基于视间相关性的多层次多视点视频编码快速预测算法

Multi-level Multi-view Video Coding Fast Prediction Algorithm Based on Inter-view Correlation

为解决多视点视频编码(MVC)计算复杂度过高的问题,提出一种多视点视频编码快速预测算法。计算当前宏块Direct模式的率失真代价并将其与自适应阈值进行比较,以提供有效的提前终止策略。当率失真代价小于自适应阈值时,Direct模式被选作最优模式,模式选择过程提前终止。当率失真代价大于自适应阈值时,利用运动复杂度将当前宏块分成不同的运动类型,每种类型的宏块仅检查指定的模式,从而使得不必要的模式检查过程提前结束。实验结果表明,该算法在保持几乎不变的编码效率基础上,大幅减少了MVC的计算量。

In order to solve greatly computational complexity in Multi-view Video Coding（ MVC）,a fast prediction algorithm is proposed based on inter-view correlation for multi-view video coding. This method computes the Rate Distortion（ RD） cost of the Direct mode and compares its RD cost with an adaptive threshold as an efficient early termination scheme. If this RD cost is smaller than the adaptive threshold,the Direct mode is selected as the optimal mode and the mode decision process is early terminated. Otherwise,motion complexity is used to divide the current macroblock into different motion types. And only assigned modes are checked for each macroblock,and the unnecessary mode checking process can be early terminated. Experimental results show that the proposed algorithm can significantly reduce the computational load for MVC,while keeping almost the same coding efficiency.