Research On Denoising Algorithm Of Salt And Pepper Noise Based On Self Learning Edge Structure Classification

Salt and pepper noise is a common type of noise in image processing,and its black and white characteristics will seriously reduce the visual quality of the image.In view of the inaccurate noise positioning of the existing filtering methods and the jagged and blurred edges of the restored image,the following two salt-and-pepper noise removal algorithms are proposed in this thesis:(1)Salt-and-pepper noise removal algorithm based on self-learning edge structure classification.A salt-and-pepper noise removal algorithm based on self-learning edge structure classification is proposed to improve the problems of inaccurate noise location and large error of denoising value in existing filtering methods.Firstly,according to the edge structure of the pixel point,it is divided into edge point and internal point,and the edge point can be divided into two types: 1:3 and 2:2,according to the number of neighboring points belonging to different areas.Then the edge structure coefficients of similar pixels are calculated by the least square method.Secondly,for the suspected noise points with extreme gray value,set the threshold for secondary detection.Finally,using the idea of the switch method,the original value of the signal point remains unchanged,and the edge structure of the noise point is determined,and the corresponding edge structure coefficient is used to calculate the denoising value.Experimental results show that the algorithm has higher peak signal-to-noise ratio and structural similarity,and has better visual performance in low and medium noise densities.(2)High-density salt-and-pepper noise removal algorithm based on multi-scale sub-image.An improved image salt-and-pepper noise removal algorithm based on selflearning edge structure classification is proposed to improve the original algorithm’s problems,such as blurred edges and poor image details under high-density noise.Firstly,the high-density noise image is divided into 2×2 pixel grids,and then the gray value of the signal points in the 2×2 pixel grids is used to restore the noise points in the pixel grids,reducing the overall noise density of the image.Secondly,a multi-scale subimage denoising method is proposed: for isolated noise points,the noise is directly denoised in the original image;for non-isolated noise points,they are segmented by sub-images,and the size of the sub-image is determined by the noise level.In addition,the pixel points whose edge structure is 2:2 type are refined,and a simpler and more effective method for judging its area is proposed.Finally,determine the edge structure of the noise point and use the corresponding edge structure coefficient to calculate the denoising value,and verify the denoising value obtained.Experimental results show that,compared with other algorithms,this algorithm has higher peak signal-to-noise ratio and structural similarity,and has better visual performance under high noise density,and is expected to be applied to image defogging,image enhancement and other applications.