CN111784605A - Image denoising method based on region guidance, computer device and computer readable storage medium - Google Patents
Image denoising method based on region guidance, computer device and computer readable storage medium Download PDFInfo
- Publication number
- CN111784605A CN111784605A CN202010611911.3A CN202010611911A CN111784605A CN 111784605 A CN111784605 A CN 111784605A CN 202010611911 A CN202010611911 A CN 202010611911A CN 111784605 A CN111784605 A CN 111784605A
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- region
- pixel
- frequency information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 81
- 230000009467 reduction Effects 0.000 claims abstract description 52
- 238000004364 calculation method Methods 0.000 claims abstract description 41
- 238000001914 filtration Methods 0.000 claims abstract description 40
- 230000004927 fusion Effects 0.000 claims abstract description 25
- 230000008569 process Effects 0.000 claims abstract description 16
- 238000007500 overflow downdraw method Methods 0.000 claims abstract description 4
- 238000004590 computer program Methods 0.000 claims description 23
- 238000001514 detection method Methods 0.000 claims description 23
- 230000011218 segmentation Effects 0.000 claims description 13
- 230000000873 masking effect Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 9
- 230000003044 adaptive effect Effects 0.000 description 8
- 230000006870 function Effects 0.000 description 7
- 230000002093 peripheral effect Effects 0.000 description 6
- 230000002829 reductive effect Effects 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 4
- 239000003086 colorant Substances 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 238000009825 accumulation Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 229910044991 metal oxide Inorganic materials 0.000 description 1
- 150000004706 metal oxides Chemical class 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000011895 specific detection Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The invention provides an image noise reduction method based on regional guidance, a computer device and a computer readable storage medium, wherein the method comprises the steps of obtaining an initial image and constructing an image pyramid; carrying out non-local mean filtering on the low-frequency information of each layer of image, and acquiring a neighborhood similarity map of the layer of image by using a similar block search result in the non-local mean filtering process; detecting an edge region, a texture region and a flat region in the image by using the neighborhood similarity map of each layer of image and the high-frequency information of the high-level image, applying the low-frequency information of at least one layer of image and the high-frequency information of at least one layer of image to the edge region, the texture region and the flat region, performing fusion calculation on the gray value of each pixel by using a corresponding fusion method, and outputting the image after noise reduction. The invention also provides a computer device and a computer readable storage medium for realizing the method. The invention can reduce the calculated amount of image noise reduction, improve the self-adaptability of noise reduction and have better noise reduction effect.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to an image denoising method based on region guidance, a computer device for realizing the method and a computer readable storage medium.
Background
Many existing intelligent electronic devices have an image capturing function, for example, a smartphone, a tablet computer, a vehicle data recorder, and the like are provided with an image capturing device, and the image capturing device is usually provided with a CMOS or CCD image sensor to obtain an image. Generally, an image includes a large number of pixels, and color information of each pixel may be represented by RGB values or YUV values.
For example, CMOS and CCD image sensors commonly used at present generally adopt a BAYER arrangement format, and color information of each pixel is usually RGB values, but the RGB values are not information of three primary colors, and the image has color distortion, so that it is necessary to perform "demosaic" processing on each pixel to obtain RGB three primary color information to restore the original color of the image.
With the increase of image resolution, the reduction of the photosensitive quantity of a single pixel and the wider application of low-illumination scenes, the image noise output by the image sensor is greatly increased. In the process of converting the RAW image into the RGB image, the demosaicing operation needs to refer to all image pixels in a certain area to obtain three primary colors of RGB of one pixel, and noise of each color component is diffused mutually in a large range, so that a large block (from several pixels to hundreds of pixels) of color spots appear in the final image, and the appearance of human eyes is seriously affected. Therefore, it is generally necessary to perform noise reduction processing on an image output from the image sensor.
Currently, the most representative image noise reduction methods are noise reduction methods implemented by Non-Local filters, such as Non-Local mean filtering (NLM) and three-dimensional Block Matching (Block Matching 3D, BM3D), which all use self-similarity of images and obtain a noise-reduced result by weighted average of similar pixel blocks. With the continuous research on noise reduction methods, the idea of combining region detection with noise reduction methods of non-local filters, that is, performing region division on an image and setting corresponding noise reduction parameters or strategies according to region information is widely accepted. In fact, the noise reduction method can obtain a better noise reduction effect and has great benefits for subsequent image analysis.
The non-local mean filtering method is a noise reduction method based on image local similarity, on the basis of a neighborhood average noise reduction method, a similarity weight coefficient of the non-local mean filtering method is determined by the similarity between a current pixel to be noise reduced and an image slice taking other pixels in a neighborhood as centers, the calculation of the weight has no substantive relation with the space positions of two pixels and is only related to the similarity of the two image slices, and therefore false information can be well prevented from being introduced. Since the noise of the image can be equivalent to additive white gaussian noise, the weighted average of the similar pixels can better remove the noise of the image. The NLM method has the characteristics of simple algorithm, superior performance, and easy improvement and expansion, and thus becomes one of the mainstream methods for reducing noise of current images.
The basic principle of the NLM method is as follows: taking a neighborhood window by taking a pixel to be denoised as a center, wherein the window size is N × N (N is generally 3, 5, 7, 9, and the like), taking an image area with a certain size around the pixel as a search area, marking the image area as M × M (the whole image can be selected, but the calculation amount is too large, and generally M is less than 41), searching for similar image blocks in the search area, and calculating the gaussian weighted euclidean distance between the image blocks by using the following formula:
wherein, Ga is a Gaussian kernel with standard deviation a, u (Ni) and u (Nj) are respectively corresponding pixels in the image block of the central window and the image block of the search area. Then, the noise-reduced pixel estimation value is calculated using the following formula:
wherein w (i, j) ═ exp (-d (i, j)/h2) In order to be the weight coefficient,
for the normalization factor, Ω (i) represents the search area of the center pixel i, and h is a similarity weight parameter, which determines the balance after image noise reduction.
Existing noise reduction methods based on regional information and non-local filters can be divided into two main categories: the first type is that gradient information is utilized to carry out edge detection on an image, the image is divided into a detail area and a flat area, and then different filtering parameters are selected according to the detail information; the second type is to divide the image into multiple regions by using the structural statistical information, and select different filters for different regions, including median filtering, bilateral filtering, non-local mean filtering, BM3D, and the like.
However, because the image containing noise in a real scene has complex scene information and large noise type and intensity variation, and the existing noise reduction methods based on region detection have insufficient adaptability, i.e. the noise reduction methods cannot effectively balance the relationship between the noise reduction effect and the noise reduction cost, specifically, on one hand, in the existing noise reduction methods, the region detection module only depends on the analysis result of a single size of the image, the detection accuracy is poor, and the selection of part of parameters has more interference of human factors, which causes the final detection result to have insufficient adaptability and affect the subsequent noise reduction effect, on the other hand, in the existing noise reduction methods, the region detection module is realized by a complex calculation method, the calculation amount is large, which causes the image noise reduction processing time to be too long, if a hardware circuit is used to realize the noise reduction calculation, the hardware circuit area is too large, the development difficulty is large.
Disclosure of Invention
The invention mainly aims to provide an image denoising method based on region guidance, which has good denoising effect and small calculation amount.
Another object of the present invention is to provide a computer device for implementing the above image denoising method based on region guidance.
It is still another object of the present invention to provide a computer-readable storage medium for implementing the above region-based guided image denoising method.
In order to achieve the main purpose of the invention, the image denoising method based on the region guidance comprises the steps of obtaining an initial image, constructing an image pyramid by applying the initial image, wherein each layer of image of the image pyramid comprises low-frequency information and high-frequency information of the layer of image; carrying out non-local mean filtering on the low-frequency information of each layer of image, and acquiring a neighborhood similarity map of the layer of image by using a similar block search result in the non-local mean filtering process; detecting an edge region, a texture region and a flat region in the image by using the neighborhood similarity map of each layer of image and the high-frequency information of the high-level image, applying the low-frequency information of at least one layer of image and the high-frequency information of at least one layer of image to the edge region, the texture region and the flat region, performing fusion calculation on the gray value of each pixel by using a corresponding fusion method, and outputting the image after noise reduction.
According to the scheme, the image pyramid is constructed and the non-local mean filtering method is applied, so that the texture information embodied by the similarity search result of the images of different layers in the non-local mean filtering can be deeply mined according to the low-frequency information and the high-frequency information of the images of different layers, and therefore different regions can be accurately judged, and a more optimal and more efficient region detection result can be obtained. And according to the result, pixel noise reduction algorithms of different regions are specified, namely different noise reduction methods are used for calculating the gray value of the noise-reduced pixel aiming at different regions, so that the noise reduction effect of the image is more reasonable.
In addition, the operation of region detection is realized by directly applying the search result in the non-local mean filtering process, and the amount of noise reduction calculation is not obviously increased, so that the amount of noise reduction calculation of the image is small, and the efficiency of the noise reduction calculation of the image is improved.
Preferably, the detecting the edge region, the texture region and the flat region in the image by using the neighborhood similarity map of each layer image and the high-frequency information of the high-level image comprises: determining pixels of the edge region according to the neighborhood similarity value of each pixel in the neighborhood similarity map of the low-level image; and determining the pixels of the texture region according to the neighborhood similarity value of each pixel in the neighborhood similarity map of the high-level image and the high-frequency information of the high-level image.
Therefore, the gray value difference between the pixels of the edge region and the peripheral pixels is large, so that the edge region in the image can be accurately detected through the neighborhood similarity value of each pixel in the neighborhood similarity map of the lower-layer image, and the detection of the texture region also comprehensively considers the neighborhood similarity map of the higher-layer image and the high-frequency information of the higher-layer image, so that the detection of the texture region is more accurate.
Further, determining the pixels of the edge region according to the neighborhood similarity values of the pixels in the neighborhood similarity map of the low-level image comprises: and performing threshold segmentation on the neighborhood similarity value of each pixel in the neighborhood similarity images of more than two layers of low-level images, merging the results of the threshold segmentation, and determining the pixels of the edge region according to the merged results.
Therefore, when the pixels of the edge area are determined, the threshold segmentation is realized according to the threshold segmentation result of the two-layer image, and the edge area can be more accurately detected.
Further, determining the pixels of the texture region according to the neighborhood similarity value of each pixel in the neighborhood similarity map of the high-level image and the high-frequency information of the high-level image comprises: and after the detection result of the edge region is subjected to masking operation, determining pixels, of which the neighborhood similarity value is not zero and the high-frequency information of the high-level image is also not zero, in the neighborhood similarity map of the high-level image as pixels of the texture region.
Therefore, the edge area of the image is determined, and then the texture area and the flat area are detected after the pixels of the edge area are shielded, so that the interference of the pixels of the edge area on the detection of the texture area and the flat area can be avoided, and the accuracy of the detection of the texture area and the flat area is improved.
Preferably, the obtaining of the neighborhood similarity map of the layer image by using the result of searching the similar blocks in the non-local mean filtering process includes: in the searching process of the similar block of the non-local mean filtering, the number of the matching windows and the similar windows corresponding to each pixel is calculated, and the number of the matching windows and the similar windows corresponding to the pixel is used as a neighborhood similarity value of the pixel in the neighborhood similarity graph.
Therefore, the neighborhood similarity graph of the image is formed by the neighborhood similarity value of each pixel, and the neighborhood similarity value of each pixel is determined when the neighborhood similarity window is detected in the non-local mean filtering process, so that the calculation amount of the neighborhood similarity graph is small.
Further, the fusion calculation of the gray values of the pixels in the edge area by using the low-frequency information of the at least one layer of image and the high-frequency information of the at least one layer of image comprises: and accumulating the low-frequency information of the low-layer image and the high-frequency information of at least two layers of high-layer images to obtain the gray value of the pixel of the edge area.
Therefore, the gray value calculation of the pixels in the edge area is realized in a direct fusion mode, namely, the low-frequency information and the high-frequency information of the multilayer image are directly accumulated, the calculation amount is small, and the high-frequency information in the edge area can be reserved.
Further, the fusion calculation of the gray value of the pixel of the flat area by using the low-frequency information of the at least one layer of image and the high-frequency information of the at least one layer of image comprises: and accumulating the low-frequency information of the low-layer image and the high-frequency information of at least two layers of high-layer images, and performing median filtering to obtain the gray value of the pixel in the flat area.
Therefore, for the pixels in the flat area, the gray values of the pixels in the flat area are more balanced in a median filtering mode, the pixel characteristics of the flat area are better met, and the image quality after filtering is improved.
The method comprises the following steps that when a texture region in an image is detected by applying a neighborhood similarity graph of each layer of image and high-frequency information of a high-level image, a similarity coefficient of each pixel of the texture region is calculated; the fusion calculation of the gray value of the pixel of the texture region by applying the low-frequency information of the at least one layer of image and the high-frequency information of the at least one layer of image comprises the following steps: and determining weighting coefficients of the multi-layer images according to the similarity coefficients, and performing weighted fusion calculation on the low-frequency information of the at least one layer of images and the high-frequency information of the at least one layer of images by using the weighting coefficients to obtain the gray value of the pixel of the texture region.
Therefore, the pixels in the texture region are subjected to fusion calculation by using the adaptive weighting coefficients, so that the pixels with different similarity coefficients adopt different weighting coefficients for the images in different layers in the fusion calculation process, the gray value calculation of the texture region is closer to the actual gray value, the texture region is clearer in the filtered image, the texture display is closer to the actual state, and the filtering effect is more rational.
In order to achieve the above another object, the present invention provides a computer device comprising a processor and a memory, wherein the memory stores a computer program, and the computer program is executed by the processor to implement the steps of the image noise reduction method based on region guidance.
To achieve the above-mentioned further object, the present invention provides a computer-readable storage medium having a computer program stored thereon, where the computer program is executed by a processor to implement the steps of the above-mentioned image noise reduction method based on region guidance.
Drawings
FIG. 1 is a flowchart of an embodiment of an image denoising method based on region guidance according to the present invention.
Fig. 2 is a schematic diagram of color information of each pixel of an initial image.
FIG. 3 is a flowchart illustrating region detection in an embodiment of a method for image denoising based on region guidance according to the present invention.
FIG. 4 is a schematic diagram of a region detection result in an embodiment of the image denoising method based on region guidance.
FIG. 5 is a flowchart illustrating fusion calculation of gray-level values of pixels according to an embodiment of the image denoising method based on region guidance.
The invention is further explained with reference to the drawings and the embodiments.
Detailed Description
The image noise reduction method based on the region guidance is applied to the intelligent electronic equipment, preferably, the intelligent electronic equipment is provided with a camera device such as a camera and the like, the camera device is provided with an image sensor such as a CMOS (complementary metal oxide semiconductor), a CCD (charge coupled device) and the like, the intelligent electronic equipment acquires an initial image by using the camera device, and the method is a processing method for reducing the noise of the initial image acquired by the image sensor. Preferably, the intelligent electronic device is provided with a processor and a memory, the memory having a computer program stored thereon, and the processor implements the image noise reduction method based on the region guidance by executing the computer program.
The embodiment of the image noise reduction method based on the region guidance comprises the following steps:
the embodiment mainly relates to a method for reducing noise of an initial image acquired by an image sensor, in particular to a method for reducing noise of an initial image acquired by an image sensor, wherein after the initial image containing noise is acquired, the initial image is firstly applied to construct an image pyramid, the image pyramid comprises a plurality of layers of images, and each layer of image comprises low-frequency information and high-frequency information; then, carrying out noise reduction processing on the low-frequency information of each layer of image of the image pyramid by using non-local mean filtering, and acquiring regional statistical information of a similarity detection result in the non-local mean filtering process of the low-frequency information of each layer of image in the noise reduction processing process; then, dividing the image into an edge region, a texture region and a flat region by using the region statistical information; and finally, calculating the gray value of the filtered pixel by using the region statistical information and adopting different fusion calculation methods for different regions respectively to obtain an output image.
Referring to fig. 1, step S1 is first performed to acquire an initial image. The initial image of the present embodiment is an image output by an image sensor such as a CMOS or a CCD, and usually, the color information of the initial image is RGB information, that is, the format of the image is a BAYER image format. BAYER image format as shown in fig. 2, the initial image has a large number of pixels, each having color information, for example, the color information of the pixels in the first row is color information of red R or green Gr, the red R pixels and the green Gr pixels are arranged at intervals, the color information of the pixels in the second row is color information of green Gb or blue B, and the green Gb pixels and the blue B pixels are arranged at intervals. The color information of each pixel is a chrominance value, which is typically a binary number of 8 bits to 16 bits. In each color channel, the chrominance value of a pixel is the gray value of the pixel, and the calculation of the embodiment is performed on the gray value of the pixel.
Because the colors of four adjacent pixels in the initial image are different, when filtering images of multiple colors, the gray values of the pixels with different chromaticities are easy to interfere with each other, and the image noise reduction effect is affected. Therefore, step S1 also needs to extract images of each color channel in the initial image, each color channel including pixels of only one color. For example, pixels of all red R pixels in the initial image are extracted to form an image of one red R pixel, pixels of all green Gr and Gb in the initial image are extracted to form an image of one green G pixel, and pixels of all blue B pixels in the initial image are extracted to form an image of one blue B pixel. Of course, the relative positional relationship of each pixel is not changed in each extracted color channel image.
The subsequent steps S2 to S6 are performed once for each color channel, that is, for the images of three color channels, the operations of steps S2 to S6 are performed, and finally, the images after filtering of the three color channels are subjected to inverse interpolation calculation in accordance with the arrangement of the respective pixels shown in fig. 2 to form an output image.
For example, for the image of the red channel, step S2 is performed to construct an image pyramid. In this embodiment, each layer of image has the same pixel as the initial image of the channel, for example, after the images of the color channels are extracted from the initial image, the pixel size of the red channel image is 1028 × 720, and the pixel size of each layer of image in the constructed image pyramid corresponding to the red channel is 1028 × 720.
Specifically, based on the initial image of the red channel, the gaussian low-pass filters with different sizes are used to perform convolution operation with the initial image of the color channel, so as to obtain multiple layers of gaussian images to form a gaussian pyramid. For example, the size of a gaussian low-pass filter corresponding to the first-layer image of the gaussian pyramid is set to be 7 × 7, and the first-layer gaussian image is obtained by performing convolution operation on the gaussian low-pass filter and the initial image of the color channel. Setting the size of a Gaussian low-pass filter corresponding to the second layer image of the Gaussian pyramid to be 9 multiplied by 9, performing convolution operation on the Gaussian low-pass filter and the initial image of the color channel to obtain the second layer Gaussian image, and so on. Preferably, the size of the gaussian low-pass filter used is gradually increased as the number of layers is gradually increased. Therefore, each layer of image of the gaussian pyramid includes low frequency information and a small portion of noise that are characteristic of most features of the original image.
After the multilayer image of the gaussian pyramid is obtained, the multilayer image of the laplacian pyramid is calculated by applying the multilayer image of the gaussian pyramid, and specifically, the value of each pixel of each layer of the laplacian image of the laplacian pyramid is the value obtained by subtracting the gray value of the pixel of the current layer of the gaussian image from the gray value of the previous layer of the gaussian image. For example, the gray values of the pixels of the first layer of gaussian image and the pixels of the second layer of gaussian image are subtracted to obtain the gray values of the pixels of the first layer of laplacian image, and so on. For the highest-level laplacian image, the highest-level laplacian image is directly used as the highest-level laplacian image. Therefore, the images of the layers of the laplacian pyramid contain much noise and high-frequency information of partial contours.
As can be seen, the image pyramid actually includes a gaussian pyramid and a laplacian pyramid, and the number of layers of the gaussian pyramid and the laplacian pyramid is the same and corresponds to each other one by one, where information of each layer of the image of the gaussian pyramid is low-frequency information, and information of each layer of the image of the laplacian pyramid is high-frequency information.
Then, step S3 is executed to perform non-local mean filtering on the low-frequency information of each layer of image in the image pyramid, specifically, perform non-local mean filtering on the information of each layer of image using the gaussian pyramid of the image pyramid. For example, for any pixel i in each layer of image in the gaussian pyramid, assuming that the gray value of the pixel is f (i), a matching window p (i) with a window size of s and a search area with a window size of t (t > s) are respectively extracted with the pixel as the center, and an image block p (j) with higher similarity to the matching window p (i) is detected in the search area in a traversing manner, wherein the window size of the image block p (j) is the same as the size of the matching window p (i).
For example, the sum of the gray values of a plurality of pixels in the matching window p (i) is calculated, and the sum of the gray values of a plurality of pixels in the window to be matched is calculated, and if the difference between the sum of the gray values of the plurality of pixels in the window to be matched and the sum of the gray values of the plurality of pixels in the current matching window p (i) is smaller than a preset threshold, the similarity between the window to be matched and the matching window p (i) is considered to be high, and the window to be matched is marked as the image block p (j).
Then, a gaussian weighted euclidean distance d (i, j) between the image block p (j) with higher similarity and the current matching window p (i) is calculated, for example, by applying the following formula:
d(i,j)＝G0*||P(i)-P(j)||2(formula 3)
Then, a weighting coefficient w (i, j) of the image block p (j) with higher similarity is calculated, and the specific formula is as follows:
and finally, calculating the gray value of the current pixel after non-local mean filtering, wherein the specific formula is as follows:
in the above formulas 4 and 5, G0H is a parameter for controlling the degree of smoothing, which is a predetermined gaussian function.
Then, step S4 is executed to apply the search result of the non-local mean filtering to obtain the neighborhood similarity map of each layer image. Each layer of image of the image pyramid corresponds to a neighborhood similarity map of the image pyramid, and the value of each pixel in the neighborhood similarity map is a neighborhood similarity value. The purpose of obtaining the neighborhood similarity map of each layer of image is to detect an edge region, a texture region and a flat region in the image, and the edge region in the image mainly comprises image contour information, so that the difference between the gray value of a pixel in the edge region and the gray value of a peripheral pixel is large, and the similarity between the gray value of the pixel in the edge region and the gray value of the peripheral pixel is low; the texture region comprises some detail information and weak texture of the image, and the similarity between the pixel gray value of the texture region and the peripheral pixel gray value is moderate; the similarity between the pixel gray-scale value of the flat area and the peripheral pixel gray-scale value is higher. Based on these characteristics, the present embodiment determines which region each pixel is in through the similarity search of the gray value of each pixel and the gray values of the peripheral pixels, that is, the region information is obtained by analyzing the similarity statistical result of the non-local mean filtering.
In step S4, when performing pixel-by-pixel non-local mean filtering on each layer of image of the gaussian pyramid, the number of image blocks p (j) with higher similarity to the matching window p (i) with the current pixel as the center point in the search window is calculated. Or, setting a difference threshold according to a noise curve value calibrated in advance, counting the number of image blocks in the search area, the difference value of which is smaller than the corresponding difference threshold from the matching window p (i), and taking the number as the neighborhood similarity value cnt (i) of the current pixel. After the neighborhood similarity value cnt (i) of each pixel is determined, a neighborhood similarity map corresponding to the layer of image can be obtained, and the value corresponding to each pixel in the neighborhood similarity map is the neighborhood similarity value cnt (i) of the pixel. Therefore, the value of each pixel in the neighborhood similarity map is independent of the gray value of the pixel, and the neighborhood similarity value cnt (i) represents the similarity degree between the pixel and the surrounding pixels.
Then, step S5 is executed to detect an edge region, a flat region and a texture region in the image according to the neighborhood similarity map. The following describes a specific detection process of each region by taking an image pyramid with three layers of images as an example and referring to fig. 3.
First, an edge region in an image is determined according to a neighborhood similarity map of a lower-layer image. Specifically, the neighborhood similarity map of the first layer image and the neighborhood similarity map of the second layer image are counted, threshold segmentation is performed, and the result of the threshold segmentation is used for detecting the edge region in the image. Specifically, step S21 is executed to obtain a neighborhood similarity map of the first layer image, and step S22 is executed to perform mean value statistics on the neighborhood similarity map of the first layer image, for example, an adaptive threshold th of the first layer image is set, where the adaptive threshold th is obtained by using the following formula:
where num is the total number of pixels of the layer image. As shown in equation 6, the adaptive threshold th of the layer image is the average value of the neighborhood similarity values of all pixels of the layer image. In addition, since the total number of pixels of each layer image is the same in this embodiment, num is the same in calculating the adaptive threshold of each layer image.
Next, step S23 is executed to perform threshold segmentation on the neighborhood similarity map of the first layer image, specifically, to mark the pixels in the neighborhood similarity map whose neighborhood similarity is smaller than the adaptive threshold th as the pixels in the edge region of the layer.
Correspondingly, the same steps are also performed for the second layer image, that is, step S24 is performed first to obtain the neighborhood similarity map of the second layer image, step S25 is then performed to perform mean value statistics on the neighborhood similarity values of the pixels in the neighborhood similarity map of the second layer image, that is, adaptive threshold calculation is performed using equation 6, step S26 is performed to perform threshold segmentation on the second layer image, and the pixels in the neighborhood similarity map whose neighborhood similarity values are smaller than the adaptive threshold th are marked as the pixels in the edge region of the layer.
Finally, step S27 is executed to combine the threshold segmentation result of the first layer image and the threshold segmentation result of the second layer image, that is, to determine the pixels in the edge region marked as the layer in the two-layer image as the pixels in the edge region of the image. Because the pixels of the first layer image and the pixels of the second layer image are in one-to-one correspondence, whether a certain pixel is marked as an edge area in the first layer image or the second layer image can be judged according to the one-to-one correspondence relationship of the pixels in the two layer images, and if so, the pixel is determined to be the pixel of the edge area.
After determining the pixels of the edge region, the pixels of the texture region are detected. Specifically, step S28 is executed to obtain a neighborhood similarity map of the higher-level image, that is, to obtain a neighborhood similarity map of the third-level image, step S29 is executed to obtain high-frequency information of the third-level image, for example, to obtain data of a laplacian image of the third-level image, and omit information of an edge region, and step S30 is executed to further determine and normalize pixels in a non-edge region.
For example, the neighborhood similarity map of the third-level image is masked, that is, pixels marked as edge regions in step S27 in the neighborhood similarity map of the third-level image are masked, for example, neighborhood similarity values of the pixels of the edge regions are directly set to 0. Then, the pixels in the neighborhood similarity map of the third layer image whose neighborhood similarity value and high frequency information of the third layer image are not 0 are marked as pixels of the texture region, that is, step S31 is performed. After determining the pixels of the texture region, the remaining pixels are determined as the pixels of the flat region, i.e., step S32 is performed. Up to this point, individual pixels of an image are divided into pixels of different areas.
While step S31 is executed, step S30 is also executed to perform a normalization calculation on the pixels of the texture region, the normalization calculation using the following formula:
wherein cnt (i) is the neighborhood similarity value of the current pixel, and max (cnt), min (cnt) are the maximum value and the minimum value of the neighborhood similarity value of each pixel in the third-layer neighborhood similarity map, respectively. It can be seen that, according to the above formula, the normalized value of the pixel in the texture region is between 0 and 1, and the normalized value of the pixel in the flat region is 0. Further, the value of the pixel of the edge area subjected to the normalization calculation may be set to 1. And, for the pixels of the texture region, each pixel has its own similarity coefficient, i.e., the calculation result g of equation 7.
In fig. 4, (a) in fig. 4 is an initial image, and an image obtained after region detection is as shown in (b) in fig. 4, and after region detection, the image is merely divided into three types of regions, but fusion weighting calculation is not performed on the gray-scale values of the pixels of each region. Therefore, step S6 needs to be executed to perform fusion calculation on the gray-level values of the pixels in different regions by applying a corresponding fusion calculation method, so as to obtain the filtered gray-level value of each pixel.
Referring to fig. 5, step S51 is first executed to obtain high frequency information and low frequency information of each layer image, that is, to obtain a value of each pixel in each layer image of the gaussian pyramid and the laplacian pyramid. Then, step S52 is executed to determine whether the current pixel is a pixel in the edge region, if so, step S59 is executed to calculate the gray level of the pixel by using a direct fusion method, specifically, the gray level of the pixel is calculated by applying the low frequency information of the at least one layer of image and the high frequency information of the at least one layer of image. For example, the pixel gray value of the first layer image of the gaussian pyramid and the values of the pixels of the second layer image and the third layer image of the laplacian pyramid are accumulated, and the accumulated result is used as the fused gray value, which is the filtered gray value of the pixel. The direct fusion mode can ensure that the characteristics of the pixels in the edge area are kept as much as possible, so that the gray value of the pixels in the edge area is more real, and the high-frequency detail characteristics in the image are kept.
If the current pixel is not the pixel of the edge region, step S53 is executed to determine whether the current pixel is the pixel of the flat region, and if so, step S60 is executed to calculate the gray value of the pixel by using a median filtering fusion algorithm. Specifically, the pixel gray value of the first layer image using the gaussian pyramid and the values of the pixels of the second layer image and the third layer image using the laplacian pyramid are accumulated, the result of performing the same accumulation operation on a plurality of pixels around the current pixel is obtained, and median filtering is performed by using the accumulation result of the plurality of pixels around the current pixel, for example, an average value of the accumulation values of the plurality of pixels around the current pixel is calculated, and the average value is used as the gray value of the flat area. In this way, the gray value of the pixel in the flat area is obtained through median filtering, so that the image in the flat area can be smoother and better conforms to the characteristics of the flat area.
If the current pixel is not a pixel of the flat region, step S54 is executed to determine whether the current pixel is a pixel of the texture region, if so, step S55 is executed to determine a weighting coefficient of the multi-layer image according to the similarity coefficient of the current pixel, for example, determine weighting coefficients of the first layer of gaussian image, the second layer of laplacian image and the third layer of laplacian image according to the similarity coefficient g, and perform weighted fusion calculation on the values of the pixel of the three-layer image according to the weighting coefficients of the three-layer image to obtain the gray value of the pixel of the texture region.
For example, if the weighting coefficient of the first layer laplacian image is g, the weighting coefficient of the second layer laplacian image is 1-g/2, and the weighting coefficient of the third layer laplacian image is 1-g/2, the gray value of the pixel of the first layer laplacian image is multiplied by the weighting coefficient of the first layer laplacian image, the value of the pixel of the second layer laplacian image is multiplied by the weighting coefficient of the second layer laplacian image, and the value of the pixel of the third layer laplacian image is multiplied by the weighting coefficient of the third layer laplacian image, and the three values are added to obtain the gray value of the pixel.
Then, step S56 is executed to determine whether the current pixel is the last pixel in the image, if not, step S58 is executed to obtain the next pixel, and step S52 is executed again, otherwise, the current pixel is the last pixel in the current image, the noise-reduced image is output, and the gray value of each pixel in the noise-reduced image is the gray value calculated according to the weighted fusion. Because the pixels in the texture region are not fused by using a fixed weighting coefficient, but the weighting coefficient is adaptively adjusted according to the similarity coefficient of the pixels, the gray value of the pixels in the texture region is more flexibly calculated, and the texture features of the image are reserved.
Since steps S2 to S6 are performed for a single color channel, steps S2 to S6 are required to be performed for three color channels, and after step S6 is performed, the gray level of each pixel of each image of the three color channels can be obtained. However, since the original image is in the BAYER format, step S7 requires that the positions of the pixels in the original image be restored in the reverse process of extracting each color channel image in step S1 according to the format of the original image, so as to form one output image, that is, to perform the inverse interpolation calculation, which is the output noise reduction image.
Because the region detection of the invention only depends on the similarity detection result of the non-local mean filtering in the noise reduction algorithm framework, the invention not only fully utilizes the characteristics of the self framework and improves the precision of the region detection, but also does not introduce additional operand, and especially compared with the traditional BM3D noise reduction method, the invention has lower computational complexity and hardware realization cost and is easy to realize. In addition, because each region is detected based on the image pyramid and the fusion calculation is carried out in a targeted manner aiming at the pixels in different regions, the situation that the adaptability to the scene is not enough can be effectively avoided, for example, the problems that the texture region in the image is excessively smooth or the noise reduction effect of the flat region is not ideal and the like are avoided. In addition, most threshold parameters of the method are set in a self-adaptive mode, for example, parameters such as a self-adaptive threshold th and a similarity coefficient g used for threshold segmentation are not fixed values, and therefore the method is high in adaptability for complex scenes with more noise.
Of course, the above-mentioned embodiments are only preferred embodiments of the present invention, and in practical applications, the following variations are possible: when the image pyramid is constructed, the construction is not limited to the method, for example, a DOG image pyramid can be constructed, or a wavelet and a curved surface wave can be used for substitution; or, in the process of carrying out region detection, the self-adaptive threshold selection mode of the images of different layers can be replaced by multiplying a fixed threshold by a fixed coefficient; alternatively, the weighting factor of the gray-level value of the texel area may be replaced by a larger fixed value.
In addition, the image noise reduction method of the invention can be used in a series of image video processing devices including vehicle-mounted camera devices, network camera devices, motion cameras and the like, and various parts in the method can be properly adjusted or deleted according to actual requirements.
The embodiment of the computer device comprises:
the computer device of this embodiment may be an intelligent electronic device, and the computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the processor implements the steps of the image noise reduction method based on the region guidance. Of course, the intelligent electronic device further includes a camera device for acquiring an initial image.
For example, a computer program may be partitioned into one or more modules that are stored in a memory and executed by a processor to implement the modules of the present invention. One or more of the modules may be a series of computer program instruction segments capable of performing certain functions, which are used to describe the execution of the computer program in the terminal device.
The Processor may be a Central Processing Unit (CPU), or may be other general-purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, a discrete Gate or transistor logic device, a discrete hardware component, or the like. The general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like, the processor being the control center of the terminal device and connecting the various parts of the entire terminal device using various interfaces and lines.
The memory may be used to store computer programs and/or modules, and the processor may implement various functions of the terminal device by running or executing the computer programs and/or modules stored in the memory and invoking data stored in the memory. The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data (such as audio data, a phonebook, etc.) created according to the use of the cellular phone, and the like. In addition, the memory may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), at least one magnetic disk storage device, a Flash memory device, or other volatile solid state storage device.
Computer-readable storage medium embodiments:
the computer program stored in the computer device may be stored in a computer-readable storage medium if it is implemented in the form of a software functional unit and sold or used as a separate product. Based on such understanding, all or part of the flow in the method according to the above embodiments may be implemented by a computer program, which may be stored in a computer readable storage medium, and when the computer program is executed by a processor, the computer program may implement the steps of the image denoising method based on the region guidance.
Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying computer program code, recording medium, U.S. disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution media, and the like. It should be noted that the computer readable medium may contain other components which may be suitably increased or decreased as required by legislation and patent practice in jurisdictions, for example, in some jurisdictions, in accordance with legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunications signals.
Finally, it should be emphasized that the present invention is not limited to the above-mentioned embodiments, such as the change of the way of obtaining the image pyramid, or the change of the way of calculating the pixel fusion of each region, and these changes should also be included in the protection scope of the claims of the present invention.
Claims (10)
1. An image denoising method based on region guidance comprises the following steps:
acquiring an initial image;
the method is characterized in that:
constructing an image pyramid by applying the initial image, wherein each layer of image of the image pyramid comprises low-frequency information and high-frequency information of the layer of image;
carrying out non-local mean filtering on the low-frequency information of each layer of image, and acquiring a neighborhood similarity map of the layer of image by using a similar block search result in the non-local mean filtering process;
detecting an edge region, a texture region and a flat region in the image by applying the neighborhood similarity map of each layer of image and the high-frequency information of the high-level image, applying the low-frequency information of at least one layer of image and the high-frequency information of at least one layer of image to the edge region, the texture region and the flat region, performing fusion calculation on the gray value of each pixel by using a corresponding fusion method, and outputting the image after noise reduction.
2. The region-based guided image denoising method according to claim 1, wherein:
the method for detecting the edge region, the texture region and the flat region in the image by applying the neighborhood similarity map of each layer of image and the high-frequency information of the high-level image comprises the following steps:
determining pixels of the edge region according to the neighborhood similarity value of each pixel in the neighborhood similarity map of the low-level image;
and determining the pixels of the texture region according to the neighborhood similarity value of each pixel in the neighborhood similarity map of the high-level image and the high-frequency information of the high-level image.
3. The region-based guided image denoising method according to claim 2, wherein:
determining the pixels of the edge region according to the neighborhood similarity values of the pixels in the neighborhood similarity map of the low-level image comprises:
and performing threshold segmentation on the neighborhood similarity value of each pixel in the neighborhood similarity images of more than two layers of low-level images, merging the results of the threshold segmentation, and determining the pixels of the edge region according to the merged results.
4. The region-based guided image denoising method according to claim 2, wherein:
determining the pixels of the texture region according to the neighborhood similarity value of each pixel in the neighborhood similarity map of the high-level image and the high-frequency information of the high-level image comprises the following steps:
and after the detection result of the edge region is subjected to masking operation, determining pixels, of which the neighborhood similarity value is not zero and the high-frequency information of the high-level image is also not zero, in the neighborhood similarity map of the high-level image as pixels of the texture region.
5. The region-based guidance image denoising method according to any one of claims 1 to 4, wherein:
the method for acquiring the neighborhood similarity map of the layer of image by using the similar block search result in the non-local mean filtering process comprises the following steps:
in the searching process of the similar block of the non-local mean filtering, the number of the matching windows and the similar windows corresponding to each pixel is calculated, and the number of the matching windows and the similar windows corresponding to the pixel is used as a neighborhood similarity value of the pixel in the neighborhood similarity graph.
6. The region-based guided image noise reduction method according to any one of claims 1 to 4, characterized in that:
the fusion calculation of the gray value of the pixel of the edge area by applying the low-frequency information of at least one layer of image and the high-frequency information of at least one layer of image comprises the following steps:
and accumulating the low-frequency information of the low-layer image and the high-frequency information of at least two layers of high-layer images to obtain the gray value of the pixel of the edge area.
7. The region-based guided image noise reduction method according to any one of claims 1 to 4, characterized in that:
the fusion calculation of the gray value of the pixel of the flat area by applying the low-frequency information of at least one layer of image and the high-frequency information of at least one layer of image comprises the following steps:
and accumulating the low-frequency information of the low-layer image and the high-frequency information of at least two layers of high-layer images, and performing median filtering to obtain the gray value of the pixel of the flat area.
8. The region-based guidance image denoising method according to any one of claims 1 to 4, wherein:
when the neighborhood similarity map of each layer of image and the high-frequency information of the high-level image are used for detecting the texture region in the image, calculating the similarity coefficient of each pixel of the texture region;
the fusion calculation of the gray value of the pixel of the texture region by applying the low-frequency information of at least one layer of image and the high-frequency information of at least one layer of image comprises the following steps:
and determining weighting coefficients of the multi-layer images according to the similarity coefficients, and performing weighted fusion calculation on the low-frequency information of at least one layer of images and the high-frequency information of at least one layer of images by using the weighting coefficients to obtain the gray value of the pixel of the texture region.
9. Computer arrangement, characterized in that it comprises a processor and a memory, said memory storing a computer program that, when executed by the processor, performs the steps of the region-based guided image noise reduction method according to any of claims 1 to 8.
10. A computer-readable storage medium having stored thereon a computer program, characterized in that: the computer program when executed by a processor implements the steps of the region-based guided image denoising method according to any one of claims 1 through 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010611911.3A CN111784605B (en) | 2020-06-30 | 2020-06-30 | Image noise reduction method based on region guidance, computer device and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010611911.3A CN111784605B (en) | 2020-06-30 | 2020-06-30 | Image noise reduction method based on region guidance, computer device and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111784605A true CN111784605A (en) | 2020-10-16 |
| CN111784605B CN111784605B (en) | 2024-01-26 |
Family
ID=72761285
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010611911.3A Active CN111784605B (en) | 2020-06-30 | 2020-06-30 | Image noise reduction method based on region guidance, computer device and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111784605B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112446838A (en) * | 2020-11-24 | 2021-03-05 | 海南大学 | Image noise detection method and device based on local statistical information |
| CN112862717A (en) * | 2021-02-10 | 2021-05-28 | 山东英信计算机技术有限公司 | Image denoising and blurring method, system and medium |
| CN112884667A (en) * | 2021-02-04 | 2021-06-01 | 湖南兴芯微电子科技有限公司 | Bayer domain noise reduction method and noise reduction system |
| CN113033574A (en) * | 2021-02-26 | 2021-06-25 | 天津大学 | Image data noise reduction system and method based on FPGA |
| CN113643198A (en) * | 2021-07-22 | 2021-11-12 | 海宁奕斯伟集成电路设计有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN114240941A (en) * | 2022-02-25 | 2022-03-25 | 浙江华诺康科技有限公司 | Endoscope image noise reduction method, device, electronic apparatus, and storage medium |
| CN114509061A (en) * | 2021-12-30 | 2022-05-17 | 重庆特斯联智慧科技股份有限公司 | Method and system for determining robot traveling path based on barrier attributes |
| CN116342891A (en) * | 2023-05-24 | 2023-06-27 | 济南科汛智能科技有限公司 | Structured teaching monitoring data management system suitable for autism children |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160086317A1 (en) * | 2014-09-23 | 2016-03-24 | Intel Corporation | Non-local means image denoising with detail preservation using self-similarity driven blending |
| CN109544478A (en) * | 2018-11-26 | 2019-03-29 | 重庆大学 | A kind of non-local mean CT image denoising method based on singular value decomposition |
| CN109785246A (en) * | 2018-12-11 | 2019-05-21 | 深圳奥比中光科技有限公司 | A kind of noise-reduction method of non-local mean filtering, device and equipment |
| CN111161188A (en) * | 2019-12-30 | 2020-05-15 | 珠海全志科技股份有限公司 | Method for reducing image color noise, computer device and computer readable storage medium |
| CN111260580A (en) * | 2020-01-17 | 2020-06-09 | 珠海全志科技股份有限公司 | Image denoising method based on image pyramid, computer device and computer readable storage medium |
-
2020
- 2020-06-30 CN CN202010611911.3A patent/CN111784605B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160086317A1 (en) * | 2014-09-23 | 2016-03-24 | Intel Corporation | Non-local means image denoising with detail preservation using self-similarity driven blending |
| CN109544478A (en) * | 2018-11-26 | 2019-03-29 | 重庆大学 | A kind of non-local mean CT image denoising method based on singular value decomposition |
| CN109785246A (en) * | 2018-12-11 | 2019-05-21 | 深圳奥比中光科技有限公司 | A kind of noise-reduction method of non-local mean filtering, device and equipment |
| CN111161188A (en) * | 2019-12-30 | 2020-05-15 | 珠海全志科技股份有限公司 | Method for reducing image color noise, computer device and computer readable storage medium |
| CN111260580A (en) * | 2020-01-17 | 2020-06-09 | 珠海全志科技股份有限公司 | Image denoising method based on image pyramid, computer device and computer readable storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 赵婧娟;周祚峰;曹剑中;王华;: "双尺度自适应非局部均值图像降噪算法", 红外与激光工程, no. 1 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112446838A (en) * | 2020-11-24 | 2021-03-05 | 海南大学 | Image noise detection method and device based on local statistical information |
| CN112884667A (en) * | 2021-02-04 | 2021-06-01 | 湖南兴芯微电子科技有限公司 | Bayer domain noise reduction method and noise reduction system |
| CN112884667B (en) * | 2021-02-04 | 2021-10-01 | 湖南兴芯微电子科技有限公司 | Bayer domain noise reduction method and noise reduction system |
| CN112862717A (en) * | 2021-02-10 | 2021-05-28 | 山东英信计算机技术有限公司 | Image denoising and blurring method, system and medium |
| CN113033574A (en) * | 2021-02-26 | 2021-06-25 | 天津大学 | Image data noise reduction system and method based on FPGA |
| CN113643198A (en) * | 2021-07-22 | 2021-11-12 | 海宁奕斯伟集成电路设计有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN114509061A (en) * | 2021-12-30 | 2022-05-17 | 重庆特斯联智慧科技股份有限公司 | Method and system for determining robot traveling path based on barrier attributes |
| CN114240941A (en) * | 2022-02-25 | 2022-03-25 | 浙江华诺康科技有限公司 | Endoscope image noise reduction method, device, electronic apparatus, and storage medium |
| CN114240941B (en) * | 2022-02-25 | 2022-05-31 | 浙江华诺康科技有限公司 | Endoscope image noise reduction method, device, electronic apparatus, and storage medium |
| CN116342891A (en) * | 2023-05-24 | 2023-06-27 | 济南科汛智能科技有限公司 | Structured teaching monitoring data management system suitable for autism children |
| CN116342891B (en) * | 2023-05-24 | 2023-08-15 | 济南科汛智能科技有限公司 | Structured teaching monitoring data management system suitable for autism children |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111784605B (en) | 2024-01-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111784605B (en) | Image noise reduction method based on region guidance, computer device and computer readable storage medium | |
| CN109712102B (en) | Image fusion method and device and image acquisition equipment | |
| CN111260580B (en) | Image denoising method, computer device and computer readable storage medium | |
| US7876956B2 (en) | Noise reduction of panchromatic and color image | |
| CN111784603B (en) | RAW domain image denoising method, computer device and computer readable storage medium | |
| US8594451B2 (en) | Edge mapping incorporating panchromatic pixels | |
| US7522782B2 (en) | Digital image denoising | |
| US20110096987A1 (en) | Noise reduced color image using panchromatic image | |
| US20080240601A1 (en) | Edge mapping using panchromatic pixels | |
| KR101225056B1 (en) | Apparatus and method for reducing noise from image sensor | |
| CN111161188A (en) | Method for reducing image color noise, computer device and computer readable storage medium | |
| CN110868548A (en) | Image processing method and electronic equipment | |
| KR20210018136A (en) | Method and apparatus for image processing | |
| CN111031241B (en) | Image processing method and device, terminal and computer readable storage medium | |
| CN111161299A (en) | Image segmentation method, computer program, storage medium, and electronic device | |
| US20050025379A1 (en) | Digital imaging systems, articles of manufacture, and digital image processing methods | |
| US20060012693A1 (en) | Imaging process system, program and memory medium | |
| AU2015259903B2 (en) | Segmentation based image transform | |
| JP2000023173A (en) | Noise elimination method for solid-state color image pickup device | |
| CN112435182B (en) | Image noise reduction method and device | |
| WO2021102704A1 (en) | Image processing method and apparatus | |
| Yamaguchi et al. | Image demosaicking via chrominance images with parallel convolutional neural networks | |
| CN111988592B (en) | Image color reduction and enhancement circuit | |
| CN115205194B (en) | Image processing-based method, system and device for detecting coverage rate of armyworm plate | |
| WO2021102702A1 (en) | Image processing method and apparatus |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |