Research On Multi-view Video Coding Thchnology

Multi-view video, as a new video, uses multiple cameras from different angles to uptake scenes information. Facing huge amount of data, we need to use time correlation and inter-view correlation to remove redundancy and to improve compression efficiency. Random access performance is used as a major measurement for multi-view video, and designing a reasonable prediction structure can balance coding efficiency and random access performance. The interference of shooting angle of cameras, the scene light intensity and other factors lead to color differences between the inter-view images, so correlation between different views is reduced and coding efficiency is affected. We need to use color correction to remove these differences.This thesis analyzes the MVC hierarchical B-frames prediction structure, and analyzes quantitatively the distribution of best reference image between different time layers and different sequences. Analyzing the time correlation and the inter-view correlation, we can improve prediction structure, so as to improve random access performance. This thesis proposes three new prediction structure, according to the time correlation and the inter-view correlation. Experiments show that the compression efficiency impact is relatively small, the proposed enhanced prediction structure can effectively improve random access performance.This thesis analyses three color correction methods during the encoding process and preprocess. Experiments show that these color correction methods make the image more similar to the reference image, reducing the color difference between inter-view image, achieving purpose of color correction.As no considering the characteristics of the image itself factors in the above methods, the correction accuracy will be affected, especially in the case of more texture detail. This thesis proposes a color correction method based on SIFT. SIFT algorithm is good at finding match points between different images, even though translation, rotation or color changing happens. This thesis introduces SIFT feature information into the traditional histogram correction and color cast correction, to fix the correction factor. Experiments show that the proposed methods can obtain a better correction effect.