Improved Methods of Image Denoising Using Non-Sub Sampled Contourlet Transform

摘要

Images get corrupted with noise during its acquisition, processing, signal conditioning and transmission through noisy channels. The process of removing noise from images is termed as image denoising. Multiresolution methods of denoising were found to be superior compared to many other denoising methods previously existed. When images contain directional features such as contours, curves, edges, lines, etc the Contourlet as well as Curvelet transform methods of denoising could perform better than the earlier denoising methods using discrete wavelet transform (DWT), the pioneer in multiresolution image denoising. Some drawbacks of the Contourlet Transform method was solved by the more recently proposed Non-Sub sampled Contourlet transform(NSCT). The Support Vector Machine(SVM) method of classification of noisy and non-noisy pixels and subsequent thresholding was found to give better performance compared to direct thresholding to remove noise from NSCT decomposed images. Further, use of Orthogonal Matching Pursuit (OMP) after SVM classification can dispense with thresholding and is seen even better than the previous methods of denoising using NSCT using SVM and thresholding.