Search

CN-122023170-A - Image noise reduction method, device, computer equipment and readable storage medium

CN122023170ACN 122023170 ACN122023170 ACN 122023170ACN-122023170-A

Abstract

The application relates to an image noise reduction method, an image noise reduction device, a computer device, a readable storage medium and a computer program product. The method comprises the steps of obtaining an image to be denoised in a video, carrying out 2D denoising treatment on the image to be denoised of a first frame of the video, outputting a target denoised image of a current frame, returning to the step of obtaining the image to be denoised in the video, carrying out 2D denoising treatment on any non-first frame of the image to be denoised of the video as an original image of the current frame, obtaining a 2D denoised image of the current frame, obtaining a target denoised image of a last frame of the current frame as a reference image of the current frame, carrying out 3D denoising treatment on the 2D denoised image of the current frame according to picture motion difference between the original image of the current frame and the reference image of the current frame, outputting the target denoised image of the current frame, and returning to the step of obtaining the image to be denoised in the video until the image to be denoised is the last frame of the video. By adopting the method, the computing resources in image noise reduction can be saved.

Inventors

  • YANG YUANFEI
  • XU HUI
  • HUANG HAONAN

Assignees

  • 珠海市杰理科技股份有限公司

Dates

Publication Date
20260512
Application Date
20260114

Claims (18)

  1. 1. A method of image denoising, the method comprising: Acquiring an image to be noise reduced in a video; under the condition that the image to be noise-reduced is the first frame of the video, 2D noise reduction processing is carried out on the image to be noise-reduced as an original image of a current frame, a target noise-reduced image of the current frame is output, and the step of obtaining the image to be noise-reduced in the video is returned; And under the condition that the image to be denoised is any non-initial frame of the video, performing 2D denoising processing on the image to be denoised as an original image of a current frame to obtain a 2D denoising image of the current frame, acquiring a target denoising image of a previous frame of the current frame to serve as a reference image of the current frame, performing 3D denoising processing on the 2D denoising image of the current frame according to picture motion difference between the original image of the current frame and the reference image of the current frame, outputting the target denoising image of the current frame, and returning to the step of acquiring the image to be denoised in the video until the image to be denoised is the last frame of the video.
  2. 2. The method according to claim 1, wherein the performing 3D noise reduction processing on the 2D noise reduction image of the current frame according to a picture motion difference between the original image of the current frame and the reference image of the current frame, and outputting the target noise reduction image of the current frame, includes: According to the similarity between a first pixel point in an original image of the current frame and a second pixel point in a reference image of the current frame, performing motion estimation and noise reduction on pixels to be noise reduced in a 2D noise reduction image of the current frame to obtain a motion estimation and noise reduction result; searching a third pixel point in the reference image of the current frame, and performing motion compensation noise reduction on a pixel to be noise reduced in the 2D noise reduction image of the current frame according to the similarity between a first pixel point in the original image of the current frame and the third pixel point in the reference image of the current frame to obtain a motion compensation noise reduction result; and obtaining a target noise reduction image of the current frame according to the motion estimation noise reduction result and the motion compensation noise reduction result.
  3. 3. The method according to claim 2, wherein the searching out the third pixel point in the reference image of the current frame includes: acquiring a similarity array between each pixel point in a preset window and the first pixel point in a preset window in a reference image of the current frame; and determining the highest similarity in the similarity array, and taking the pixel point corresponding to the highest similarity as the third pixel point.
  4. 4. The method according to claim 2, wherein the performing motion estimation denoising on the pixel to be denoised in the 2D denoising image of the current frame according to the similarity between the first pixel point in the original image of the current frame and the second pixel point in the reference image of the current frame to obtain the motion estimation denoising result includes: the similarity between a first pixel point in an original image of the current frame and a second pixel point in a reference image of the current frame is used as the motion estimation similarity of the current frame; Performing edge compensation on the motion estimation similarity to obtain the motion estimation similarity after edge compensation; and according to the edge-compensated motion estimation similarity, performing motion estimation and noise reduction on pixels to be noise reduced in the 2D noise reduction image of the current frame to obtain a motion estimation and noise reduction result.
  5. 5. The method of claim 4, wherein edge compensating the motion estimation similarity to obtain an edge-compensated motion estimation similarity comprises: Mapping the position of the pixel to be reduced in the original image of the current frame into a current frame edge value based on a preset edge detection operator, and mapping the position of the pixel to be reduced in the reference image of the current frame into a reference frame edge value of the pixel to be reduced; And taking the absolute value of the difference value between the current frame edge value and the reference frame edge value of the pixel to be noise reduced as a first edge compensation value, and fusing the first edge compensation value and the motion estimation similarity to obtain the motion estimation similarity after edge compensation.
  6. 6. The method of claim 4, wherein performing motion estimation denoising on the 2D denoising image of the current frame according to the edge-compensated motion estimation similarity to obtain a motion estimation denoising result, comprises: inquiring corresponding motion estimation noise reduction weights from a first weight table constructed in advance according to the motion estimation similarity after the edge compensation; according to the motion estimation noise reduction weight, performing motion estimation noise reduction on pixels to be noise reduced in the 2D noise reduction image of the current frame to obtain a motion estimation noise reduction result; the first weight table is used for recording a mapping relation between the motion estimation similarity and the motion estimation noise reduction weight.
  7. 7. The method of claim 6, wherein the querying the corresponding motion estimation noise reduction weight from the pre-constructed first weight table according to the edge compensated motion estimation similarity comprises: obtaining the maximum value in the first weight table from the first weight table; Determining a first motion estimation weight from the first weight table according to the motion estimation similarity after the edge compensation, wherein the first motion estimation weight is used for representing the weight for performing motion estimation and noise reduction on the reference image of the current frame; taking the difference value between the maximum value in the first weight table and the first motion estimation weight as the second motion estimation weight, wherein the second motion estimation weight is used for representing the weight for performing motion estimation and noise reduction on the 2D noise reduction image of the current frame; and taking the first motion estimation weight and the second motion estimation weight as the motion estimation noise reduction weight.
  8. 8. The method of claim 7, wherein performing motion estimation denoising on pixels to be denoised in the 2D denoised image of the current frame according to the motion estimation denoising weight to obtain a motion estimation denoising result, comprises: acquiring a to-be-denoised position of a to-be-denoised pixel in a 2D denoised image of the current frame; weighting pixels of the corresponding position of the position to be reduced on the reference image of the current frame by using the first motion estimation weight to obtain a first motion estimation weighting result; weighting pixels to be noise reduced in the 2D noise reduction image of the current frame by using the second motion estimation weight to obtain a motion estimation second weighting result; And fusing the motion estimation first weighted result and the motion estimation second weighted result to obtain the motion estimation noise reduction result.
  9. 9. The method according to claim 2, wherein the performing motion compensation noise reduction on the pixel to be reduced in the 2D noise reduction image of the current frame according to the similarity between the first pixel point in the original image of the current frame and the third pixel point in the reference image of the current frame to obtain a motion compensation noise reduction result includes: The similarity between a first pixel point in an original image of the current frame and a third pixel point in a reference image of the current frame is used as the motion compensation similarity of the current frame; Performing edge compensation on the motion compensation similarity to obtain edge-compensated motion compensation similarity; and according to the edge-compensated motion compensation similarity, performing motion compensation noise reduction on pixels to be noise reduced in the 2D noise reduction image of the current frame to obtain a motion compensation noise reduction result.
  10. 10. The method of claim 9, wherein edge compensating the motion compensated similarity to obtain an edge compensated motion compensated similarity comprises: mapping the position of the pixel to be reduced in the original image of the current frame into a current frame edge value based on a preset edge detection operator, and mapping the position of the third pixel point in the reference image of the current frame into a reference frame edge value of the third pixel point; And taking the absolute value of the difference value between the edge value of the current frame and the edge value of the reference frame of the third pixel point as a second edge compensation value, and fusing the second edge compensation value and the motion compensation similarity to obtain the motion compensation similarity after edge compensation.
  11. 11. The method according to claim 9, wherein performing motion compensation noise reduction on the 2D noise reduction image of the current frame according to the edge compensated motion compensation similarity to obtain a motion compensation noise reduction result comprises: inquiring corresponding motion compensation noise reduction weight from a second weight table constructed in advance according to the motion compensation similarity after the edge compensation; according to the motion compensation noise reduction weight, performing motion compensation noise reduction on pixels to be noise reduced in the 2D noise reduction image of the current frame to obtain a motion compensation noise reduction result; The second weight table is used for recording a mapping relation between the motion compensation similarity and the motion compensation noise reduction weight.
  12. 12. The method of claim 11, wherein the querying the corresponding motion compensated noise reduction weight from the pre-constructed second weight table according to the edge compensated motion compensation similarity comprises: Obtaining the maximum value in the second weight table from the second weight table; Determining a first motion compensation weight from the second weight table according to the motion compensation similarity after the edge compensation, wherein the first motion compensation weight is used for representing the weight for performing motion compensation noise reduction on the reference image of the current frame; The difference value between the maximum value in the second weight table and the first motion compensation weight is used as the second motion compensation weight, wherein the second motion compensation weight is used for representing the weight for carrying out motion compensation noise reduction on the 2D noise reduction image of the current frame; and taking the first motion compensation weight and the second motion compensation weight as the motion compensation noise reduction weight.
  13. 13. The method according to claim 12, wherein the performing motion compensation noise reduction on the pixels to be noise reduced in the 2D noise reduced image of the current frame according to the motion compensation noise reduction weight to obtain a motion compensation noise reduction result includes: acquiring a to-be-denoised position of a to-be-denoised pixel in a 2D denoised image of the current frame; Weighting pixels of the corresponding positions of the third pixel points on the reference image of the current frame by using the first motion compensation weights to obtain a motion compensation first weighting result; weighting pixels to be noise reduced in the 2D noise reduction image of the current frame by using the second motion compensation weight to obtain a motion compensation second weighting result; And fusing the motion compensation first weighting result and the motion compensation second weighting result to obtain the motion compensation noise reduction result.
  14. 14. The method according to claim 2, wherein obtaining the target noise reduction image of the current frame according to the motion estimation noise reduction result and the motion compensation noise reduction result comprises: obtaining the motion grade of the pixel to be reduced according to the similarity between a first pixel point in the original image of the current frame and a second pixel point in the reference image of the current frame; And according to the fusion weight matched with the motion grade, fusing the motion estimation noise reduction result and the motion compensation noise reduction result to obtain a target noise reduction image of the current frame.
  15. 15. An image noise reduction device, the device comprising: the image acquisition module is used for acquiring an image to be noise reduced in the video; The first frame denoising module is used for performing 2D denoising processing on the image to be denoised as an original image of a current frame under the condition that the image to be denoised is the first frame of the video, outputting a target denoising image of the current frame, and returning to the step of acquiring the image to be denoised in the video; The non-first frame denoising module is used for performing 2D denoising processing on the image to be denoised serving as an original image of a current frame under the condition that the image to be denoised is any non-first frame of the video to obtain a 2D denoising image of the current frame, acquiring a target denoising image of a previous frame of the current frame to serve as a reference image of the current frame, performing 3D denoising processing on the 2D denoising image of the current frame according to the picture motion difference between the original image of the current frame and the reference image of the current frame, outputting the target denoising image of the current frame, and returning to the step of acquiring the image to be denoised in the video until the image to be denoised is the last frame of the video.
  16. 16. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 14 when the computer program is executed.
  17. 17. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 14.
  18. 18. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 14.

Description

Image noise reduction method, device, computer equipment and readable storage medium Technical Field The present application relates to the field of computer vision, and in particular, to an image noise reduction method, an image noise reduction apparatus, a computer device, a readable storage medium, and a computer program product. Background In the field of image processing and computer vision, noise is a ubiquitous interference factor, and is generated due to the influence of sensors, illumination conditions, circuit thermal noise and the like in the image acquisition process, so that the visual quality of an image can be seriously reduced, and the accuracy of subsequent processing tasks (such as target detection, image segmentation, target identification and the like) can be influenced. Therefore, image noise reduction technology has been the focus of research in this field. Traditional image denoising techniques can be largely divided into two main categories, 2D (single frame) image denoising and 3D (multi-frame/video) image denoising. However, the conventional 2D image denoising technology and 3D image denoising technology have the problem of large computational complexity, and the 2D image denoising technology and the 3D image denoising technology cannot be effectively combined, so that the image denoising process is complex in flow, a large amount of cache resources (line_buffer) are wasted, and the computational resources in the image denoising process are consumed in a large amount, so that the image denoising process efficiency is reduced. Therefore, the conventional technology has a problem of large consumption of computing resources in terms of image noise reduction. Disclosure of Invention In view of the foregoing, it is desirable to provide an image noise reduction method, apparatus, computer device, readable storage medium, and computer program product that can save computing resources. In a first aspect, the present application provides an image noise reduction method, the method comprising the steps of: Acquiring an image to be noise reduced in a video; under the condition that the image to be noise-reduced is the first frame of the video, 2D noise reduction processing is carried out on the image to be noise-reduced as an original image of a current frame, a target noise-reduced image of the current frame is output, and the step of obtaining the image to be noise-reduced in the video is returned; And under the condition that the image to be denoised is any non-initial frame of the video, performing 2D denoising processing on the image to be denoised as an original image of a current frame to obtain a 2D denoising image of the current frame, acquiring a target denoising image of a previous frame of the current frame to serve as a reference image of the current frame, performing 3D denoising processing on the 2D denoising image of the current frame according to picture motion difference between the original image of the current frame and the reference image of the current frame, outputting the target denoising image of the current frame, and returning to the step of acquiring the image to be denoised in the video until the image to be denoised is the last frame of the video. In one embodiment, the performing 3D noise reduction processing on the 2D noise reduction image of the current frame according to a picture motion difference between the original image of the current frame and the reference image of the current frame, and outputting the target noise reduction image of the current frame includes: According to the similarity between a first pixel point in an original image of the current frame and a second pixel point in a reference image of the current frame, performing motion estimation and noise reduction on pixels to be noise reduced in a 2D noise reduction image of the current frame to obtain a motion estimation and noise reduction result; searching a third pixel point in the reference image of the current frame, and performing motion compensation noise reduction on a pixel to be noise reduced in the 2D noise reduction image of the current frame according to the similarity between a first pixel point in the original image of the current frame and the third pixel point in the reference image of the current frame to obtain a motion compensation noise reduction result; and obtaining a target noise reduction image of the current frame according to the motion estimation noise reduction result and the motion compensation noise reduction result. In a second aspect, the present application provides an image noise reduction apparatus, the apparatus comprising: the image acquisition module is used for acquiring an image to be noise reduced in the video; The first frame denoising module is used for performing 2D denoising processing on the image to be denoised as an original image of a current frame under the condition that the image to be denoised is the first frame of the video, outputting a