Research On Inter Prediction Engine For Real-Time High-Definition Video Encoders

As the development of communications, video processing and communications are widely used today, and the quantity of video signal data also increases sharply due to the increase of video resolution. In such a situation, the video coding technologies are improved rapidly, which leads to the improved compression ratio and the multiplied complexity. However, when designing a high-definition real-time encoder, not all of the encoding tools supported by the standards could be implemented in the encoder due to the encoding algorithm characteristics and the limited hardware resources, so the coding gain is not as good as that of the best case. Considering that the inter prediction engine consumes the most of the hardware resources, this paper focused on the inter prediction engine and improved its key modules to improve the coding gain. On the other hand, video coding standards are competing and updating, which results to the existence of multiple standards. So algorithms and structures that could be used by multiple standards are more valuable. In this paper, the flexibility of the designed algorithms and structures are considered.By analyzing the similar hybrid coding algorithms adopted by different video coding standards, a frame structure of video encoder for multiple standards was proposed in this paper. Based on the frame structure, this paper studied on several key technologies of the inter prediction engine. First, models of "resources supply - processing capability" relationship for fast motion search algorithms and full search algorithms were established in this paper, which are helpful for designers to choose the proper monition search algorithm for VLSI based inter prediction engine. Second, this paper proposed a new measure of block matching to choose the best motion vector in motion search. The VLSI implementation of the new measure saves nearly 50% of the area consumption compared to the traditional designs. Third, A new all-zero block detection algorithm based on the proposed new measure was presented in this paper, and the all-zero block detection rate was increased by 20%-30%, compared to the existing algorithms. Besides, a new method based on the idea of iterative and intermediate data sharing was proposed for chroma interpolation. The resource consumption of the high-speed interpolation module was reduced by about 48% at the same processing capability.The development of these new technologies can effectively reduce the hardware resource consumption of the inter prediction engine, and improve the processing capabilities of the engine. These technologies make it easier for the video encoder to implement more coding tools and improve the coding gain under the same constraints of hardware resource.