Research On Image And Video Image Denoising Algorithm Based On Visual Saliency

Image and video images are disturbed by noise during sampling,processing,transmission and storage,resulting in the degradation of visual information.The human eye has different characteristics of different levels of sensitivity to noise in different regions of the image,and the image can be divided into significant and non-significant regions.The main content of the paper's research is to use visual saliency to pre-process the noise image according to the visually significant characteristics of the human eye,to obtain the region of interest in the noise image that is of interest to the human eye,and then use a denoising algorithm with good image texture protection to process it.The non-interesting areas are denoised using filtering algorithms with faster computing speeds.The denoising algorithms involved in this paper include BM3D(block-matching and 3-D filtering)algorithm,VBM3D(video block-matching and 3-D filtering)algorithm and mean value filtering algorithm.The main content and innovations of this article are as follows:1.Image denoising algorithm based on visual saliencyAiming at the problem that there are limitations to denoising algorithms nowadays,that is,the denoising algorithm that has better protection for image texture details often takes a long time,and the denoising effect of a filtering algorithm with a fast denoising processing speed is not ideal.In this paper,the visually significant characteristics of the human eye are applied to the field of image denoising.The noise image is pre-classified by saliency calculation,so that the block matching search is performed in the significant image block group instead of all the image blocks,and the search range of the narrower block matching reduces the workload of BM3 D.At the same time,the non-significant image blocks are processed with fast mean filtering,and finally the denoising results are synthesized.The results show that the proposed method is not only denoising compared to using BM3 D algorithm alone,but also has a significantly higher processing speed and can obtain higher subjective evaluation results.2.Threshold improvementThe threshold of hard threshold classification is fixed.When the threshold is too large,the significant image blocks after classification are often insufficient,resulting in the area that can be searched by the block matching part of the subsequent denoising algorithm is too small,and finally the denoising effect is not ideal.If the threshold is too small,the efficiency of the algorithm cannot be effectively improved.Inspired by the OTSU threshold algorithm,this paper improves the original hard threshold classification into an adaptive threshold classification algorithm.Experimental results show that the improved threshold classification algorithm effectively protects more image details and ensures the accuracy of image block saliency classification.3.Video denoising algorithm based on visual saliencyThere is rich redundant information between frames of the video sequence,so the video denoising algorithm is different from the image processing method.The time domain information is added in the denoising,which makes VBM3 D no longer performs search block matching in a single image.Searching between frames results in a larger search range and more time-consuming algorithms.To solve this problem,based on the above idea of image denoising improvement,the significant characteristics of human eyes are applied to the denoising of video images,and block matching is performed within significant image block groups,which narrows the search scope.And use the fast filter to filter the non-significant image block to denoise and accelerate it further.Experiments show that the proposed algorithm has a competitive advantage compared with previous algorithms in computing time and subjective evaluation.