Multi-view Video Coding Method Based On Epipolar Geometry

As the latest and most promising video coding standard, H.264/AVC is efficient in video coding and ensures meanwhile satisfactory video quality. Captured by synchronized cameras from different viewpoints, multi-view video sequences contain rich depth information and provide more realistic feelings for viewers observing natural scenes. In recent years, multi-view video technique is drawing rising attention in the realms of academia and industries, and has become the hotspot of the video area. The huge amount of data from multi-view video makes it difficult for transmission and thus needs efficient compression. One key approach of data compression is to remove the spatially and temporally relevant information in the video sequences. In multi-view video, there is huge time and spatial redundancy. To reduce that, simultaneous motion and disparity estimation can be useful in multi-view video coding.As a key concept in computer vision, the epipolar geometry characterizes the constraint relations of the same object between two viewpoints. In multi-view video sequences, any two frames at the same time satisfy the constraints of epipolar geometry. The motion vector is generally determined by the displacement of objects in the frames at different times, while the disparity vector reflects the displacement of objects in the different views at the same time. When the motion and disparity vectors relate to the same content, all the associated pixels are contained in both temporally and spatially neighboring frames, leading to a closed-form quadrilateral constraints relationship.With consideration to the temporal and spatial redundancy, a fast motion and disparity estimation search method is proposed in this paper based on the quadrilateral constraints of epipolar geometry from multi-views. This method reduces the encoding time from motion and disparity estimation, yielding the increase in coding efficiency. In addition, this method is tested and analysized on the platform of JMVC(Joint Multi-view Video Coding) .Compared with the TZ fast search algorithm, this method ensures little influence upon image quality and bit rate, while the total encoding time is reduced by 45-60% averagely.