【摘要】 不同硬件设备具有不同的计算能力,能否在任意给定计算能力约束下达到最好的编码效率,是当前视频编码研究领域的一个极具挑战性问题.同时,随着分块结构越来越灵活的编码标准不断出现(如:HEVC,H.264等),运动估计不得不反复地应用在大小不同的各种分块上,导致其对编码总体计算复杂度的影响愈加重要.在此背景下,本文提出了一种针对运动估计的计算能力可伸缩(Complexity scalable)优化算法.我们通过对运动估计过程中预测失真度和计算复杂度的变化规律建模,发现根据各宏块的特性设置不同的预测失真度阈值可以优化地分配计算资源.而该阈值的大小则恰恰是各宏块的最小预测失真度加上一个由复杂度约束统一决定的偏移量.有鉴于此,我们进一步构造了计算能力可伸缩的优化运动估计算法,在不增加额外计算量的前提下,快速地得到各个宏块所对应的优化阈值,并完成运动估计.通过实验分析,该算法不仅具备自动适应不同计算复杂度约束的能力,而且在任意给定的复杂度约束下,都能提供优化的编码性能.更多还原

【Abstract】 One of the major challenges of video coding is how to attain the best coding performance under different constraints of computational complexity with various hardware. The computational scalability of coding algorithms is particularly important on this occasion. On the other hand,more and more complicated coding structures are employed to repeatedly applied motion estimation on each block. And the algorithms of motion estimation exerts significant influence on the overall performance of video coding. Therefore,we proposed an optimized motion estimation algorithm with scalable complexity.This algorithm can automatically adapt to different video contents and optimally allocate the computing resources by setting appropriate distortion thresholds. These thresholds are the minimal predicted distortion of each macroblocks plus a unified offset. And all of these thresholds can be computed easily with negligible costs. According to our experiments,our proposed algorithm not only provides scalable complexity,but also achieves better coding performance at the same computational expenses.更多还原