Stereo image coding using dynamic template-matching prediction

Template matching (TM) has been originally proposed as a texture synthesis tool. However, it has been successfully exploited for spatial and inter-frame prediction in video coding. In this paper we investigate the use of TM prediction for stereo image coding. In order to efficiently encode stereo images, the TM algorithm was optimized for stereo disparity prediction. Additionally, we have used the state-of-the-art pattern matching based algorithm, the Multidimensional Multiscale Parser (MMP), to encode images predicted with the proposed scheme. Experimental results of the developed stereo image encoder show that TM is able to efficiently exploit the redundancy between stereo views. The use of the new prediction method with MMP also achieves important coding gains over the state-of-the-art transform based H.264/AVC standard, for stereo image coding.