CN-121998857-A - Method, device, equipment, storage medium and program product for image noise reduction

CN121998857ACN 121998857 ACN121998857 ACN 121998857ACN-121998857-A

Abstract

The embodiment of the application provides a method, a device, equipment, a storage medium and a program product for image noise reduction. The method comprises the steps of obtaining an original image, wherein the original image comprises at least two channels, determining image data of the channels under at least two preset resolutions according to the original image for each channel, determining noise-reduced data corresponding to the channels according to the image data of the channels under at least two preset resolutions, and determining noise-reduced images corresponding to the original image according to the noise-reduced data corresponding to each channel. For different channels in an original image, image data under different resolutions are obtained, information of multiple scales in the image is reserved, and for each channel, the image data under each preset resolution are combined to perform independent noise reduction on the channel, so that the mutual influence of the channels is avoided, and the noise reduction precision of the image is effectively improved.

Inventors

YUE HAO
GU YI

Assignees

上海途擎微电子有限公司

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. A method of image denoising, comprising: Acquiring an original image, wherein the original image comprises at least two channels; for each channel, determining image data of the channel under at least two preset resolutions according to the original image; determining noise-reduced data corresponding to the channel according to the image data of the channel under at least two preset resolutions; and determining a noise-reduced image corresponding to the original image according to the noise-reduced data corresponding to each channel.
2. The method of claim 1, wherein determining the noise-reduced data corresponding to the channel from the image data of the channel at least two preset resolutions, comprises: For each preset resolution, traversing a first window from the image data of the channel under the preset resolution, and determining a second window corresponding to the first window, wherein the first window and the second window are both image windows with preset shapes and preset sizes, one first window corresponds to a plurality of second windows, and the second windows are image windows except the first window in the image data; Determining noise-reduced data of the channel under the preset resolution according to the current pixel value of the pixel point in each first window and the current pixel value of the pixel point in a second window corresponding to each first window; and determining the noise-reduced data corresponding to the channel according to the noise-reduced data of the channel under each preset resolution.
3. The method of claim 2, wherein determining the noise reduced data of the channel at the preset resolution based on the current pixel value of the pixel point in each of the first windows and the current pixel value of the pixel point in the second window corresponding to each of the first windows comprises: For each first window, determining target noise reduction intensity of the first window according to the current pixel value of a pixel point in the first window, wherein the target noise reduction intensity represents the degree of noise reduction treatment on the first window; For each second window corresponding to the first window, determining weight information of the second window according to the current pixel value of the pixel point in the first window, the current pixel value of the pixel point in the second window and the target noise reduction intensity of the first window, wherein the weight information represents the influence degree of the second window on the first window; Determining a target pixel value of a central point of the first window according to weight information of each second window corresponding to the first window and a current pixel value of the central point of each second window corresponding to the first window; and determining the noise-reduced data of the channel under the preset resolution according to the target pixel value of the central point in each first window.
4. A method according to claim 3, wherein determining the target noise reduction intensity for the first window based on the current pixel values for the pixels in the first window comprises: determining preset index information of the first window according to the current pixel value of the pixel point in the first window, wherein the preset index information comprises at least one of average brightness, variance and radius; Determining a noise reduction intensity influence factor corresponding to the preset index information according to a preset association relation, wherein the noise reduction intensity influence factor represents the influence degree of the preset index information on noise reduction, and the preset association relation represents the association relation between the preset index information and the noise reduction intensity influence factor; And determining the target noise reduction intensity of the first window based on the preset basic noise reduction intensity according to the noise reduction intensity influence factors corresponding to the preset index information.
5. A method according to claim 3, wherein determining the weight information of the second window based on the current pixel value of the pixel in the first window, the current pixel value of the pixel in the second window, and the target noise reduction intensity of the first window comprises: Determining an absolute difference sum of the second window according to the current pixel value of the pixel point in the first window and the current pixel value of the pixel point in the second window, wherein the absolute difference sum represents the similarity degree between the first window and the second window; And determining weight information of the second window according to the sum of the target noise reduction intensity of the first window and the absolute difference value of the second window.
6. A method according to claim 3, wherein determining the target pixel value of the center point of the first window according to the weight information of each second window corresponding to the first window and the current pixel value of the center point of each second window corresponding to the first window comprises: Determining an initial noise reduction pixel value corresponding to a second window according to the weight information of the second window corresponding to the first window and the current pixel value of the center point of the second window, wherein the initial noise reduction pixel value represents a pixel value obtained by noise reduction processing of the center point of the second window; and determining a target pixel value of the central point of the first window according to the initial noise reduction pixel value corresponding to each second window.
7. The method of claim 2, wherein the predetermined resolution includes at least an original resolution, 1/4 resolution, and 1/16 resolution of the original image, and determining the noise-reduced data corresponding to the channel based on the noise-reduced data of the channel at each predetermined resolution comprises: Determining initial noise reduction data based on a preset pyramid fusion strategy according to the noise reduction data of the channel under 1/16 resolution and the noise reduction data of the channel under 1/4 resolution, wherein the preset pyramid fusion strategy represents an image processing strategy for carrying out pyramid fusion on a multi-scale image, and the initial noise reduction data represents the noise reduction data fused with 1/4 resolution and 1/16 resolution; and determining the noise-reduced data corresponding to the channel based on a preset pyramid fusion strategy according to the initial noise-reduced data and the noise-reduced data of the channel under the original resolution.
8. The method according to any one of claims 1-7, wherein the preset resolution comprises at least an original resolution, 1/4 resolution, 1/16 resolution of the original image, wherein determining image data of the channel at least two preset resolutions from the original image comprises: determining image data of the channel under the original resolution according to the original image; Performing 1/4 downsampling processing on the original image to obtain image data of the channel under 1/4 resolution; And carrying out 1/16 downsampling processing on the original image to obtain image data of the channel under 1/16 resolution.
9. An apparatus for image noise reduction, comprising: the image acquisition unit is used for acquiring an original image, wherein the original image comprises at least two channels; A data determining unit, configured to determine, for each channel, image data of the channel at least two preset resolutions according to the original image; the image denoising unit is used for determining denoising data corresponding to the channel according to the image data of the channel under at least two preset resolutions; And the image determining unit is used for determining the noise-reduced image corresponding to the original image according to the noise-reduced data corresponding to each channel.
10. An electronic device/computer-readable storage medium/computer program product, characterized in that, The electronic device comprising a memory, a processor, the memory storing computer-executable instructions, the processor executing the computer-executable instructions stored in the memory, causing the processor to perform the method of any one of claims 1-8, and/or, The computer readable storage medium having stored therein computer executable instructions which, when executed by a processor, are adapted to carry out the method of any one of claims 1-8, and/or, The computer program product comprising a computer program which, when executed by a processor, implements the method of any of claims 1-8.

Description

Method, device, equipment, storage medium and program product for image noise reduction Technical Field The present application relates to the field of artificial intelligence, and in particular, to a method, apparatus, device, storage medium, and program product for image noise reduction. Background With the development of the information age, the requirements of people on image quality are increasing. In the process of acquiring an image by the CMOS sensor, various noises are introduced due to the influence of sensor material properties, working environment, electronic components, circuit structures and the like, so that the image is blurred, details are lost or edges are distorted. How to reduce the loss of detail and noise residue in the noise reduction process is a problem to be solved. Disclosure of Invention The embodiment of the application provides a method, a device, equipment, a storage medium and a program product for image noise reduction, which are used for improving the processing precision of image noise reduction. In a first aspect, an embodiment of the present application provides a method for image noise reduction, including: Acquiring an original image, wherein the original image comprises at least two channels; for each channel, determining image data of the channel under at least two preset resolutions according to the original image; determining noise-reduced data corresponding to the channel according to the image data of the channel under at least two preset resolutions; and determining a noise-reduced image corresponding to the original image according to the noise-reduced data corresponding to each channel. In one example, determining the noise-reduced data corresponding to the channel according to the image data of the channel under at least two preset resolutions includes: For each preset resolution, traversing a first window from the image data of the channel under the preset resolution, and determining a second window corresponding to the first window, wherein the first window and the second window are both image windows with preset shapes and preset sizes, one first window corresponds to a plurality of second windows, and the second windows are image windows except the first window in the image data; Determining noise-reduced data of the channel under the preset resolution according to the current pixel value of the pixel point in each first window and the current pixel value of the pixel point in a second window corresponding to each first window; and determining the noise-reduced data corresponding to the channel according to the noise-reduced data of the channel under each preset resolution. The beneficial effects of setting up like this lie in, when carrying out non-local mean filtering under single resolution, regard first window as anchor point, many second windows as similar block source, combine multiscale pyramid framework, solve traditional NLM and have promoted whole image signal to noise ratio and image quality because of the intensity mismatch problem of making an uproar that falls that single scale matches leads to in complicated texture region. In one example, determining the noise-reduced data of the channel under the preset resolution according to the current pixel value of the pixel point in each first window and the current pixel value of the pixel point in the second window corresponding to each first window includes: For each first window, determining target noise reduction intensity of the first window according to the current pixel value of a pixel point in the first window, wherein the target noise reduction intensity represents the degree of noise reduction treatment on the first window; For each second window corresponding to the first window, determining weight information of the second window according to the current pixel value of the pixel point in the first window, the current pixel value of the pixel point in the second window and the target noise reduction intensity of the first window, wherein the weight information represents the influence degree of the second window on the first window; Determining a target pixel value of a central point of the first window according to weight information of each second window corresponding to the first window and a current pixel value of the central point of each second window corresponding to the first window; and determining the noise-reduced data of the channel under the preset resolution according to the target pixel value of the central point in each first window. The beneficial effects of the arrangement are that in the RAW domain image noise reduction process, the target noise reduction intensity which is strictly matched with the content characteristics of each first window is dynamically generated, the similarity weight distribution of the second window is regulated and controlled, the self-adaptive reconstruction of the center pixel is further completed, the dual effects of realizing refined nois