Adaptive Predictive and Bi-Predictive Eigenspace Approach for Video Coding

Abstract

In this paper, a video coding method which is based on the adaptive eigenspace introduces. The eigenspace is updated to adopt into the changing scenes in the video sequence. The method encodes the video frames by using the eigenspace which is generated by incorporating the original whole frames as well as the predicted and bi-predicted frames. Two alternative approaches have been proposed. The first approach, which uses the past frames to generate the adaptive eigenspace, is referred to as the predictive adaptive video coding (P-AVC) and uses a PSNR based threshold to regulate the quality of the coded frame. The second approach, which is referred to as the bi-directional predictive adaptive video coding (BP-AVC), uses both past and future frames for updating the eigenspace and also employs a PSNR based threshold to regulate the video quality. The visual and quantitative results show that BP-AVC outperforms P-AVC approach. Furthermore, when compared with H.264 AVC, the BP-AVC and P-AVC methods generate superior or competitive bit rate performance for the tested benchmark video sequences.