Compression artifacts removal by signal adaptive weighted sum technique

Abstract

Many mobile devices compress images excessively to meet limited bandwidth requirements and adopt the Block-based Discrete Cosine Transform (BDCT) coding structure. This produces inevitably the visually annoying noises including blocking artifacts in the decoded images. We present a Signal Adaptive Weighted Sum (SAWS) technique of block boundary pixels to alleviate the blocking artifacts encountered in highly compressed images. The weights are adjusted adaptively according to the directional correlation and activities of local areas. To avoid blurring the original details, we employ a parameter that can control the strength of deblocking. Comprehensive experiments demonstrate that the proposed approach achieves excellent visual quality and PSNR compared to a number of deblocking methods in the literature. Moreover, the low complexity of the proposed algorithm facilitates the application to mobile devices.