Improving Intra Coding in H.264\AVC by Image Epitome

Abstract

In the most advanced video coding standard H.264\AVC, intra prediction is adopted to eliminate the spatial correlation in intra coded frame. But compared with the inter-frame prediction, the performance of intra prediction is much less than that of inter-frame prediction, and how to improve the performance of intra prediction is of great significance to further enhance the H.264\AVC coding efficiency. In this paper, a new intra prediction method based on image epitome is proposed. In the new approach, image epitome is extracted to describle the most essential representation of the image or the video, and the traditional DC prediction mode is modified to generate prediction from the image epitome through template matching. Image epitome is also required to be losslessly coded and transmitted to the decoder side later for generating the same prediction with the encoder side. The new intra prediction method presented in this paper not only can eliminate the correlation between the current block and the overall image without any syntax change, but also can reduce the temporal redundacy among adjacent images to some extent while avoiding dependency between current image and other coded images. Experimental results show that, the proposed intra coding method can significantly improve the intra coding efficiency in H.264\AVC comparing with that without our method, and at the same bit-rate, the PSNR can be raised by more than 0.8dB at most. The new intra coding method has a great potential in the applications such as video editing, video streaming and high-definition movie compression.