Auto-scaled Incremental Tensor Subspace Learning for Region Based Rate Control Application

摘要

In this paper, we proposed a method that employs the auto-scaled incremental eigenspace learning to locate the salient distortion areas continually in the video to serve the purpose of region based rate control application. Compared to other locating methods, the auto-scaled incremental eigenspace learning locating method can achieve locating the salient distortion areas robustly and accurately, and specifically in realtime. In addition, for the case that there exists the overlap/occlusion between different salient distortion areas, the proposed method can also obtain accurate location information which could make the region based rate control and bit allocation to reach higher efficiency in many applications. The experiment results of the proposed algorithm demonstrate the subject visual quality of the video has been improved greatly.