Search

CN-116523768-B - Denoising method of depth image, image processing system and image processing device

CN116523768BCN 116523768 BCN116523768 BCN 116523768BCN-116523768-B

Abstract

The application provides a denoising method, an image processing system and an image processing device for a depth image, which relate to the technical field of image processing, wherein, according to the denoising method of the depth image, the abrupt number of the single pixel in the adjacent direction is used as the confidence weight, so that the computational resource consumed by denoising is greatly reduced. Meanwhile, the denoising method of the depth image does not need parameter adjustment, and the robustness of the denoising method is greatly improved. In addition, noise and edges are taken as mutation, and the mutation number in the adjacent direction of the pixels is taken as a confidence weight, so that better noise filtering effect is realized, and more texture details and other information are reserved. The method has the advantages of reducing the consumption of computing resources, enhancing the robustness of the denoising method, saving the detailed information such as the textures and edges of the image and improving the denoising effect.

Inventors

  • LEI SHUYU

Assignees

  • 宁波飞芯电子科技有限公司

Dates

Publication Date
20260508
Application Date
20230322

Claims (9)

  1. 1. The denoising method of the depth image is characterized by comprising the following steps of 1, performing edge pixel filling on an acquired depth image of a real scene; step 2, calculating the confidence coefficient of each pixel in the depth image; step 3, calculating a confidence weight matrix of each pixel according to the confidence, and calculating a denoised pixel value according to the confidence weight matrix; The confidence is the difference between the number of abrupt pixels in the adjacent pixels of the pixel and the number of the adjacent pixels.
  2. 2. The method according to claim 1, wherein an absolute value of a difference between a pixel value of the neighboring pixel and a pixel value of the present pixel is used as a mutation amount of the neighboring pixel.
  3. 3. The method according to claim 2, wherein a mutation threshold is preset, and the mutation amount is compared with the mutation threshold to determine whether the adjacent pixel is mutated.
  4. 4. A denoising method of a depth image according to claim 3, wherein the threshold is set by the number of adjacent pixels of the present pixel and the abrupt amount of the adjacent pixels.
  5. 5. The method according to claim 1, wherein the element value of the confidence weight matrix is the confidence level of the element as the present pixel.
  6. 6. The method of denoising a depth image according to claim 1, wherein the confidence weight matrix is normalized such that the sum of all element values of the confidence weight matrix is 1.
  7. 7. The method according to claim 1, wherein the denoised pixel values are obtained from the element values in the weight matrix, the pixel values of the corresponding depth image, and the number of elements of the weight matrix.
  8. 8. An image processing system, characterized in that the image processing system comprises an image acquisition module and an image processing module, the image processing module using the denoising method of the depth image according to any one of claims 1 to 7.
  9. 9. An image processing apparatus comprising a memory for storing an original image and a processor for performing the denoising method of a depth image according to any one of claims 1 to 7.

Description

Denoising method of depth image, image processing system and image processing device Technical Field The present application relates to the field of image processing technologies, and in particular, to a denoising method for a depth image, an image processing system, and an image processing apparatus. Background With the development of portable and inexpensive depth cameras, depth images have an increasingly important meaning in the basic research and application of the image processing field. By applying information on the depth image, performance of related research and application in the field of machine vision, such as image segmentation, object tracking, image recognition, image reconstruction, and the like, can be improved. However, due to the limitations of the existing depth camera technology principles, the quality of the depth image obtained from the depth camera is inferior to that of the visual image, and there is much noise interference, usually some random noise and different shapes of 'black holes' generated at the edges of the object, black surfaces and the like, namely, areas where the depth information is lost. These problems interfere with the application of depth information for depth images. Therefore, to obtain more accurate information, it is necessary to perform depth enhancement processing on the depth image to remove noise. At present, the denoising method of the depth image is mainly divided into a traditional image denoising method and a denoising method based on depth learning. The denoising method based on deep learning has higher complexity and higher requirement on hardware. The conventional image denoising method mainly comprises a filtering method, a transform domain denoising method and the like. The filtering method realizes weighted summation filtering of the image by utilizing local spatial correlation of the image, such as non-local mean filtering, bilateral filtering, guided filtering and the like. Although the method can effectively remove noise, the corresponding filter window size and weight parameters are required to be selected according to different noise images, and in order to improve the denoising effect and keep certain edge texture information, the complexity of an algorithm is high, and a large amount of calculation resources are required to be consumed. Unlike filtering methods, transform domain denoising often employs wavelet transform or sparse representation to perform preprocessing, and then uses the characteristics of the transform domain or redundancy of the representation for further processing. For example, the K-SVD algorithm, wavelet denoising and the like, but the transform domain denoising method also has the problems of needing to select a transform domain and a wavelet coefficient according to a noise image, and the like, and meanwhile, the complexity is high. How to solve the problems of large consumption of computing resources, low robustness and loss of detailed information such as texture, image edges and the like in the existing depth image denoising technology is a technical problem to be solved. Disclosure of Invention The application provides a denoising method, an image processing system and an image processing device for a depth image, which solve the problems of long time consumption, high complexity, low robustness and loss of detailed information such as texture, image edges and the like of the denoising method in the prior art. In order to achieve the above object, in a first aspect, an embodiment of the present application provides a denoising method for a depth image, which is characterized by comprising the following steps: step 1, filling edge pixels in an obtained depth image of a real scene; step 2, calculating the confidence coefficient of each pixel in the depth image; And 3, calculating a confidence weight matrix of each pixel according to the confidence, and calculating a denoised pixel value according to the confidence weight matrix. Optionally, the confidence is a difference between the number of abrupt pixels and the number of adjacent pixels in the adjacent pixels of the present pixel. Optionally, the absolute value of the difference between the pixel value of the adjacent pixel and the pixel value of the present pixel is used as the mutation amount of the adjacent pixel. Optionally, a mutation threshold is preset, and the mutation amount is compared with the mutation threshold to determine whether the adjacent pixel is mutated. Optionally, the threshold is set by the number of neighboring pixels of the present pixel and the abrupt amount of the neighboring pixels. Optionally, the element value of the confidence weight matrix is the confidence level when the element is used as the pixel. Optionally, the normalizing the confidence weight matrix is performed, so that the sum of all element values of the confidence weight matrix is 1. Optionally, the denoised pixel value is obtained from an element value in the weigh