Directionality-based modified coefficient scanning for image coding

Abstract

Coefficient scanning is one of the most significant parts in block-based image and video compression coding standards. Zigzag scanning is applied to convert the 2-D alternating current (AC) coefficients into a 1-D data stream in an energy decreasing way, but it cannot reflect the particularity for every image blocks. In this paper, we design six scanning templates (including zigzag scanning template), which are depending on the directional texture information of image, so we call it directionality-based modified coefficient scanning method. In the experiment, five categories of images with different directional texture information are trained to design scanning template. Then we apply the six templates to each 8×8 block and each block will choose one template which makes its code stream shortest. Experiment results show that the method in this paper achieves better coding efficiency obviously than not only the zigzag scanning method but also the context-based modified zigzag scanning method adopted in JPEG.