GPU accelerated motion and disparity estimations for multiview coding

Multiview video coding (MVC) is an extension of H.264/AVC standard. Its purpose is to further increase compression efficiency when multiple video sources are presented for coding. The computational complexity of MVC, however, is extremely high due to the motion estimation (ME) between the frames and disparity estimation (DE) between the views, contributing to more than 99% of overall complexity of the coder. For MVC to find a common place for real-time applications requires acceleration of ME and DE engines. On the other hand, there is a recent uptake of graphical processing unit (GPU) computing to significantly enhance the performance of signal and image processing applications, through massive-parallel processing. This paper presents the development and GPU implementation of a parallel full search algorithm to significantly reduce the computational complexity of ME and DE over the full search and TZsearch estimations on a sequential processor, by a factor of 300 and 4, respectively.