Temporal-Spatial Hybrid Prediction Model For Video Coding

With the synchronous development process of both international and domestic next generation video coding standards including H.265 and AVS 2, video coding technology has entered a new stage of rapid progress. The next generation video coding standards have brought forward higher requirement to the coding performance. Therefore, to achieve more significant performance improvement through update of existing means or proposal of new methods has currently been the focus of both academia and industry.As one of the core technologies used by the hybrid video coding frameworks, prediction eliminates video redundancy and thus improves the coding efficiency. In mainstream video compression standards like H.264/AVC, the prediction of a macroblock utilizes either spatial or temporal correlation only. However, different regions within one macroblock may come from different objects and therefore may have different spatial and temporal correlation. Some regions may have strong spatial correlation and some regions may have strong temporal correlation. In this situation, neither intra nor inter prediction can provide accurate prediction results for all pixels in the macroblock. Consequently, from the perspective of comprehensive utilization of all available correlation, current prediction means might not be the most efficient.In order to further enhance the prediction capability, this paper proposes a temporal-spatial hybrid prediction model which utilizes both spatial and temporal correlation simultaneously within one macroblock. In the proposed model, based upon the best prediction mode determined through Rate-Distortion Optimization, the encoder correspondingly adds spatial or temporal information into current prediction macroblock and obtains new hybrid prediction macroblock. Meanwhile, a weighted temporal-spatial hybrid prediction model which is adaptive to the spatial correlation is proposed as optimization to the basic hybrid prediction strategy. Furthermore, experimental results demonstrate that the proposed hybrid prediction model is able to gain a performance improvement of about 0.1dB without bit-rate increase compared with traditional prediction methods in H.264/AVC. As a conclusion, the proposed novel prediction method is able to significantly improve the coding efficiency of video coding frameworks.