New norm-based image noise removal method

Abstract

The invention discloses a new norm-based image noise removal method comprising the following steps: block matching operation is performed on a target image, matched similar image blocks are laminatedinto a tensor form, and a corresponding principal component analysis model is established; a definition of a new norm is given, the definition of the new norm is generalized from the matrix to a tensor case, a rank function in the principal component analysis model established in the previous step is replaced with the new norm, an original NP difficult problem is converted into a problem that canbe solved, and an analytical solution for a new optimization problem is given; a low rank tensor obtained from the solution for the new problem is expanded into a matrix form, a de-noising result of each image subblock is obtained, an average value of overlapping areas can be obtained, and a final de-noising result can be obtained. Via implementation of the new norm-based image noise removal method, local and non-local statistics characteristics of images can be effectively fused, space structure information is effectively used, robustness for noise and singular points can be improved, and improved de-noising effects of a signal to noise ratio can be realized.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an image noise removal method based on a new norm. Background technique [0002] During the acquisition and transmission of image signals, the quality is often degraded due to the interference of various external noises, which seriously affects the subsequent processing of images, such as edge detection, target recognition, feature extraction, image segmentation, etc. Therefore, image denoising has become an image The most basic and important link in the processing process, and has attracted widespread attention. [0003] The current image denoising methods are mainly divided into local methods and non-local methods. A local approach usually involves convolving the image with some kind of kernel. However, because the local method only uses the relationship between the pixel in the area where the current pixel is located and the pixel for denoising, it will easily lose ...

AI Technical Summary

Problems solved by technology

[0004] However, in the existing non-local image denoising method, the inventors found that there are the following disadvantages: one, when the image block is expanded into a vector, it is easy to lose the spatial distribution information of the image; two, due to the principal component analysis model (matrix rank minimization problem) is a non-deterministic polynomial (NP) hard problem, and when the strong incoherence condition is satisfied, the nuclear norm minimization problem can obtain the solution of the original problem with a high probability

But unfortunately, in practical situations, the non-coherence condition is difficult to be satisfied

Images