Paper ID | A-2-3.6 |
Paper Title |
RSP-BT:An Advanced Parallel Method for Depth Map Motion Estimation |
Authors |
Xiaoyan Xie, Anqi Wang, Yun Zhu, Chuanzhan Hu, Zhuolin Du, Xi'an University of Posts and Telecommunications, China |
Session |
A-2-3: Reconfigurable Computing and Performance Evaluation |
Time | Wednesday, 09 December, 17:15 - 19:15 |
Presentation Time: | Wednesday, 09 December, 18:30 - 18:45 Check your Time Zone |
|
All times are in New Zealand Time (UTC +13) |
Topic |
Signal Processing Systems: Design and Implementation (SPS): Special Session: Reconfigurable Computing and Performance Evaluation |
Abstract |
Aiming at the high computational complexity and long encoding time caused by insufficient consideration of the homogeneous area in depth map motion estimation of 3D High Efficiency Video Coding(3D-HEVC), An optimization method to reduce search points based on block type (RSP-BT) is proposed. Firstly, according to the pixel distribution features of coding blocks, the coding blocks are classified into homogeneous or edge areas. And then the different search strategies are selected adaptively in the light of center-biased characteristic of the motion vectors for different types of area. Thereby, the overall calculation overhead is reduced. Experimental results show that, compared with the full-search algorithm, the proposed algorithm has an average Peak Signal to Noise Ratio (PSNR) loss of only 0.0519dB; Compared with the TZSearch (Test Zone Search) algorithm, the search points is reduced by 11.25% on average. In order to further improve the coding speed, a parallel implementation is given on the video array processor of the optimized algorithm, in the view of there is no data correlation among the matching operates for different coding blocks.The experimental results show that the average speedup of the designed parallel scheme is 2.8940. |