Fast Motion Estimation for H.264/AVC using Dynamic Search Window
Sandro Moiron, M. Ghanbari · 2007
The motion estimation is one of the most time con- suming task in the encoding of inter predicted frames. Min- imising the computational complexity of this task is manda- tory in order to reduce the implementation cost, either in hardware or software. This paper proposes a dynamically adapted search algorithm for motion estimation to reduce the computational complexity. Moreover, a complementary re- finement method is proposed to improve the objective qual- ity of the video. Experimental results show that, when com- pared with the original fast full search, the proposed algo- rithm is able to achieve a significant reduction of the computa- tional complexity up to 98%, introducing only a small quality penalty near 0.08dB and a bitrate increase up to 3%. ity increases linearly with the number of reference frames available, the quality improvement has a logarithmic trend. Overall, combining the new tools significantly improves the compression gain making the H.264/AVC the flag-ship in compression efficiency. Compared with the previous stan- dards, the H.264/AVC is capable to obtain the same visual quality as its predecessor MPEG-2 (6) with 2 to 3 times lower bitrate (7). The most popular algorithm for motion estimation is the block-based full search algorithm due to a number of rea- sons. Its implementation is straightforward and fits well with the rectangular frames used in video. Furthermore, it effectively combines with other techniques such as block- based image transforms. However, the high computational load associated with this algorithm represents 60-80% of the H.264/AVC encoder complexity. Therefore, a signif- icant effort has been devoted in minimising the computa- tional complexity of this algorithm. This is performed in order to reduce the implementation cost of hardware and software based encoders, as well as unshielding the possi- bility to use it in portable devices.