NON LOCAL MEANS IMAGE DENOISING ALGORITHM BASED ON TEXTURE FEATURES

Abstract

Abstract While transmission of an image it generally happens that it gets affected by noise. Well then applying non local means denoising algorithm in order to recover the original image, structural information such as texture and edges are easily lost due to the smoothening effect which it creates. Our work is basically centered at embedding the texture features in the filtered image so that PSNR of the filtered image is increased. Firstly, we extract the texture features using entropy and standard deviation functions and then these texture features are used in modified weight function to get the modified weights values in non local means algorithm. In order to increase the weights of the similar structure we have not only considered the Euclidean weighted Gaussian distance but texture structure has also been contemplated and merged with the Euclidean weighted Gaussian distance function. Experimental results have validated that our method is superior to non local means algorithm. Using standard deviation texture feature we are getting a PSNR of 36.2774 and using local entropy texture feature we are getting a PSNR of 36.34. and when these features are used simultaneously i.e local entropy as well as standard deviation then we are getting a PSNR of 34.88. Please note that above values have been calculated for a single image and for a single value of standard deviation.