A Patch-Number and Bandwidth Adaptive Non-Local Kernel Regression Algorithm For Multiview Image Denoising

Abstract

This paper presents an automatic patch number selection method for bandwidth adaptive non-local kernel regression (BA-NLKR) algorithm, which was recently proposed for improving the performance of conventional non-local kernel regression (NLKR) in image processing. Although BA-NLKR addressed the important issue of bandwidth selection, the number of non-local patches, which impacts the integration of local and non-local information, however is chosen empirically. In this paper, we propose a new algorithm for automatic patch number selection based on the intersecting confidence intervals (ICI) rule in order to achieve better performance. Moreover, the proposed patch number and bandwidth adaptive NLKR (PBA-NLKR) is applied to the denoising problem of multiview images. The effectiveness of the proposed algorithm is illustrated by experimental results on denoising for both single-view and multi-view images.