An Improved Directional Intra Prediction Method Based on Sparse Approximation for Image Compression

Abstract

In image and video coding, intra image prediction has attracted considerable attention in recent years. This paper describes a new intra prediction method that aims to improve image coding efficiency, based on directional intra prediction and sparse approximation, denominated directional sparse intra prediction (DSIP). In this method, instead of searching for the best prediction in a causal neighborhood, the directional characteristic of the block to be predicted is included with the causal neighborhood during sparse prediction. An image block having self-similarity directional characteristics with the block to be predicted is included with the causal neighborhood for better prediction results. The prediction and coding performance of this new intra prediction method have been evaluated through a block based image compression algorithm. Simulation results show that the proposed method yields both improved image prediction quality and coding efficiency when compared with existing intra prediction methods. Results also demonstrate that the image compression method with proposed intra prediction outperforms JPEG and JPEG2000 in terms of objective quality.