Vaguelette-Wavelet Deconvolution via Compressive Sampling

Abstract

Vaguelette-wavelet deconvolution (VWD) is known as a transform-based image restoration technique that involves applying wavelet-domain denoising to an observed image, followed by the Fourier-domain blur inversion, which can prevent noise amplification in conventional Fourier-domain deconvolution techniques. However, the direct application of VWD often results in poorly restored images because of the artifacts that result from the denoising and inversion stages. In this paper, we thus propose a new image deconvolution technique based on VWD that applies a cycle-spinning and averaging technique and a compressive-sampling-based recovery technique to suppress these artifacts. The experimental results revealed that the proposed technique outperforms the existing deconvolution techniques in terms of both restored image quality and computational cost.