基于模型分割和小波变换的立体视频编码研究

Digital stereoscopic television will be become the next generation of TV after the arrival of DTV and HDTV. Therefore, MPEG-4 also support the stereo video stream. Since the stereo video has a left stream and a right stream, the obvious price for this increased realism is the doubling of data rate as compared to mono channel case. The transmitting bandwidth will be bottleneck of the transmission for stereo video. The method compressing the bandwidth is coding the left video stream (also named primary stream) as reference to compress the right video stream (also named auxiliary stream). The auxiliary stream as additional stream, only occupies little additional bandwidth as compared to primary stream. Therefore, the international research hotspot focuses on reducing the coding bit-rates of the additional stream. The goal is to transmit an additional stream with 30% or lower bandwidth of the primary stream by maintaining the stereo video with reasonable definition. The additional stream is also named target stream.This project creates a stereo video coding scheme based on a mixed-resolution model by theoretic and experimental research on disparity estimation based on hierarchical Markov random field (MRF) model, adaptive overlapped block disparity compensation, residue coding based on wavelet transform, frame estimation and interpolation model, and the subspace projection technique. In which, the additional bandwidth transmitting the target stream can be adjusted according to need of views at the decoder. The primary stream is independently coded by using a MPEG-type method. However, the auxiliary stream is coded and transmitted by exploiting disparity compensation prediction (performed to frame I), joint prediction of disparity and motion compensation (performed to frame P), and by using of frame estimation and interpolation at the decoder to reconstruct the frame B. Our algorithm has the performance of encoding with higher compression rates and quickly decoding. The auxiliary stream has a high compression ratio of 160~600 with very low additional bandwidth of about 8%~30% over single view bit stream.The created coding system of stereo images and stereo video sequences by us has following characteristics and innovation in theory and methods: proposing an algorithm for disparity estimation and compensation based on MRF model and adaptive overlapped bock, selecting a wavelet filters suitable to coding stereoscopic residual image. Also, we design a frame estimation mathematic model and propose an interpolation algorithm for coding stereo video based on least cost function, which can address the problem of disparity estimation in occluded regions, therefore the B frames of auxiliary stream are unnecessarily transmitted but reconstructed directly at the decoder with reasonable good quality.The research results of this project contribute its theoretic proof and performing scheme to transmission stereo video sequences with lower additional bandwidth, which is valuable for subsequent research and application.

通过对Markov随机场和率失真分割补偿模型、任意形状视差场分割及边界逼近和轮廓跟踪、分割信息及视差图的表达和编码、立体残差图象的小波变换等研究，发展一种边界估计误差小、可视干扰低、压缩效率高的立体视频编码新方法，且可作为多视点图象序列压缩编码的核心方法，在3D远程医学、3D可视通信、3DHDTV和3D虚拟现实等领域有良好应用前景。