Image Denoising Algorithm Based On PDE And Non-local Means

With the progress of information,image processing technology is widely used in military,medical,transportation and many other fields,which is inseparable from daily life.However,in the process of image collection and transmission,due to the equipment aging,interference from the external environment and other inevitable reasons will produce noise in the image,resulting in a decline in image quality,which will have an impact on the subsequent image processing.Therefore,in order to better analyze the image and meet the needs of high-quality image in real life,the first task is to denoise the image.For a long time,many scholars from some universities and enterprises at home and abroad have been studying the image denoising methods with the goal of improving the quality of image denoising.In recent years,the theoretical knowledge of partial differential equation denoising algorithm and non-local means denoising algorithm is relatively complete,and the denoising effect is relatively good.In this paper,partial differential equation and non-local means denoising algorithm are taken as research objects to carry out theoretical research on image denoising.The main research contents are as follows:(1)A new hybrid model was proposed to solve the ladder effect of PM model and the residual spots of YK model.In the PM denoised image,texture detection operator is introduced to construct a new diffusion function,calculate the local variance of the remaining image,use the relationship between gradient and residual local variance,adaptively select the diffusion coefficient,and introduce a weight parameter ? to combine the NAPM model with the fourth-order YK model.The experimental results show that the new model has better denoising performance.(2)The weight function of non-local means filtering algorithm is exponential function,and the weight distribution is unreasonable,which is easy to cause image blur after denoising.Aiming at this phenomenon,an improved weight non-local means filtering algorithm is proposed.In the first step,a new weight function is proposed,which makes the weight proportion of the lower pixel neighborhood similarity smaller,and that of the higher pixel neighborhood similarity In the second part,the neighborhood similarity function is introducedto improve the neighborhood similarity more accurately.Experimental results show that the algorithm can effectively remove noise and retain more edge texture information.(3)The non-local means denoising algorithm is very slow in calculating the similarity between pixels,and the denoising effect is not good,An improved algorithm based on DCT is proposed.In this algorithm,the image block is transformed into discrete cosine,and the low-frequency coefficient subspace is used to filter the data.Then the weight function of non-local means is combined with the spatial proximity function of bilateral filtering to construct a new weight function.Finally,the pixel similarity is calculated by the new weight function.The feasibility of the new algorithm is verified from two aspects of subjective evaluation criteria and objective evaluation criteria,and then the residual structural edge information in the image is observed through the method noise.The experiment shows that the image quality after denoising is significantly improved.