How Can Intra Correlation Be Exploited Better?

Abstract

This paper studies how to better exploit intra correlation in image/intra-frame coding. Unlike the previous assumption, where the correlation between samples is determined only by their distance, image is considered to present local orientation property. The correlation between samples is determined by not only the distance but also the link orientation. Thus, a directional Filtering Transform (dFT) is proposed in this paper to exploit local orientation correlation among samples. It consists of directional filtering and optional transform. Furthermore, this paper analyzes the theoretical coding gain of the proposed directional filtering in terms of signal power spectral density (PSD). Both numerical results and actual compression results of the intra-frame coding in H.264 demonstrate the advantages of the proposed dFT with decoding complexity increased slightly.