image data by converting light received by a photoelectric charge conversion device (imaging device) such as a CCD or CMOS sensor into a digital signal.

  In the process of generating digital image data, dark current noise, thermal noise, shot noise, and the like are generated due to the characteristics of the imaging device and circuit, and the noise is mixed into the digital image data. With recent miniaturization of image sensors and higher pixel counts, the pixel pitch has been minimized, making noise more conspicuous, especially when shooting sensitivity is increased, resulting in noticeable noise and significant deterioration in image quality. It is a factor. Therefore, in order to obtain a high-quality image, it is necessary to reduce mixed noise.

  Conventionally, a technique for reducing noise by applying a low-pass filter that passes a signal component below a noise frequency is known. However, in the technique of reducing noise by applying this low-pass filter, it is difficult to obtain a high-quality image because the edge portion is blurred. Therefore, many methods have been proposed for determining noise and edge information by some method and adaptively reducing noise.

  In general, in adaptive noise reduction, in order to reduce noise of a target pixel, a plurality of pixels near the target pixel are selected as reference pixels, and the target pixel is replaced with an appropriate weighted average value of the selected reference pixels. Yes. As one of adaptive noise reduction methods, a region including a target pixel (target region) is determined, the similarity between the target pixel and the reference pixel is obtained for each region, and a weighted average value corresponding to the similarity is obtained. A method for realizing noise reduction by using these has been proposed (Patent Documents 1 and 2).

Special table 2007-536662 gazette JP 2011-39675 A

  However, the methods disclosed in Patent Document 1 and Patent Document 2 have a problem that the noise reduction effect varies for each image area. For example, if there are many similar pixels in the vicinity of the pixel of interest, the noise reduction effect will be high, but it will be low if there are few. In addition, it is effective for a region having a repetitive pattern such as a lattice, but there is a problem that a region having a random texture such as a tree is easily blurred. This is because, in an area having a random texture, reference pixels having a high similarity between the target pixel and the reference pixel are decreased, while reference pixels having a medium similarity are increased, and a large weight is obtained with a small weight in the weighted average process. This is because the pixels are added.

  The image processing apparatus according to the present invention generates a histogram of similarity between a target pixel to be subjected to noise reduction processing and a plurality of reference pixels corresponding to the target pixel, and based on the generated similarity histogram Estimating means for estimating the characteristics of the neighboring area of the target pixel; and determining weights for the plurality of reference pixels based on the characteristics of the neighboring area of the target pixel estimated by the estimating means; Noise reduction processing means for performing noise reduction processing.

  According to the present invention, it is possible to obtain a suitable noise-reduced image by estimating the characteristics of the neighborhood region of the target pixel to which the noise reduction process is applied and determining the content of the noise reduction process based on the estimation result. Become.

1 is a diagram illustrating an example of a hardware configuration of an image processing apparatus according to Embodiment 1. FIG. It is a figure explaining the calculation method of the weight of a reference pixel. It is a figure explaining the calculation method of the weight of a reference pixel. 1 is a schematic diagram illustrating a logical configuration of an image processing apparatus according to Embodiment 1. FIG. 6 is a flowchart illustrating a flow of processing for generating a noise reduced image in the first embodiment. It is a figure which shows an example of distribution of a reference histogram. 6 is a flowchart illustrating details of processing for estimating characteristics of a pixel-of-interest neighborhood area according to the first embodiment. 3 is a flowchart illustrating details of noise reduction processing according to the first embodiment. FIG. 10 is a schematic diagram illustrating a logical configuration of an image processing apparatus according to a second embodiment. It is a figure which shows how the characteristic of a similarity histogram is determined. 12 is a flowchart illustrating a flow of main processing for generating a noise-reduced image according to the second embodiment. 10 is a flowchart illustrating details of processing for estimating characteristics of a pixel-of-interest neighborhood according to a second embodiment. 10 is a flowchart illustrating details of noise reduction processing according to the second embodiment.

(Example 1)
FIG. 1 is a diagram illustrating an example of a hardware configuration of the image processing apparatus according to the present embodiment.

  In FIG. 1, the image processing apparatus 100 includes a CPU 101, a RAM 102, an HDD 103, a general-purpose interface (I / F) 104, a monitor 105, and a main bus 106. The general-purpose I / F 104 connects an imaging device 111 such as a camera, an input device 112 such as a mouse and a keyboard, and an external memory 113 such as a memory card to the main bus 106.

  First, the CPU 101 activates an image processing application stored in the HDD 103, expands it in the RAM 102, and displays a user interface screen (hereinafter referred to as UI screen) on the monitor 105. Subsequently, various data stored in the HDD 103 and the external memory 113, image data captured by the imaging device 111, instructions from the input device 112, and the like are transferred to the RAM 102. Furthermore, according to the processing in the image processing application, the data stored in the RAM 102 performs various calculations based on commands from the CPU 101. The calculation result is displayed on the monitor 105 or stored in the HDD 103 or the external memory 113.

  In the above configuration, based on a command from the CPU 101, image data is input to the image processing application to determine a region where the effect of the noise reduction processing is high and a region where the effect is low, and the noise reduction processing is used based on the determination result. A method for changing the weight will be described.

First, the Non-Local Means method (hereinafter, NLM method), which is a noise reduction processing method in the present embodiment, will be described. In this method, weighting is performed based on the similarity between a block centered on a pixel of interest targeted for noise reduction and a block centered on a reference pixel around it, and the value of the pixel of interest is calculated by a weighted average of the reference pixels. It is a technology that reduces noise by replacing it. Now, assuming that the number of reference pixels used for the weighted average is N _S , the pixel value of the reference pixel is I _j (j = 1 to N _S ), and the weight of the reference pixel is w _j (j = 1 to N _S ). The pixel value I _new of the pixel of interest after the noise reduction process is obtained by the following equation (1).

  Next, a method for calculating the weight w of the reference pixel will be described with reference to FIGS.

In FIG. 2A, an image 200 represents the pixel value of each pixel as I (x, y) with the upper left pixel as the origin. Here, the black rectangle indicates the target pixel 201, and the pixel value thereof is I (4, 4). A rectangular area of 3 × 3 pixels (N _b = 9) centering on the pixel of interest 201 indicates the region of interest 202. Each gray rectangle in a 5 × 5 pixel (N _b = 25) rectangular area including the pixel of interest 201 indicates a reference pixel 203. A rectangular area indicated by a broken line is a reference area 204 of the reference pixel 203, and is a 3 × 3 pixel area centered on the reference pixel having the pixel value I (2, 2). The reference area exists for all reference pixels (25 reference pixels in the example of FIG. 2A), but here, only the reference area of the reference pixel 203 having a pixel value of I (2, 2). Is shown.

Now, in order to obtain the weight w of the reference pixel 203 with the pixel value I (2, 2), first, the attention area 202 and the reference area 204 are compared to calculate the similarity. The similarity may be obtained by any method. For example, as shown in FIG. 2B, the pixels in the region of interest 202 are b _s (p, q), and the pixels in the reference region 204 are b _j (p, q) (j = 1 to N _b ). And Then, assuming that the difference between the spatially corresponding pixels of the attention area 202 and the reference area 204 is the similarity, the similarity C _j is obtained by the following equation (2).

The smaller the value of the similarity C _{j, the} higher the similarity between the attention area and the reference area. Therefore, the weight w is determined according to the similarity. Weight w has a large weight as the value of C _j is small, such as the function shown in FIG. 3, may be determined as the weight as the value of C _j is greater is reduced, for example, determined by the following equation (3) .

  Here, h is a parameter for controlling the magnitude of the weight w. When h is increased, the noise reduction effect is enhanced, but the edge is blurred.

  Then, the weight w for each reference pixel is obtained by sequentially comparing the attention area 202 and the reference area 204 of each reference pixel.

  The noise reduction processing in the present embodiment may be processing in which the weight of the reference pixel is determined based on the similarity between the focus area and the reference area, and the similarity and weight calculation methods are limited to the methods described here. It is not something that can be done.

(Logical configuration of image processing device)
FIG. 4 is a schematic diagram illustrating a logical configuration of the image processing apparatus 100 according to the present embodiment, and includes a reference histogram generation unit 401, an estimation unit 402, and a noise reduction processing unit 405.

  The reference histogram generation unit 401 acquires a block matching parameter, and generates a reference histogram according to a statistical distribution corresponding to the block matching parameter.

  The estimation unit 402 includes a similarity histogram generation unit 403 and an inter-histogram approximation derivation unit 404. Based on an instruction from the CPU 101, the estimation unit 402 estimates the characteristics of the neighborhood region of the target pixel using a block matching parameter and a similarity histogram.

  The similarity histogram generation unit 403 derives the similarity between the target pixel and a plurality of reference pixels corresponding to the target pixel using the input image data and block matching parameters, and generates a similarity histogram. Here, the image data is input from the imaging device 111, the HDD 103, or the external memory 113 based on an instruction from the CPU 101. Of course, the image data captured by the imaging device 111 may be input after being temporarily stored in a storage device such as the HDD 103. The block matching parameter is input from the HDD 103 or the external memory 113 based on an instruction from the CPU 101. Further, it may be specified and input directly by the input device 112 such as a keyboard or a mouse via the UI screen on the monitor 105.

  An inter-histogram approximation derivation unit 404 acquires the similarity histogram generated by the similarity histogram generation unit 403 and the reference histogram generated by the reference histogram generation unit 401, and derives the approximation between the two histograms.

  The noise reduction processing unit 405 includes a weight control parameter setting unit 406 and an NLM application unit 407.

  The weight control parameter setting unit 406 sets the value of the parameter h that controls the weight w used in the noise reduction process applied to the pixel of interest, based on the estimation result of the characteristics of the region near the pixel of interest estimated by the estimation unit 402.

  The NLM application unit 407 acquires the weight control parameter h set / updated by the weight control parameter setting unit 407, applies noise reduction processing by the NLM method to the input image data, and generates output image data. The generated output image data is output to the HDD 103. Note that the output destination is not limited to the HDD 103 and may be output to the external memory 113 connected to the general-purpose I / F 104, for example.

  Next, processing performed according to the logical configuration of the image processing apparatus 100 described with reference to FIG. 4 will be described. FIG. 5 is a flowchart showing a flow of processing for generating a noise reduced image in the present embodiment. This series of processing is implemented by reading a computer-executable program describing the following procedure from the HDD 103 onto the RAM 102 and then executing the program by the CPU 101.

  In step 501, the image processing apparatus 100 acquires input image data. The acquired input image data is sent to the estimation unit 402 and the noise reduction processing unit 405.

In step 502, the image processing apparatus 100 acquires block matching parameters. The acquired block matching parameters are sent to the reference histogram generation unit 401 and the estimation unit 402. Here, the block matching parameters input is the number of pixels N _b and the reference pixel number N _S of the target region in the previous NLM method. In the following description, it is assumed that the number of pixels N _b = 9 is a region of interest in a 3 × 3 pixel rectangular region, and the number of reference pixels is the number of pixels N _S = 25 included in the 5 × 5 pixel rectangular region.

In step 503, the reference histogram generation unit 401 generates a reference histogram based on the acquired block matching parameter. Note that the reference histogram may be determined as a statistical distribution of values that the similarity _Cj can take. For example, it is assumed that the above equation (2) is a chi-square distribution with variance 1 and p × q degrees of freedom. In this case, the statistical distribution of values that the similarity C _j can take can be determined as a probability density function of a chi-square distribution with p × q degrees of freedom. FIG. 6 is a diagram illustrating an example of the distribution of the reference histogram. 6A shows the probability density function of the chi-square distribution with 9 degrees of freedom, and FIG. 6B shows the probability density function of the chi-square distribution with 25 degrees of freedom. Here, the degree of freedom p × q matches the number of pixels N _b of the target area acquired in step 502. That is, the target pixel, the reference histogram consisting of statistical distribution of the similarity C _j can take values of each the N _S reference pixels based on the probability density function of the chi-square distribution for the N _b and freedom Determined. The reference histogram may be determined as a statistical distribution of values that the similarity C _j can take, and is not limited to this.

  In step 504, the estimation unit 402 performs processing for estimating the characteristics of the pixel-of-interest neighborhood region. FIG. 7 is a flowchart illustrating details of the process for estimating the characteristics of the region near the target pixel according to the present embodiment.

  In step 701, the estimation unit 402 selects a pixel (target pixel) that is subject to noise reduction processing from the input image data acquired in step 501.

In step 702, the similarity histogram generation unit 403 uses a 3 × 3 pixel region of interest centered on the selected pixel of interest, and performs similarity C _j for each reference pixel of 5 × 5 pixels by block matching. And a similarity histogram is generated. If the noise variance of the input image data is 1, the similarity histogram has a shape close to the above-described FIG.

  In step 703, the estimation unit 402 acquires the reference histogram generated in step 503.

  In step 704, the inter-histogram approximation degree deriving unit 404 obtains an approximation degree between the reference histogram obtained in step 703 and the similarity histogram obtained in step 702. For example, when the histogram is divided into n sections, the approximation d between the histograms can be obtained by the following equation (4).

Here, H _ref [i] is the i-th frequency of the reference histogram, H _calc [i] is the i-th frequency of the similarity histogram obtained in step 702, and min returns a smaller value of the two elements. It is a function. Then, when the value of d becomes the maximum (d _max ), the two histograms match. Since the reference pixel is 5 × 5 pixels, d _max = 25. That is, approximation when the histograms match (for example, n = 2, H _ref [1] = 5, H _ref [2] = 20, H _calc [1] = 5, H _calc [2] = 20) The degree d is as follows.

Also, approximation when histograms do not match (for example, n = 2, H _ref [1] = 5, H _ref [2] = 20, H _calc [1] = 20, H _calc [2] = 5) The degree d is as follows.

Thus, the value of the approximation d when the histograms do not match is smaller than d _max . The method for deriving the degree of approximation between histograms is not limited to the method using the above formula (4), and any method such as a method for obtaining a correlation or a method based on a difference can be used. Depending on how the degree of approximation between histograms is derived, the way of evaluating the derived result may vary (for example, the smaller the derived value, the higher the degree of approximation may be evaluated). The description will be continued on the assumption that the degree of approximation d is obtained using equation (4). In the case of the present embodiment, as the value of the approximation d between histograms increases, the possibility that the characteristic is similar to the pixel of interest increases, for example, the region near the pixel of interest is a flat region. On the other hand, the smaller the value of the approximation d between histograms, the higher the possibility that the characteristics are different from the target pixel, for example, an edge region different from the target pixel is included in the region near the target pixel. Therefore, in this embodiment, the degree of approximation d between the histograms derived in this way is used as an index for estimating the characteristics of the neighborhood region of the selected target pixel.

  In step 705, the estimation unit 402 determines whether or not the processing has been completed for all pixels in the image data. If there is an unprocessed pixel, the process returns to step 701 to continue the process. On the other hand, if the processing for all the pixels has been completed, the present processing is exited.

  The above is the content of the process for estimating the characteristics of the region near the target pixel according to the present embodiment.

  Returning to the flowchart of FIG.

  In step 505, the noise reduction processing unit 405 applies noise reduction processing by the NLM method to the input image data based on the weight control parameter updated in accordance with the characteristics of the pixel-of-interest neighborhood estimated in step 504. . FIG. 8 is a flowchart showing details of the noise reduction processing according to the present embodiment.

In step 801, the noise reduction processing unit 405 sets a reference weight control parameter _hbase . The value of the weight control parameter h _base may be a predetermined value that provides a standard noise reduction effect, for example.

  In step 802, the noise reduction processing unit 405 selects a pixel (target pixel) that is a target of noise reduction processing from the input image data acquired in step 501.

  In step 803, the noise reduction processing unit 405 obtains the estimation result (in the present embodiment, the degree of approximation d) of the area near the target pixel derived in step 504.

In step 804, the weight control parameter setting unit 406 updates the value of the weight control parameter h used in the noise reduction processing by the NLM method, based on the degree of approximation d as the obtained estimation result. Here, when the value of the estimation result (approximation d) is large, it is highly possible that the region near the target pixel is a flat region as described above. Therefore, a new value is set for the weight control parameter h so as to obtain a high noise reduction effect. On the other hand, when the value of the estimation result (approximation d) is small, there is a high possibility that an edge region or the like is included in the pixel vicinity region as described above. Therefore, by setting a smaller value for the weight control parameter h, the noise reduction effect is weakened and the edge is stored. Specifically, when the reference parameter is h _base , the changed parameter is h _new , and the maximum value of the estimation result is d _max , the weight control parameter h to be set is obtained by the following equation (5). Can do.

Here, it is assumed that the reference pixel is 5 × 5 pixels (d _max = 25), h _base = 5, d = 25 when the estimation result (approximation d) is large, and d = 10 when the estimation result is small. H _new derived based on the above equation (5) is h _new = 5 when the degree of approximation d is large, and h _new = 2 when the degree of approximation d is small. Thus, when the degree of approximation d as the estimation result is large, the value of h _new is also large, and when the degree of approximation d as the estimation result is small, the value of h _new is also small. The method for determining the value of the weight control parameter h is not limited to the above example. For example, one or more threshold values may be provided in advance, and the degree of approximation d as an estimation result may be classified based on the magnitude relationship with the threshold value, and the value of the weight control parameter h may be determined.

In step 805, the NLM application unit 407 derives the weight w by the above equation (3) according to the new weight control parameter h _new updated in step 804, and the pixel value after replacement of the pixel of interest selected in step 802 I _new is obtained using the above equation (1).

  In step 806, the noise reduction processing unit 405 determines whether the noise reduction processing by the NLM method has been completed for all the pixels of the input image data. If there is an unprocessed pixel, the process returns to step 802 to continue the process. On the other hand, if the process has been completed for all pixels, the process is exited.

  The above is the content of the noise reduction processing according to the present embodiment.

  Returning to the flowchart of FIG.

  In step 506, the image processing apparatus 100 outputs the image data generated in step 505 (image data on which noise reduction processing has been performed).

  In the flowchart of FIG. 5 described in the present embodiment, the noise reduction processing is performed after performing the characteristic estimation processing of the region near the target pixel for all the pixels in the input image data. It is not limited. For example, when the characteristic estimation of the neighboring area for the target pixel selected in the characteristic estimation process of the target pixel neighboring area is completed, the noise reduction process for the target pixel is continuously performed, and then the next target pixel You may make it transfer to the characteristic estimation process of the near region about.

  According to the present embodiment, the characteristics of the vicinity region of the target pixel are estimated, and the weight control parameter used for the noise reduction process is set based on the estimation result. This makes it possible to obtain a suitable noise reduction effect.

(Example 2)
In the first embodiment, the degree of approximation between the reference histogram and the similarity histogram is estimated as the characteristic of the pixel-of-interest neighborhood region. In the method according to the first embodiment, when the target pixel is similar to all the reference pixels, the characteristics of the region near the target pixel can be accurately estimated. However, when a periodic pattern such as a lattice exists in the input image data, the shape of the similarity histogram changes, and the characteristics of the pixel-of-interest neighborhood that can be estimated are limited. Therefore, an embodiment in which the characteristics of the vicinity pixel of interest are estimated in more detail by determining the characteristics of the similarity histogram and using this will be described as a second embodiment. The description of the parts common to the first embodiment will be omitted or simplified, and the differences will be mainly described below.

  FIG. 9 is a schematic diagram illustrating a logical configuration of the image processing apparatus 100 according to the present embodiment, which is the same as the first embodiment in that the reference histogram generation unit 401, the estimation unit 402, and the noise reduction processing unit 405 are configured. It is. However, a histogram feature determination unit 901 is added to the estimation unit 402 according to the present embodiment, and feature determination information is input. The feature determination information is read from the HDD 103 or the external memory 107 based on an instruction from the CPU 101. Further, the user may directly input from the input device 112 such as a keyboard or a mouse via the UI screen.

The histogram feature determination unit 901 determines the feature of the similarity histogram generated for each pixel of interest by the similarity histogram generation unit 403 according to the feature determination information. Here, the feature of the similarity histogram includes any one of a peak position, a peak size, and a peak number in a statistical distribution of values that can be taken by the similarity C _j described above, or a combination thereof. FIGS. 10A and 10B are diagrams showing how the characteristics of the similarity histogram are determined. The solid line indicates the distribution of the reference histogram, and the dotted line indicates the distribution of the similarity histogram. .

FIG. 10A is an example of a distribution of values that can be taken by the similarity C _j when the vicinity region of the target pixel is a lattice pattern. When a periodic pattern such as a lattice pattern is in the region near the target pixel, the similarity histogram has a plurality of peaks according to the number of periodic patterns, and one of the peak positions matches the peak position in the reference histogram. It has the characteristics. This is because the pixel is divided into a pixel having high similarity and a pixel having low similarity to the target pixel. Therefore, by giving the number of peaks in the similarity histogram as feature determination information, it is possible to estimate whether or not the region near the pixel of interest is a periodic pattern.

FIG. 10B is an example of a distribution of values that can be taken by the similarity C _j when the vicinity region of the pixel of interest has a random texture. When there is no specific pattern as in a random texture, the similarity histogram and the reference histogram have different peak positions, and the similarity histogram has a feature that the peak size is larger. This is because reference pixels having a high degree of similarity with the target pixel are reduced. Therefore, it is possible to estimate whether or not the neighborhood region of the pixel of interest is a random texture by giving the peak position and peak size of the reference histogram as feature determination information.

  As described above, by determining and using the feature of the similarity histogram using the feature determination information, it is possible to estimate the characteristics of the region near the target pixel in more detail. Note that what features are derived from the similarity histogram is not limited to the above-described example, and can be arbitrarily set.

  Next, processing performed according to the logical configuration of the image processing apparatus 100 described with reference to FIG. 9 will be described. FIG. 11 is a flowchart showing the flow of main processing for generating a noise-reduced image in the present embodiment. A major difference from the flowchart of FIG. 5 according to the first embodiment is that a process (step 1104) for acquiring feature determination information is added. In this embodiment, after the reference histogram is generated, the above-described feature determination information is acquired. The other points are the same as in the first embodiment. That is, steps 1101 to 1103 correspond to steps 501 to 503 in FIG. 5, and steps 1105 to 1107 correspond to steps 504 to 506 in FIG. However, as described below, the contents of the characteristic estimation process and the noise reduction process in the vicinity region of the target pixel are different from those in the first embodiment.

(Characteristic estimation process for the area near the target pixel)
FIG. 12 is a flowchart illustrating details of the process for estimating the characteristics of the region near the target pixel according to the present embodiment.

  In step 1201, the estimation unit 402 selects a pixel (target pixel) to be subjected to noise reduction processing from the input image data acquired in step 1101.

In step 1202, the similarity histogram generating unit 403, by using a region of interest centered on the target pixel selected for each reference pixel to derive a similarity C _j by block matching, to generate a similarity histogram .

  In step 1203, the estimation unit 402 acquires the reference histogram generated in step 1103.

  In step 1204, the inter-histogram approximation degree deriving unit 404 obtains an approximation degree between the reference histogram acquired in step 1203 and the similarity histogram generated in step 1202.

  In step 1205, the histogram feature determination unit 901 determines the feature of the similarity histogram obtained in step 1202 according to the feature determination information acquired in step 1104.

  With this processing, in this embodiment, the degree of approximation d between the histograms obtained in step 1204 and the characteristics of the similarity histogram determined in step 1205 are estimated as the characteristics of the neighborhood region of the pixel of interest being selected. become.

  In step 1206, the estimation unit 402 determines whether or not the processing has been completed for all pixels in the input image data. If there is an unprocessed pixel, the process returns to step 1201 to continue the process. On the other hand, if the processing for all the pixels has been completed, the present processing is exited.

  The above is the content of the characteristic estimation process of the region near the target pixel according to the present embodiment.

(Noise reduction processing)
FIG. 13 is a flowchart illustrating details of noise reduction processing according to the present embodiment.

In step 1301, the noise reduction processing unit 405 sets a weight control parameter h _base as a reference.

  In step 1302, the noise reduction processing unit 405 selects a pixel (target pixel) that is a target of noise reduction processing from the input image data acquired in step 1101.

  In step 1303, the noise reduction processing unit 405 acquires the estimation result of the region near the target pixel derived in step 1105 (in this embodiment, the feature of the approximation d and the similarity histogram).

  In step 1304, the weight control parameter setting unit 406 determines whether or not the neighborhood area for the selected pixel of interest is a random texture area based on the acquired feature of the similarity histogram. Specifically, it is determined that the region is a random texture region when there is a shift in peak position compared to the reference histogram or when the peak of the similarity histogram is larger. If it is determined that the region near the target pixel is a random texture region, the process proceeds to step 1305. On the other hand, if it is determined that the region is not a random texture region, the process proceeds to step 1306.

In step 1305, the weight control parameter setting unit 406 sets the weight control parameter h _{s to} a weight control parameter h _{s having} a predetermined value (a value smaller than a value that can be set in step 1307, which will be described later). Set. This is to reduce blurring of the image because there are few reference pixels having high similarity to the target pixel in the random texture region. For example, if the value of the reference weight control parameter h _base set in step 1301 is 3, 1.5 which is a half value (3 * 0.5) is set as h _s .

In step 1306, the weight control parameter setting unit 406 determines whether or not the neighboring region for the selected pixel of interest is a periodic pattern region based on the acquired feature histogram feature. Specifically, if the reference histogram and the similarity histogram have the same peak position or the number of peaks in the similarity histogram is plural, it is determined that the region is a periodic pattern region. If it is determined that the periodic pattern area proceeds to step 1308 and performs the noise reduction process by the NLM method under standards become weighting control parameter h _base. This is because in the periodic pattern region, the difference in similarity between the target pixel and the reference pixel becomes large, and therefore a preferable noise reduction result can be obtained without updating the parameter h. On the other hand, if it is determined that the region is not a periodic pattern region, the process proceeds to step 1307.

In step 1307, the weight control parameter setting unit 406 obtains and updates a _new value of the weight control parameter h _new using the above-described equation (5) based on the degree of approximation d as the obtained estimation result. As described above, when the degree of approximation d as the estimation result is large, the value of h _new is also large, and when the degree of approximation d as the estimation result is small, the value of h _new is also small.

In step 1308, the NLM application unit 407 derives the weight w by the above-described equation (3) according to the updated new weight control parameter h _new (or h _base ), and the pixel after replacement of the pixel of interest being selected The value I _new is obtained using the above-described equation (1).

  In step 1309, the noise reduction processing unit 405 determines whether or not noise reduction processing by the NLM method has been completed for all pixels of the input image data. If there is an unprocessed pixel, the process returns to step 1301 to continue the process. On the other hand, if the process has been completed for all pixels, the process is exited.

The above is the content of the noise reduction processing according to the present embodiment.
According to the present embodiment, it is possible to estimate the characteristics of the region near the target pixel in more detail by determining and using the feature of the similarity histogram generated for each target pixel. And it becomes possible to acquire a suitable noise reduction effect by updating a noise reduction parameter based on an estimation result.

(Example 3)
In the first and second embodiments, the value of the weight control parameter for the noise reduction processing by the NLM method is appropriately changed according to the estimation result. Instead of such a method of changing the value of the weight control parameter, it may be configured to apply noise reduction processing (for example, noise reduction processing using a Gaussian filter or a median filter) of a different method depending on the estimation result. I do not care. Furthermore, it is possible to adopt a configuration in which the noise reduction process itself is not applied to a region that is easily blurred, such as the random texture region described in the second embodiment.

(Other examples)
The present invention is also realized by executing the following processing. That is, software (program) for realizing the functions of the above-described embodiments is supplied to a system or apparatus via a network or various storage media, and a computer (or CPU, MPU, etc.) of the system or apparatus reads the program. It is a process to be executed.

Claims (13)

Generate a histogram of the similarity between the target pixel for noise reduction processing and a plurality of reference pixels corresponding to the target pixel, and estimate the characteristics of the neighborhood region of the target pixel based on the generated similarity histogram An estimation means to
Noise reduction processing means for determining weights for the plurality of reference pixels based on characteristics of the vicinity region of the target pixel estimated by the estimation means, and performing noise reduction processing on the target pixel;
An image processing apparatus comprising: The noise reduction processing means includes
Parameter setting means for setting a control parameter for controlling the weight according to the characteristics of the neighborhood region of the target pixel estimated by the estimation means;
The image processing apparatus according to claim 1, wherein the weight is determined based on a control parameter set by the parameter setting unit, and noise reduction processing is performed on the pixel of interest. Reference histogram generation means for generating a histogram serving as a reference according to a statistical distribution according to the block matching parameter,
The estimation means includes
Including inter-histogram approximation derivation means for deriving an approximation between the similarity histogram and the reference histogram, and estimating the approximation between the derived histograms as a characteristic of a neighborhood region of the target pixel. The image processing apparatus according to claim 1, wherein the image processing apparatus is characterized.   The parameter setting means sets the control parameter such that the higher the degree of approximation derived by the inter-histogram approximation degree derivation means, the stronger the noise reduction effect, and the lower the degree of approximation, the weaker the noise reduction effect. The image processing apparatus according to claim 3. The estimation means includes
Further comprising a feature determination means for determining a feature of the similarity histogram,
The feature of the similarity histogram determined by the feature determination unit and the degree of approximation between the histograms derived by the inter-histogram approximation deriving unit are estimated as characteristics of the neighborhood region of the target pixel. Item 5. The image processing apparatus according to Item 3 or 4.   The image processing according to claim 5, wherein the feature of the similarity histogram determined by the feature determination unit is any one of a peak position, a peak size, a peak number, or a combination thereof. apparatus.   When the noise reduction processing unit determines whether or not a region near the target pixel is a periodic pattern region based on the feature of the similarity histogram, and determines that the region is a periodic pattern region In addition, a noise reduction process is performed on the target pixel using a predetermined weight without determining a weight for the plurality of reference pixels based on the characteristics of the vicinity region of the target pixel estimated by the estimation unit. The image processing apparatus according to 5 or 6, wherein the image processing apparatus is performed.   The noise reduction processing means determines whether or not the neighborhood region of the target pixel is a random texture region based on the feature of the similarity histogram, and when it is determined that the region is a random texture region, the estimation The predetermined pixel is used for the target pixel by using a predetermined weight such that the noise reduction effect is weaker than when the weight for the plurality of reference pixels is determined based on the characteristics of the vicinity region of the target pixel estimated by the means. The image processing apparatus according to 5 or 6, wherein noise reduction processing is performed.   The image processing apparatus according to any one of 1 to 8, wherein the noise reduction processing is noise reduction processing by a Non-Local Means method. Generate a histogram of the similarity between the target pixel for noise reduction processing and a plurality of reference pixels corresponding to the target pixel, and estimate the characteristics of the neighborhood region of the target pixel based on the generated similarity histogram An estimation means to
Noise reduction processing means for performing noise reduction processing on the target pixel, wherein the noise reduction processing means performs noise reduction processing of different contents according to the characteristics of the vicinity region of the target pixel estimated by the estimation means;
An image processing apparatus comprising: Generate a histogram of the similarity between the target pixel for noise reduction processing and a plurality of reference pixels corresponding to the target pixel, and estimate the characteristics of the neighborhood region of the target pixel based on the generated similarity histogram An estimation means to
Means for performing noise reduction processing on the target pixel, wherein the noise reduction processing unit switches whether to perform noise reduction processing according to the characteristics of the neighborhood region of the target pixel estimated by the estimation unit;
An image processing apparatus comprising: Generate a histogram of the similarity between the target pixel for noise reduction processing and a plurality of reference pixels corresponding to the target pixel, and estimate the characteristics of the neighborhood region of the target pixel based on the generated similarity histogram An estimation step to
A noise reduction processing step of determining a weight for the plurality of reference pixels based on characteristics of a neighborhood region of the target pixel estimated in the estimation step, and performing a noise reduction process on the target pixel;
An image processing method comprising:   A program for causing a computer to function as the image processing apparatus according to any one of claims 1 to 11.

Images