Study On Key Technology Of Representations For Multiview Mixed Resolution Video Based On Super Resolution

With the rapid development of 3D processing, 3D displaying and micro-processor technologies, 3D video processing technology has drawn a lot of attention, and has developed rapidly from theory research to applications. And it offers consumers an opportunity to have a high quality visual experience in 3D video applications, such as 3DTV and Free-viewpoint TV, but the huge challenge is followed, that is, in the face of tremendous of multi-view video data, how to represent these data efficiently so that these video data can be transmitted in the band limited channel. At the same time, high resolution and high quality video images after decoding are required. Therefore, it is necessary to research the representations for multi-view video to reduce data size and bit rate. This dissertation is going to have a research on the key technology of representations for multi-view mixed resolution video based on super resolution, according to the existing multi-view video framework and image super resolution reconstruction technology.Firstly, the dissertation introduces several typical SR methods in mixed resolution multi-view video plus depth(MR-MVD) framework. And then it does a research on the correlation between different views in multi-view video, and proposes a novel SR scheme based on correlation between views in MR-MVD format. Compared with traditional SR method, which only borrows adjacent high resolution(HR) image as reference frame in multi-view mixed resolution video, it makes use of all HR views. According to correlation coefficient between views, the dissertation proposes a high frequency and low frequency fusion method based on optimal weight in transform domain, to provide LR frames with detail information. It also introduces several common representation formats for multi-view mixed resolution video. Finally, according to the human eyes temporal masking and binocular vision suppression principle, the dissertation demonstrates a novel representation model intra-interview union mixed resolution for multi-view video content, and briefly introduces the transmission system based on the proposed model in the existing codec framework.Simulation results indicate that the proposed novel SR method and the intra-inter multi-view video representation model have better performance both in subjective and objective.