Rate distortion optimization-based video coding rapid search method in GPU

Abstract

The invention relates to a rate distortion optimization-based video coding rapid search method in GPU. The method comprises a 64 * 64 PU searching step, other PU searching steps and a fractional pixelsearching step. The method is applied to 64 * 64 PU search and other PU search. Two-step search based on rate distortion optimization is provided for different sizes of PUs; first, obtaining an optimal matching position, calculating rate distortion cost corresponding to the MV, the PMV and the SAD by using a rate distortion cost calculation method, storing the rate distortion cost into Shard Memory, calculating minimum cost by using an iterative algorithm, and storing the minimum cost, the SAD and MV corresponding to the minimum cost into the Global Memory. Compared with the prior art, the method has the advantages that the calculation complexity is reduced, and the calculation time is greatly shortened.

Description

technical field [0001] The invention relates to the technical field of GPU video coding, in particular to a rate-distortion optimization-based fast search method for video coding in GPU. Background technique [0002] To improve computing performance, HEVC (High Efficiency Video Coding) video coding can be executed on the basis of heterogeneous platform parallel architecture, and based on multi-core CPU and GPU framework. Among them, the HEVC video encoder generally uses the full search method to search when performing motion estimation in the GPU. Although the full search method can find the best partition mode, it needs to traverse all partitions, and the computational complexity is very high, making the motion estimation process difficult. It takes a long time. Another search method is the TZSearch algorithm. The TZSearch algorithm uses a diamond search method. The process is: [0003] (1) Determine the initial search point; [0004] (2) Start with a step size of 1, and...

AI Technical Summary

Problems solved by technology

[0008]According to the principle of the TZSearch fast search algorithm, a diamond search algorithm will also be designed in the GPU to reduce the computational complexity, but this method cannot obtain PMV (Motion Vector Prediction, prediction Motion vector), because PMV is the motion vector obtained from the motion prediction of adjacent CTU (CodingTreeUnit, coding tree unit, the size is 64×64), and in GPU motion estimation, the adjacent CTU processing is processed in parallel at the same time, The corresponding PMV cannot be obtained from adjacent CTUs, so the rate-distortion optimization technology cannot be used to select the best CTU partition mode, resulting in a decrease in the rate-distortion performance. Therefore, finding out the PMV is the key to solving the rate-distortion optimization problem in the GPU.

[0009]Because the CPU has a strong versatility to handle various data types, at the same time logical judgment will introduce a large number of branch jumps and interrupt processing, which are Makes the internal structure of the CPU extremely complex

What the GPU is facing is a highly uniform type, large-scale data that is independent of each other, and a pure computing environment that does not need to be interrupted. This also makes the GPU unable to have efficient logical judgment performance like the CPU, so it is difficult in the GPU. Implementing a search algorithm with a large number of logical judgments like TZSearch will take a lot of time even if it can be realized, which is contrary to the original intention of reducing computational complexity and improving encoding speed, so the fast search algorithm in GPU cannot fully use TZSearch fast algorithm , need to improve

Images