Efficient hybrid image denoising scheme based on SVM classification

Abstract

An efficient hybrid image denoising method based on support vector machine (SVM) classification is designed and developed in this communication. Here, the input noisy image is first divided into large number of overlapping patches followed by extraction of local features from each patches using scale invariant feature transform (SIFT). Based on a predefined threshold, SVM is used to classify the patches into two classes such as texture patches and flat patches. The texture patches are processed through gradient histogram preservation (GHP) where flat patches are reconstructed using sparse based denoising method using the analysis, K-SVD. Finally, the reconstructed image is obtained by merging the results of the two denoising process. To find the efficiency of our proposed hybrid denoising scheme, we perform several experiments on some standard noisy images and compare the results with some state-of-the-art denoising methods Experiments indicate that the proposed hybrid scheme performs better denoising performance, particularly in preserving edges and textures as compared to existing denoising method.