一种新的自适应图像组结构算法

A Novel Adaptive Group of Pictures Structure

运动补偿时域滤波(MCTF)将若干连续的视频帧组成一个图像组进行时域小波分解以获得时域可分级性。与图像组大小固定的结构相比,自适应图像组结构(AGS)提高了编码效率,但实现复杂度也大大增加。提出了一种新的自适应图像组结构(NAGS),利用帧间预测判断视频序列的时域变化特性,并根据视频序列的时域变化特性选择不同的图像组结构。新的自适应图像组结构大大降低了自适应图像组结构的计算复杂度,并解决了自适应图像组结构时延的振荡性问题。实验结果表明,新的自适应图像组结构的编码性能与自适应图像组结构基本一致。

Motion-compensated temporal filtering （MCTF） decomposed a group of pictures （GOP） consisted of several successive frames to support temporal scalability by using subband/wavelet transform. Compared with fixed -size GOP structure, adaptive GOP structure （AGS） improves the coding performance, but the computational complexity of AGS is also increased significantly. A novel adaptive GOP structure （NAGS） is proposed to estimates the temporal characters of video sequences by inter - prediction. According to the temporal characters, it adopts different size of GOP. NAGS significantly reduces the computational complexity of AGS, and fixes the problem of the jitter of temporal decomposition delay in AGS. The experimental results show that NAGS has almost the same coding performance as AGS.