DCT domain video foveation and transcoding for heterogeneous video communication.

摘要

Video transcoding converts a compressed video bit stream from one format to another format for purposes such as channel bandwidth adaptation and receiver adaptation in heterogeneous video communications. Due to the high data volume to be processed, efficient algorithms are needed for real-time video transcoding, especially for software based implementations. Moreover, since most video content is ultimately consumed by human beings, human vision system (HVS) based video transcoding can be desirable to achieve optimal visual quality at a given bit rate. This dissertation presents fast algorithms for efficient video transcoding that exploit the foveation property of the HVS to transcode a uniform resolution video stream into a foveated video stream for better visual quality at a given bit rate.; Foveation is attributed to the spatially-varying sampling density of the HVS, in which the resolution is highest within a few degrees of eccentricity around the fixation point, and drops quadratically away from this central region, or fovea, as a function of eccentricity. I develop an efficient DCT domain image foveation filtering technique to foveate a block DCT coded video stream. A serious problem for DCT domain foveation is blocking artifacts in peripheral areas far away from the foveation point. I propose an efficient algorithm that first measures the local visibility of blocking artifact at each block edge, and then removes the artifact if its visibility is above certain threshold determined experimentally. Since all operations are performed in the compressed DCT domain, the algorithm has low computational complexity.; DCT domain video transcoding has been shown to be more efficient than pixel domain methods. However, since the organization of data is on a block basis in the DCT domain, DCT domain inverse motion compensation becomes the bottleneck of DCT domain methods. In this work, two fast algorithms for DCT domain inverse motion compensation are proposed to improve the computational efficiency by an order of magnitude without compromising visual quality. A prototype of foveation embedded DCT domain video transcoder is developed, which supports both homogeneous and heterogeneous video transcoding. Two applications, foveated multipoint video conferencing and video transcoding for wireless communications, are implemented by using the foveation embedded video transcoding techniques.