Research On Improved BM3D Image Denoising Algorithm Based On Region Division

As one of the best image denoising algorithms,BM3 D can effectively remove noise in images,but there are still some problems,such as loss of image details,poor denoising effect on high-noise images,and high time complexity of the algorithm.The purpose of image denoising is to keep the useful information of the image as much as possible while removing the noise.Since the edge,texture and other details of the image and noise are mainly concentrated in the high-frequency part of the image signal,the main goal of image denoising is to restore the high-frequency information of the image,which can be divided into two steps.The first step is to extract the high-frequency features of the image,and the second step is to separate the signal and noise in the high-frequency.Based on the above two steps,based on the idea of zoning,improved the BM3 D image denoising algorithm,first proposed a modified BM3 D algorithm based on multiscale geometric analysis,through the selection of subregional domain transform method,fully and sparse extract image high-frequency characteristics,thus more to restore the image edge and texture detail information;Secondly,an improved BM3 D algorithm based on adaptive threshold is proposed.By selecting the threshold calculation method in different regions,the high frequency signal and noise of the image can be separated more accurately,so that the image details can be preserved more completely while the noise can be removed effectively.The main innovation work and research results of this thesis are summarized as follows:(1)Aiming at the problem of image high-frequency detail loss existing in BM3 D algorithm,this thesis proposes an improved BM3 D algorithm based on multi-scale geometric analysis,and improves the joint hard threshold filtering in BM3 D algorithm.First,image segmentation algorithm is used to divide the image into smooth region and edge region.Secondly,different domain transformation methods are selected in different regions.In the smooth region,the Hare wavelet transform is selected to extract the local point features of the sub-blocks on different scales.In the edge region,the nonsubsampled contour transform is selected to extract the linear structure features of the sub-image blocks from different scales and directions.Finally,a joint hard threshold filter is applied to the high frequency sub-blocks.Experimental results show that,compared with the standard BM3 D algorithm,the proposed algorithm improves the denoising image quality by 2.62 d B on average,and improves the structural similarity by 0.06 on average,and better retains the edge,texture and other details of the image.(2)Aiming at the problem that BM3 D algorithm has low ability to remove high intensity noise,this thesis proposes an improved BM3 D algorithm based on adaptive threshold,and further improves the joint hard threshold filtering in BM3 D algorithm.First,image segmentation algorithm is used to divide the image into smooth region and edge region.Secondly,different threshold calculation methods are selected in different regions.In the smooth region,the improved unified threshold method is selected to calculate the thresholds of the high-frequency sub-blocks at different scales separately.In the edge region,the improved maximal inter-class variance method is selected to calculate the threshold values of the high-frequency sub-block groups in different scales and directions.Finally,a joint adaptive threshold filter is applied to the high frequency sub-blocks.Experimental results show that,compared with the standard BM3 D algorithm,the denoising image quality of the proposed algorithm is improved by 8.52 d B on average,and the structural similarity is improved by 0.24 on average,which further improves the denoising effect of high-noise images.