Research Of Non-local Mean Image Denoising Algorithm Based On Structural Similarity

Noise as a common interference in image processing,often accompanied by the formation of the image,transmission and recording process.The presence of noise affects not only human visual perception,but also affects the subsequent image processing.Therefore,in order to deal with the noise in the image,it is an urgent to take effective way to denoise and improve the subsequent image processing capabilities.The goal of image denoising is to remove the noise while preserving the image edge structure information better.Based on the study of the non local mean denoising algorithm,a series of improvement measures are proposed by means of the similarity between the local blocks of the image.Specific research work is as follows:(1)Improvement of similarity block comparison using LBP texture feature.Aiming at the limitation of the gray scale vector and the local block in the horizontal azimuth translation,the similar block of the original non-local mean denoising algorithm is proposed to describe the image block structure information based on RHLBP texture feature.The RHLBP descriptor not only effectively expresses the texture of the image block in the horizontal direction,but also expresses the image block after a certain angle of rotation.The experimental results show that the non-local mean denoising algorithm based on RHLBP texture feature can better maintain the edge texture texture information and reduce the PSNR ratio by 1.6 dB.(2)Improve the comparison of similar blocks using bilateral structural tensors.The advantage of non-local mean denoising algorithm makes use of the similarity of image blocks,and the similarity measure directly affects the denoising effect.Therefore,aiming at the shortcomings of the original algorithm in the gray scale similarity,the method of describing the local information of the image block with the texture structure of the bilateral structure tensor is proposed.In order to improve the computational cost of similarity measure,the similarity of matrix is proposed as a local block similarity measure,and the algorithm is denoised by similarity weight adaptive parameter filtering and noise variance estimation,and finally improved by 1.2 dB Peak signal to noise ratio.(3)Image denoising experiment and evaluation index calculation.Firstly,the non-local mean algorithm based on RHLBP texture feature and the non-local mean algorithm based on bilateral structure tensor are experimented respectively.The experimental parameters are set and the experimental results are analyzed subjectively.The results show that the improved method has good denoising effect.Secondly,through the image evaluation index,including the peak signal to noise ratio and the similarity of the structure,the objective effect of the denoising effect is evaluated.The results show that the improved method has a certain degree of improvement on each index.Finally,we compare the structure similarity measure based on RHLBP and the denoising effect of structural similarity measure based on bilateral structure tensor.