Motion Estimation And Refinement In Video Frame-skipping Transcoder

With Internet and mobile communication's high development, various client devices having different performance appear, such as: mobile phone, PDA, handheld computer, etc. These client devices hope to have access to Internet in wireless way, and share Internet's content. As Internet and wireless network have different bandwidth, they have different bit-rate accordingly. If compressed video bit stream is transmitted in wireless channel from Internet to client devices, coded video bit stream will not match the wireless channel. Thus, proxies need to be set between generic WWW servers and client devices, transcoding video bit stream which has been coded and compressed. Proxy can transform high bitrate into lower bitrate to make video bit stream transmitted, accurately in mobile wireless network, and provide client devices with different quality of service.This thesis researches following several aspects:(1) Various transcoding models in video transcoding are analysed. Video transcoding can be carried out not only in pixel domain but also in transform domain, namely DCT domain. Video Transcoding have three means: code rate transformation, resolution transformation, transformation between coding standard. Code rate transformation commonly is reducing video's code rate. Resolution transformation commonly is reducing video's temporal resolution and spatial resolution. Transformation of coding standard is to transcode coded video with a standard into another coding standard.(2) In transcoding, reusing motion vectors extracted from an incoming video bit stream is studied in this thesis. Three composed methods to compose a motion vector from the incoming motion vectors in frame-skipping transcoding, such as bilinear interpolation method , FDVS and forward vector method are also studied in detail. Experimental results show that the forward vector method is computationally more efficient and improves image quality better compared with other methods.(3) Motion vector refinement for high-performance transcoding is studied in thelast chapter. We show that the incoming motion vectors become nonoptimal due to the reconstruction errors. To achieve the best video quality possible, a new motion estimation should be performed in the transcoder. Motion vector refinement for high-performance frame-skipping transcoding, especially, is studied. We propose a fast-search motion vector refinement scheme and an optimal search window for refinement in frame-skipping transcoding. Experimental results show that the proposed fast-search method of HAVS can improve PSNR of picture greatly with less additional computation.