Efficient Prediction Structure for Key Pictures in Multiview Video Coding

Compared to simulcast, the utilization of inter-view prediction at key pictures can dramatically improve the compression efficiency of multiview video coding (MVC). The reconstructed key picture in previous adjacent view is used as the inter-view reference picture of the key picture in the current coding view. The efficiency of inter-view prediction may deteriorate due to the existence of mismatch between views. An efficient prediction structure for key pictures is proposed by introducing multiple reference pictures prediction scheme. Experimental results show that average gains of 0.219dB can be achieved for key pictures compared to the reference prediction structure of MVC.

I. Introduction

Multiview video (MVV), captured by a set of cameras simultaneously, comprises rich 3-D information of a scene and has gained increasing research interests due to its wide variety of applications, including free-viewpoint television, 3D television, immersive teleconference and surveillance [1]. Since MVV leads to an enormous amount of data, a crucial issue for its wide application is how to encode all views efficiently. Multiview Video Coding (MVC) is then put forward with a focus on improving the compression efficiency with regard to all views. A new standard for MVC [2] is developed by the Joint Video Team (JVT) of ISO/IEC MPEG and ITU-T VCEG, which has been included in the latest version of H.264/AVC [3].