Performance evaluation of the Hypercube based Prediction Algorithm for Multi-View Video Coding

Abstract

Multi-view video coding (MVC) is showing a new demand in the video communications, video surveillance systems, video teleconferencing, TV communications, and 3D video games. New algorithms are needed to improve video compression and reduce the complexity of multiple views in order to improve the MVC systems. This paper presents the performance evaluation of the hypercube prediction algorithm (HPA) for MVC. The main objective is to minimize the chain view of the dependencies based on the hypercube structure. Using spatio-temporal predictions based on the hypercube structure we show that MVC compression efficiency can be improved while keeping the dependencies low. The proposed HPA for MVC provides increased flexibility in selecting prediction references by reducing the view dependency by a factor of 2. The performance is compared with a linear prediction algorithm (LPA) that uses one spatial and one temporal reference frame. We show that HPA can be used to allow flexible prediction structures that improve the encoding performance while keeping the sufficient the PSNR video quality.