Improved NLM Denoising Algorithm Based on Classification Preprocess in NSCT Domain

Non-local mean(NLM)algorithm for image denoising exists some problems such as the high computation complexity and the smooth restored image,etc.In this paper,we propose an improved NLM image denoising algorithm based on classification preprocess in the non-subsampled contourlet transform(NSCT)domain.Firstly,high frequency coefficients of the image in NSCT domain are obtained and classified into two classes using fuzzy support vector machine(FSVM),i.e.,non-noisy coefficients and noisy coefficients.Secondly,to reduce the computational complexity of the overall algorithm,only the noisy coefficients are retained for subsequent NLM processing.The polar harmonic transform(PHT)decomposition coefficients are used to replace the pixel values of the traditional NLM to calculate the similarity,which makes the computation process have better resistance to the change of direction.Finally,the modified bisquare function is used as the kernel function of similarity calculation.This is due to the modified bisquare function can be more in line with the residual characteristics between PHT decomposition coefficients,which makes the weight values of the similarity calculation are more accurate.Experiments are conducted on standard gray images and remote sensing images,and extensive experimental results show that the proposed algorithm can not only accelerate the computation speed of the traditional image denoising algorithms,but also has a better ability to preserve edges and structures,and the overall denoising performance of the proposed algorithm has been significantly improved.