Motion vector derivation of deformable block

Na Zhang,Xiaopeng Fan,Debin Zhao,Wen Gao
DOI: https://doi.org/10.1109/ICIP.2012.6467168
2012-01-01
ICIP
Abstract:Motion estimation (ME) plays an important role in most video encoding systems since it could significantly affect coding performance. However, both the next generation video coding standard High Efficiency Video Coding (HEVC) and the current video coding standard H.264/MPEG-4 AVC employ block matching motion estimation (BMME) which is based on translation motion model. This makes it difficult to represent the complex motion accurately such as rotation, zoom, and etc. In this paper, we propose an adjacent-block-based prediction model to improve the prediction performance of a deformable block. Based on this model, the motion information of each 4×4 block, i.e. the minimum partition (MP) in a coding unit (CU) of HEVC, is derived from the motion information of the nearest neighbors to the four corners of current prediction unit (PU). We integrate our method into HEVC as an additional choice of its merge mode. Simulation results show that our proposed method has better performance compared to HM4.0, the BD bit rate saving is up to 15.4%, while the encoding and decoding complexities are almost the same.
What problem does this paper attempt to address?