Image Denoising Based on Statistical Nearest Neighbor and Wave Atom Transform

Abstract

In this research, a hybrid multiresolution method is proposed for image denoising, which is the combination of Statistical Nearest Neighbor (SNN) and Wave Atom Transform (WAT) approaches. The proposed model captures both patterns across oscillation and coherence of the pattern along with the oscillations. In this research, KODAK database is utilized for analyzing the proposed model performance, where the acquired color images are contaminated with Gaussian noise of σϵ [5,10,15,20,25,35,40 and 50] and noise range (0, 0.1, 0.35, 0.65, 0.8. 0.9, and 1.0). The denoising performance of the SNN-WAT model is analyzed by means of Gradient Magnitude Similarity Deviation (GMSD), Feature Similarity Index (FSIM), FSIM with chromatic information (FSIMC), Structural Similarity Index (SSIM), Mean SSIM index (MSSIM) and Peak Signal-to-Noise Ratio (PSNR). In the experimental phase, proposed SNN-WAT model averagely enhanced maximum of 4 dB and minimum of 0.1 dB of PSNR compared to the existing models like Non-Local-Means (NLM) with SNN technique.