3D noise reduction method based on guided filtering

Abstract

The invention discloses a 3D noise reduction method based on guided filtering, and relates to the technical field of digital image processing, and the method comprises the steps: carrying out the matching of each pixel point of an input frame through motion estimation, and obtaining a corresponding pixel point in a reference frame; based on each pixel point, respectively extracting image blocks of the pixel point in R, G and B channels as target image blocks, and extracting corresponding guide image blocks in the reference frame for matching; and filtering the target image block and the guide image block in the reference frame by using guide filtering to obtain a denoised output image block. According to the method, a high-quality noise reduction effect can be achieved, image distortion caused by non-translational motion is reduced, hardware implementation is easy, and low-cost implementation is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of digital image processing, and in particular to a 3D noise reduction method based on guided filtering. Background Art

[0002] Digital image acquisition is the process of converting optical signals into electrical signals using sensors such as cameras and storing, transmitting and displaying them in digital form. Digital image processing is the process of processing and optimizing the acquired digital images according to the purpose and scenario of use. Common methods include: image enhancement and restoration, image coding compression, image description, etc.

[0003] Image noise reduction is a crucial aspect of image enhancement and restoration. It's widely used in a variety of fields, including automotive imaging and surveillance cameras. In low-light environments, such as outdoors in the evening or at night, cameras generate a significant amount of noise due to insufficient light sensitivity. To ensure clear display, noise reduction is necessary in low-light images. This allows the naked eye to clearly capture real-time information and store clear images for later use.

[0004] 3D noise reduction is a key technology in image processing. It exploits the temporal correlation of image sequences to reduce noise. It uses motion estimation to detect corresponding pixel blocks between two frames, then performs temporal fusion on these matching pixel blocks, leveraging the statistical randomness of noise to reduce its impact on image quality. Currently, mainstream 3D noise reduction methods primarily employ averaging or weighted averaging for temporal fusion. However, for non-translational objects, simple averaging can result in blurring and artifacts, which can negatively impact image quality.

[0005] Guided filtering is an important image processing filtering technique. Its core principle is to filter the target image using information from a guide image. It calculates the filtered output of the target image by defining a local linear model on the guide image. This filtering method exhibits edge-preserving properties, effectively preserving image edges and details while removing noise and smoothing the image. Local linearity also ensures that the local basic structure of the image remains unchanged, preventing image distortion caused by misalignment due to non-translational motion.

[0006] Nowadays, image processing plays an increasingly important role in many fields such as smart cars, smart homes, and smart monitoring, and also puts higher requirements on image quality. A new and more effective 3D noise reduction time domain fusion method has practical value.

[0007] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0008] In view of the problems in the related art, the present invention proposes a 3D noise reduction method based on guided filtering to overcome the above technical problems existing in the existing related art.

[0009] To this end, the specific technical solutions adopted in the present invention are as follows:

[0010] A 3D noise reduction method based on guided filtering, comprising:

[0011] Use motion estimation to match each pixel in the input frame to obtain the corresponding pixel in the reference frame;

[0012] Based on each pixel point, its image blocks in R, G, and B channels are extracted as target image blocks, and the corresponding guide image blocks are extracted in the reference frame for matching;

[0013] The target image block and the guide image block in the reference frame are filtered using guided filtering to obtain a denoised output image block.

[0014] Furthermore, motion estimation is used to match each pixel of the input frame to obtain the corresponding pixel in the reference frame, including:

[0015] The current image frame to be denoised is used as the input frame, and the image that has completed denoising is used as the reference frame;

[0016] For each pixel of the input frame, select the pixel with the same position in the reference frame, and use the pixel with the same position as the center to define the matching range preset by the reference frame;

[0017] Centering the pixel in the input frame, a preset image block is selected as the matching block, and a search is performed within the preset matching range of the reference frame to find several similar reference blocks.

[0018] Calculate the distance between the image block in the reference frame and the matching block in the input frame, compare the obtained distance with a preset threshold, and retain the image blocks with a distance less than the threshold;

[0019] Through the motion estimation and matching process, the matching block in the reference frame corresponding to each pixel point in the input frame is obtained.

[0020] Furthermore, the formula for calculating the distance between the image block in the reference frame and the matching block in the input frame is:

[0021]

[0022] Where d represents the distance between the image block in the reference frame and the matching block in the input frame; I represents the matching block; I(i, j, c) represents the pixel value of the matching block at coordinates (i, j, c) in the input frame; I′(i, j, c) represents the pixel value of the image block at coordinates (i, j, c) in the reference frame; and m represents the side length of the image block.

[0023] Furthermore, based on each pixel point, extracting its image blocks in the R, G, and B channels as target image blocks, and extracting corresponding guide image blocks in the reference frame for matching includes:

[0024] For each pixel, extract its image blocks in R, G, and B channels as the target image blocks;

[0025] In the reference frame, the position matching the target image block is extracted, and the guide image blocks of the R, G, and B channels are extracted respectively, and it is ensured that the extracted target image block and the guide image block have the same size.

[0026] Furthermore, the target image block and the guide image block in the reference frame are filtered using guided filtering to obtain a noise-reduced output image block including:

[0027] Calculating the coefficients in the guided filter linear model based on the relationship between the guided image block and the target image block;

[0028] Based on the obtained coefficients, the pixel value of the center point of the output image block is calculated, and for each pixel, the output pixel values ​​of its R, G, and B channels are calculated respectively;

[0029] In the filtering process, the non-local mean idea is used to set weights to balance the contributions of different guide image blocks;

[0030] Traverse all pixels in the image, perform filtering on the target image block and the guide image block respectively, and finally generate a complete output image block after noise reduction.

[0031] Furthermore, the formula for calculating the coefficients in the guided filter linear model is:

[0032]

[0033] Where a and b are coefficients; I′ rl represents the guidance image block; I′ rl (i, j) represents the pixel value of the pixel at coordinate (i, j) in the guide image block; I r Represents the target image block; I r(i, j) represents the pixel value of the pixel at coordinate (i, j) in the target image block; s represents the distance from the center point to the edge of the image block; ε represents the regularization parameter that limits the size of a; m represents the side length of the image block; n represents the matching range; Represents weight.

[0034] Furthermore, the formula for calculating the pixel value of the center point of the output image block is:

[0035]

[0036] Where, Represents the pixel value of the center point of the output image block; a and b both represent coefficients; represents the weight; I′ rl Represents the guidance image patch.

[0037] Furthermore, during the filtering process, the non-local mean concept is used to set weights to balance the contributions of different guide image blocks, including:

[0038] Multiple guide image patches are used instead of a single guide image patch, and the translation similarity within the image is exploited to improve the noise reduction effect in smooth areas.

[0039] By setting a weight function to balance the contributions of different guide image blocks, the noise reduction effect is improved by combining the structural similarity and pixel value similarity of the image blocks;

[0040] Based on computing resource limitations, the weight function is replaced with a monotonically decreasing function, and the weight calculation is optimized through a lookup table.

[0041] The beneficial effects of the present invention are:

[0042] 1. This invention uses a block-matching-based method for motion estimation. By comparing the similarity of image blocks with adjacent positions in the previous frame, it finds several matching blocks for time-domain fusion. This method does not affect compatibility with other modules. Improvements have been made in the time-domain fusion process, replacing the mean fusion method with a guided filtering method to enhance noise reduction. This method achieves high-quality noise reduction, reduces image distortion caused by non-translational motion, and is easy to implement in hardware, making it cost-effective.

[0043] 2. The present invention replaces mean fusion with guided filtering, and modifies the general guided filtering of a single guided image into guided filtering of multiple guided images, making full use of spatial similarity to further improve the noise reduction effect. After obtaining a number of matching blocks through motion estimation, these matching blocks are used as guide images, and the current frame is used as the target image. The smoothing property of the previous frame that has completed noise reduction is combined with the structural information of the current frame to avoid image distortion caused by changes in the local structure of the image due to non-translational motion. At the same time, the effect of maintaining edges and reducing noise is better than traditional mean fusion. By replacing a single guided image with multiple guided images, the spatial similarity within the image is further utilized, and the impact of accidental matching errors caused by randomness is reduced. At the same time, the guided filtering itself consumes little computing resources. Compared with traditional 3D noise reduction, the main computing resources are also consumed in the motion estimation part, which is easy to implement at low cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 is a flow chart of a 3D noise reduction method based on guided filtering according to an embodiment of the present invention;

[0046] Figure 2 2 is a schematic diagram of a process of RGB image motion estimation and obtaining a guide image block and a target image block according to an embodiment of the present invention;

[0047] Figure 3 3 is a schematic diagram of a process of Bayer image motion estimation and obtaining a guide image block and a target image block according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. By referring to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and the advantages of the present invention.

[0049] According to an embodiment of the present invention, a 3D noise reduction method based on guided filtering is provided.

[0050] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, the 3D noise reduction method based on guided filtering according to an embodiment of the present invention includes:

[0051] S1. Use motion estimation to match each pixel of the input frame to obtain the corresponding pixel in the reference frame;

[0052] S2. Based on each pixel point, extract its image blocks in the R, G, and B channels as the target image blocks, and extract the corresponding guide image blocks in the reference frame for matching;

[0053] S3. Perform filtering processing on the target image block and the guide image block in the reference frame using guided filtering to obtain a denoised output image block.

[0054] In this optional embodiment, matching each pixel point of the input frame using motion estimation to obtain the corresponding pixel point in the reference frame includes:

[0055] The current image frame to be denoised is used as the input frame, and the image that has completed denoising is used as the reference frame;

[0056] For each pixel of the input frame, select the pixel with the same position in the reference frame, and use the pixel with the same position as the center to define the matching range preset by the reference frame;

[0057] Centering the pixel in the input frame, a preset image block is selected as the matching block, and a search is performed within the preset matching range of the reference frame to find several similar reference blocks.

[0058] Calculate the distance between the image block in the reference frame and the matching block in the input frame, compare the obtained distance with a preset threshold, and retain the image blocks with a distance less than the threshold;

[0059] Through the motion estimation and matching process, the matching block in the reference frame corresponding to each pixel point in the input frame is obtained.

[0060] In this optional embodiment, based on each pixel point, extracting its image blocks in R, G, and B channels as target image blocks, and extracting corresponding guide image blocks in the reference frame for matching includes:

[0061] For each pixel, extract its image blocks in R, G, and B channels as the target image blocks;

[0062] In the reference frame, the position matching the target image block is extracted, and the guide image blocks of the R, G, and B channels are extracted respectively, and it is ensured that the extracted target image block and the guide image block have the same size.

[0063] In this optional embodiment, filtering the target image block and the guide image block in the reference frame using guided filtering to obtain a noise-reduced output image block includes:

[0064] Calculating the coefficients in the guided filter linear model based on the relationship between the guided image block and the target image block;

[0065] Based on the obtained coefficients, the pixel value of the center point of the output image block is calculated, and for each pixel, the output pixel values ​​of its R, G, and B channels are calculated respectively;

[0066] In the filtering process, the non-local mean idea is used to set weights to balance the contributions of different guide image blocks;

[0067] Traverse all pixels in the image, perform filtering on the target image block and the guide image block respectively, and finally generate a complete output image block after noise reduction.

[0068] In this optional embodiment, during the filtering process, using the non-local mean concept, setting weights to balance contributions of different guide image blocks includes:

[0069] Multiple guide image patches are used instead of a single guide image patch, and the translation similarity within the image is exploited to improve the noise reduction effect in smooth areas.

[0070] By setting a weight function to balance the contributions of different guide image blocks, the noise reduction effect is improved by combining the structural similarity and pixel value similarity of the image blocks;

[0071] Based on computing resource limitations, the weight function is replaced with a monotonically decreasing function, and the weight calculation is optimized through a lookup table.

[0072] It should be noted that when performing 3D noise reduction, the previous frame is used as a reference frame. After inputting a new frame, 3D noise reduction is performed on it according to the following steps. The present invention is applicable to both RGB format images and Bayer format images, which are explained below:

[0073] 1) For RGB images, perform 3D noise reduction as follows:

[0074] The first step is to take each pixel in the input frame and take a pixel block centered on it. Motion estimation is performed by searching for matching blocks in the reference frame to find the corresponding pixel. Specifically, the current input image frame to be denoised is called the input frame, and the previous image frame that has undergone denoising and is stored in the device register is called the reference frame. For each pixel in the input frame, a pixel in the reference frame with the same position is selected as the matching range, with it as the center. An m×m image block centered on the pixel to be matched in the input frame is selected as the matching block, and several matching reference blocks are found by searching within the aforementioned n×n matching range. The side length m of the image block used for matching usually corresponds to the selected matching range n and can be adjusted as needed. Common values ​​of m and n include m=5, n=9, 11, 13; m=7, n=11, 13, 15, 17, etc.

[0075] When performing matching, m×m image blocks within the n×n matching range are selected in sequence, and the search step size is set to 1 for images in RGB format.

[0076] Calculate the difference between the image block in the selected reference frame and the matching block. Specifically, for an RGB format image, let the matching block be I, where the pixel value of the pixel with coordinates (i, j, c) is I(i, j, c), and let the image block in the selected reference frame be I′, where the pixel value of the pixel with coordinates (i, j, c) is I′(i, j, c). Here, c = 1, 2, 3, representing the channel values ​​1, 2, and 3 corresponding to the red, green, and blue channels respectively. The formula for calculating the distance between the image block in the reference frame and the matching block in the input frame is:

[0077]

[0078] Where d represents the distance between the image block in the reference frame and the matching block in the input frame; I represents the matching block; I(i, j, c) represents the pixel value of the matching block at coordinates (i, j, c) in the input frame; I′(i, j, c) represents the pixel value of the image block at coordinates (i, j, c) in the reference frame; and m represents the side length of the image block.

[0079] Compare this distance with the threshold th, and retain the image blocks whose distance is less than the threshold. The threshold is determined based on the value range of the pixel points and the noise level of the image. For example, for 8-bit data with a data range of [0, 255], the noise level is generally selected from 8 to 32 from low to high. The noise level can be determined by the exposure gain during exposure and other methods. After the first step above, several reference blocks can be obtained. The number of reference blocks that meet the conditions is denoted as k, and these reference blocks are denoted as I1, ...I k , and remember the distances obtained are d1,...d k .

[0080] The second step is to obtain the guide image block and the target image block. Guided filtering is suitable for single-channel scenes. For multiple color channels, the channels need to be filtered separately. Specifically, for RGB format images, each pixel has three channels: R, G, and B. Take the R channel image block I centered on the pixel. r , G channel image I g , and B channel image block I b , and then take the image block I′ centered on the point it matches in the reference frame r1 ...I′ rk ..., I′ g1 ...I′ gk ..., I′ b1 ...I′ bk ...these image patches have the same size, i.e. m×m.

[0081] The results of the first and second steps are shown in the following figure: Figure 2 As shown in the figure, the left side is the input frame, the pixel marked with a black dot is the current pixel, and the pixel block centered on it is selected, which is represented by a box in the figure. The arrow below represents the RGB three-channel image block centered on it as the target image block I r , I g , I b The right side is the reference frame. The pixel blocks at the same position as the input frame are represented by solid boxes. The dotted boxes represent the pixel blocks obtained by motion estimation matching, and their center points are the matched pixels. The arrows indicate that the RGB three-channel image blocks represented by the dotted boxes are extracted as the guide image blocks I′. r , I′ g , I′ b The same operation is performed on the remaining reference blocks to obtain other guide image blocks of the corresponding channels.

[0082] The third step is to perform guided filtering to obtain the output image block through the guided image block and the target image block. Take the R channel as an example:

[0083] Take I' r1 ...I′ rk is the guide image block, and the guide image block is I′ rl The pixel value I' of the pixel at coordinate (i, j) rl (i,j). Take I r is the target image block, and the pixel value of the pixel at coordinate (i, j) is I r (i, j). Where -s≤i, j≤s, s is the distance from the center point to the edge of the image block. In the image block, the coordinates of the center point are (0, 0). The formula for calculating the coefficients in the guided filter linear model is:

[0084]

[0085] Where a and b are coefficients; k is the number of reference blocks obtained in the first step, I′ rl I′ represents the guidance image block obtained in the first step, l = 1, 2, ..., k; rl (i, j) represents the pixel value of the pixel at coordinate (i, j) in the guide image block, l = 1, 2, ..., k; I r Represents the target image block; I r (i, j) represents the pixel value of the pixel at coordinates (i, j) in the target image block; s represents the distance from the center point to the edge of the image block; ε represents the regularization parameter that limits the size of a. A larger value of ε is equivalent to a larger proportion of the guided image information in the output image, which makes the image smoother. However, if the value is too large, it is prone to distortion. In this invention, the value is generally [0.5, 2]. m represents the side length of the image block, m = 2s + 1; n represents the matching range. represents weights, d1,...d k That is the reference block I1, ...I obtained in the first step k The corresponding distance, r is used as a subscript to indicate that the operation is now being performed on the R channel.

[0086] After obtaining the coefficients in the third step above, use the formula for calculating the pixel value of the center point of the output image block to find the pixel value of the center point in the output image:

[0087]

[0088] Where, Represents the pixel value of the center point of the output image block; a and b both represent coefficients; represents the weight; k is the number of reference blocks obtained in the first step, I′ rl represents the guide image block, l = 1, 2, ..., k.

[0089] Similarly, after applying the same operation to the G channel and the B channel, the corresponding output value of the pixel can be obtained.

[0090] On the basis of guided filtering, the present invention introduces the idea of ​​non-local means (NLM) and uses multiple guided image blocks instead of the previous single guided image block. On the one hand, it can make full use of the translation similarity within the image and achieve better noise reduction effect for the smooth part; on the other hand, it can make full use of the information in the reference frame to further improve the noise reduction effect and avoid the influence caused by the non-optimal matching of some image blocks due to the randomness of the noise. At the same time, by setting the weight e -dl, which combines the structural similarity of image blocks represented by guided filtering and the pixel value similarity of image blocks represented by distance, can further make better use of image information while reducing the impact of pixel blocks with large differences. In order to save resources, Replace with In practical applications, the corresponding weight W(d l ) to further optimize the effect and save resources.

[0091] After traversing all pixels in the image, the complete output image can be obtained.

[0092] 2) For Bayer format images, perform 3D denoising as follows:

[0093] Step 1: For each pixel point of the input frame, take the pixel block centered on it, perform motion estimation by searching for matching blocks in the reference frame, and find the corresponding pixel point. Specifically, the current input image frame to be denoised is called the input frame, and the previous image frame that has completed denoising and is stored in the device register is called the reference frame. For each pixel point of the input frame, select the pixel point with the same position as it in the reference frame, take it as the center, and select an n×n range as the matching range. Select the m×m image block centered on the pixel point to be matched in the input frame as the matching block, and find several matching reference blocks by searching within the aforementioned n×n matching range. The side length m of the image block used for matching usually has a corresponding relationship with the selected matching range n, and can be adjusted as needed. Common values ​​of m and n are m=5, n=9, 11, 13; m=7, n=11, 13, 15, 17, etc.

[0094] When performing matching, m×m image blocks within the n×n matching range are selected in sequence. For Bayer format image blocks, it is important to note that the same color channels must correspond to each other during the search, and the search step size is set to 2.

[0095] Calculate the difference between the image block in the selected reference frame and the matching block. Specifically, for the Bayer format image, let the matching block be I, where the pixel value of the pixel with coordinates (i, j) is recorded as I(i, j), let the image block in the selected reference frame be I′, where the pixel value of the pixel with coordinates (i, j) is recorded as I′(i, j), then the formula for calculating the distance between the image block in the reference frame and the matching block in the input frame is:

[0096]

[0097] Where d represents the distance between the image block in the reference frame and the matching block in the input frame; I represents the matching block; I(i, j, c) represents the pixel value of the matching block at coordinates (i, j, c) in the input frame; I′(i, j, c) represents the pixel value of the image block at coordinates (i, j, c) in the reference frame; and m represents the side length of the image block.

[0098] Compare this distance with the threshold th, and retain the image blocks whose distance is less than the threshold. The threshold is determined based on the pixel value range and the noise level of the image. For example, for 8-bit data with a data range of [0, 255], the noise level is generally selected from 8 to 32 from low to high. The noise level can be determined by the exposure gain during exposure and other methods. After the above step 1, several reference blocks can be obtained, denoted as I1, ...I k , and remember the distances obtained are d1,...d k .

[0099] Step 2: Obtain the guide and target image blocks. For Bayer format images, each pixel has only one channel, and the surrounding pixels store pixel values ​​for other channels. When selecting an image block, use the block as the center and select with a step size of 2, that is, only select the surrounding pixels with the same channel.

[0100] The results of step one and step two are shown in the following figure: Figure 2 As shown, the left side of the figure is the input frame, and the pixel block used for matching is marked with a black frame, and its center point is the current pixel point. The surrounding pixels of the same channel are selected with a step size of 2 and merged into an image block as the target image, denoted as I. The right side is the reference frame, and the matched image block is also marked with a black frame. The surrounding pixels of the same channel are selected with a step size of 2 and merged into an image block as the guide image, denoted as I'. The figure selects a 3×3 image block as an example as the target image block and the guide image block. In practical applications, it is more common to select a 5×5 range. For the reference blocks I1,...I in step 1 k Perform this operation separately to obtain the target image block I in the current frame and the guide image blocks I1′, ...I in the reference frame. k ' is used in subsequent steps.

[0101] Step 3: Perform guided filtering. k ′ is the guide image block, and the guide image block I l The pixel value of the pixel point with coordinates (i, j) in ′ is I l'(i, j). Let I be the target image block, and the pixel value of the pixel at coordinate (i, j) be I(i, j). Where -s≤i, j≤s, s is the distance from the center point to the edge of the image block. In the image block, the coordinates of the center point are (0, 0). First, calculate the coefficients in the guided filter linear model according to the following formula:

[0102]

[0103] Where a and b are coefficients; k is the number of reference blocks obtained in step 1, I l ′ represents the guide image block obtained in step 1, l = 1, 2, ..., k; I l ′(i, j) represents the pixel value of the pixel at coordinate (i, j) in the guide image block; I l Represents the target image block; I l (i, j) represents the pixel value of the pixel at coordinates (i, j) in the target image block; s represents the distance from the center point to the edge of the image block; ε represents the regularization parameter that limits the size of a. A larger value of ε is equivalent to a larger proportion of the guided image information in the output image, which makes the image smoother. However, if the value is too large, it is prone to distortion. In this invention, the value is generally [0.5, 2]. m represents the side length of the image block, m = 2s + 1; n represents the matching range. represents weights, d1,...d k That is the reference block I1, ...I obtained in the first step k The corresponding distance.

[0104] After obtaining the coefficients in step 3 above, use the following formula to calculate the pixel value of the center point in the output image:

[0105]

[0106] Where, Represents the pixel value of the center point of the output image block; a and b both represent coefficients; represents the weight; k is the number of reference blocks obtained in step 1; I l ′ represents the guide image block, l=1, 2, ..., k.

[0107] Similarly, after applying the same operation to the G channel and the B channel, the corresponding output value of the pixel can be obtained.

[0108] On the basis of guided filtering, the present invention introduces the idea of ​​non-local means (NLM), and uses multiple guided image blocks instead of the previous single guided image block. On the one hand, it can make full use of the translation similarity within the image and achieve better noise reduction effect for the smooth part; on the other hand, it makes more full use of the information in the reference frame, further improves the noise reduction effect, and avoids the influence caused by the non-optimal matching of some image blocks due to the randomness of noise. At the same time, by setting the weights Combining the structural similarity of image blocks represented by guided filtering and the pixel value similarity of image blocks represented by distance can further make better use of image information while reducing the impact of pixel blocks with large differences. Replace with In practical applications, the corresponding weight W(d l ) to further optimize the effect and save resources.

[0109] After traversing all pixels in the image, the complete output image can be obtained.

[0110] In summary, with the help of the above-mentioned technical solution of the present invention, the present invention replaces mean fusion with guided filtering, and modifies the general guided filtering of a single guided image into guided filtering of multiple guided images, making full use of spatial similarity to further improve the noise reduction effect. After obtaining several matching blocks through motion estimation, these matching blocks are used as guide images, and the current frame is used as the target image. The smoothness of the noise reduction completed in the previous frame is combined with the structural information of the current frame to avoid image distortion caused by changes in the local structure of the image due to non-translational motion. At the same time, the effect of edge preservation and noise reduction is better than traditional mean fusion. By replacing a single guided image with multiple guided images, the spatial similarity within the image is further utilized and the impact of accidental matching errors caused by randomness is reduced. At the same time, guided filtering itself consumes little computing resources. Compared with traditional 3D noise reduction, the main computing resources are also consumed in the motion estimation part, which is easy to implement at low cost.

[0111] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A 3D noise reduction method based on guided filtering, characterized in that: include: Use motion estimation to match each pixel in the input frame to obtain the corresponding pixel in the reference frame; Based on each pixel point, its image blocks in R, G, and B channels are extracted as target image blocks, and the corresponding guide image blocks are extracted in the reference frame for matching; The target image block and the guide image block in the reference frame are filtered using guided filtering to obtain a denoised output image block.

2. The 3D noise reduction method based on guided filtering according to claim 1, characterized in that: The matching of each pixel point of the input frame by using motion estimation to obtain the corresponding pixel point in the reference frame includes: The current image frame to be denoised is used as the input frame, and the image that has completed denoising is used as the reference frame; For each pixel of the input frame, select the pixel with the same position in the reference frame, and use the pixel with the same position as the center to define the matching range preset by the reference frame; Centering the pixel in the input frame, a preset image block is selected as the matching block, and a search is performed within the preset matching range of the reference frame to find several similar reference blocks. Calculate the distance between the image block in the reference frame and the matching block in the input frame, compare the obtained distance with a preset threshold, and retain the image blocks with a distance less than the threshold; Through the motion estimation and matching process, the matching block in the reference frame corresponding to each pixel point in the input frame is obtained.

3. The 3D noise reduction method based on guided filtering according to claim 2, characterized in that: The formula for calculating the distance between the image block in the reference frame and the matching block in the input frame is: Where d represents the distance between the image block in the reference frame and the matching block in the input frame; I represents the matching block; I(i, j, c) represents the pixel value of the matching block at coordinates (i, j, c) in the input frame; I′(i, j, c) represents the pixel value of the image block at coordinates (i, j, c) in the reference frame; and m represents the side length of the image block.

4. The 3D noise reduction method based on guided filtering according to claim 1, characterized in that: The method of extracting the image blocks in the R, G, and B channels of each pixel as target image blocks and extracting the corresponding guide image blocks in the reference frame for matching includes: For each pixel, extract its image blocks in R, G, and B channels as the target image blocks; In the reference frame, the position matching the target image block is extracted, and the guide image blocks of the R, G, and B channels are extracted respectively, and it is ensured that the extracted target image block and the guide image block have the same size.

5. The 3D noise reduction method based on guided filtering according to claim 1, characterized in that: The filtering process of the target image block and the guide image block in the reference frame by using the guided filter to obtain the output image block after noise reduction comprises: Calculating the coefficients in the guided filter linear model based on the relationship between the guided image block and the target image block; Based on the obtained coefficients, the pixel value of the center point of the output image block is calculated, and for each pixel, the output pixel values ​​of its R, G, and B channels are calculated respectively; In the filtering process, the non-local mean idea is used to set weights to balance the contributions of different guide image blocks; Traverse all pixels in the image, perform filtering on the target image block and the guide image block respectively, and finally generate a complete output image block after noise reduction.

6. The 3D noise reduction method based on guided filtering according to claim 5, characterized in that: The formula for calculating the coefficients in the guided filtering linear model is: Where a and b are coefficients; I′ rl represents the guidance image block; I′ rl (i, j) represents the pixel value of the pixel at coordinate (i, j) in the guide image block; I r Represents the target image block; I r (i, j) represents the pixel value of the pixel at coordinate (i, j) in the target image block; s represents the distance from the center point to the edge of the image block; ε represents the regularization parameter that limits the size of a; m represents the side length of the image block; n represents the matching range; Represents weight.

7. The 3D noise reduction method based on guided filtering according to claim 5, characterized in that: The formula for calculating the pixel value of the center point of the output image block is: Where, Represents the pixel value of the center point of the output image block; a and b both represent coefficients; represents the weight; I′ rl Represents the guidance image patch.

8. The 3D noise reduction method based on guided filtering according to claim 5, characterized in that: In the filtering process, the non-local mean concept is used to set weights to balance the contributions of different guide image blocks, including: Multiple guide image patches are used instead of a single guide image patch, and the translation similarity within the image is exploited to improve the noise reduction effect in smooth areas. By setting a weight function to balance the contributions of different guide image blocks, the noise reduction effect is improved by combining the structural similarity and pixel value similarity of the image blocks; Based on computing resource limitations, the weight function is replaced with a monotonically decreasing function, and the weight calculation is optimized through a lookup table.

Images