| Summary: | 碩士 === 國立高雄應用科技大學 === 電機工程系碩士班 === 95 === In recent years, ISO proposed various video coding standards, like MPEG-4, H.261, H.263 and H.264 etc. H.264 is better than the others since it provides the higher compression rate while simultaneously maintains better video quality. However, H.264 employs various partition modes to encode different macro-blocks. To get the best visual quality and the lowest coding bit-rate, it is necessary to evaluate all of possible partition modes in terms of RD-cost, and then choose the best one for the target macro-block. Consequently, the computational complexity of H.264 is very high. Therefore, how to effectively reduce the cost of mode selection is a key issue for the performance of H.264 encoders.
To address this issue, this paper is aimed at developing a fast mode selection algorithm, called Spatial and Temporal Association Rule algorithm (STAR), for H.264 encoders. The basic concept of this algorithm is to mine spatial and temporal association rules from the patterns of mode selection, and predict a candidate mode according to the discovered association rules for the target macro-block instead of evaluating all of possible modes. To increase the correction of association rules, the proposed algorithm periodically flushes the collected patterns of mode selection. Moreover, it partition a frame into four subsection, and then collect the patterns of mode selection and mine association rules for each subsection due to considering the space locality of object motion. Finally, it also takes into account the modes of the macro-blocks adjacent to the target macro-block in order to decrease the errors of mode prediction. The proposed algorithm has been applied to a real H.264 encoder. The performance evaluation shows that this algorithm indeed effectively reduces the cost of mode selection, and maintains quality as well as the full-searching algorithm, and decreases the bit-rate of video encoding at the same time.
|
|---|
