Overview of View Synthesis Prediction for Multi-view Video Coding

With a wide range of viewing angles, the multi-view video can provide more realistic feeling at arbitrary viewpoints. However, because of a huge amount of data from multiple cameras, we need to develop an efficient coding method. One of the promising approaches for multi-view video coding is view synthesis prediction which generates an additional reference frame. In this paper, we explain the principal idea of view synthesis prediction and propose a view interpolation and coding scheme. For view interpolation, we employ initial disparity estimation, variable block matching, and disparity error correction methods. After generating an intermediate view image, we apply a ‘VIP Ppicture’ coding method with the additional reference frame, where we have five additional motion estimation modes and a modified motion vector prediction scheme. By computer simulations, we have improved the coding gain by 0.66 dB on average for the well synthesized sequences.