Image denoising based on iterative generalized cross-validation and fast translation invariant

摘要

Wavelet shrinkage is a promising method in image denoising, the key factor of which lies in the threshold selection. A fast and effective wavelet denoising method, called Iterative Generalized Cross-Validation and Fast Translation Invariant (IGCV-FTI) is proposed, which reduces the computation cost of the standard Generalized Cross-Validation (GCV) method and efficiently suppresses the Pseudo-Gibbs phenomena with an extra gain of 1-1.87 dB in PSNR compared with GCV. In the proposed approach, we establish a novel functional relation between the GCV results of two neighboring thresholds based on integer wavelet transform, and combine it with threshold-search interval optimization. As a result, the proposed IGCV reduces the time complexity of original GCV algorithm by two orders of magnitude. In addition, a recursion strategy is applied to expedite the translation invariant. The high efficiency and proficient capacity to remove noise make IGCV-FTI a good choice for image denoising. (C) 2015 Elsevier Inc. All rights reserved.