Video Coding Based On Texture Analysis And Synthesis

In the last two decades, image and video compression has made a great improvement. The state-of-the-art JPEG2000 and MPEG-4 AVC/H.264 greatly outperform their previous generations in terms of coding efficiency. Whereas, one problem in all mainstream coding schemes is that the statistical redundancy among pixels is considered as the only adversary of compression, while the perceptual redundancy is totally neglected, though it is well known that any criterion is not as eloquent as the visual quality of an image. It is to say, the Rate Distortion Performance is mainly used as the evaluation criterion for the previous video compression standard. Although Rate Distortion is adopted abroadly, it does not completely reflect human vision speciality. Essentially, compression schemes and vision systems face a similar problem, that is, how to represent visual object in ways that are efficient and effective. It is possible to apply certain vision technologies in compression system toward perceptual fidelity rather than pixel-wise fidelity. So, this paper does some research on subjective vision quality and proposes a hybrid video encoder and decoder based on texture analysis and synthesis. It first identifies the texture regions by texture analyzer, and then synthesizes these areas by texture synthesizer. Other regions are encoded by traditional encoding methods.In order to detect and part the texture non-sensitive regions, this paper utilizes Edge Histogram Descriptor (EHD) and Scalable Color Descriptor (SCC) as similarity to segment and combine the video images. It computes the distance between different macorblocks utilizing EHD and SCC as similarity. If the distance is below a preset threshold, they are considered as similar, and are combined into a collection.The affine model, which is a kind of model in global motion estimation, is analyzed and researched in this paper. In order to synthesize the texture regions, the affine model is utilized to get the affine parameters. In the decoder, the corresponding texture macorblocks are synthesized by the affine parameter, which is sent into the stream.Besides, another texture synthesis method based on graph cut is proposed in...