Stereoscopic Video Coding

Abstract

Coding of the stereoscopic video source has received significant interest recently. The MPEG committee decided to form an ad hoc group to define a new profile which is referred to as Multiview Profile (MVP) [4]. The importance of multiview video representation is also recognized by the MPEG 4 committee as one of the eight functionalities to be addressed in the near future. In this paper, we will first review the technical results using temporal scalability (disparity analysis) in MPEG-2 as pioneering by [9] and [10]. Based on temporal scalability, the concept is further generalized to affine transformation to consider the deformation and foreshortening due to the change of view point. Estimation of the affine parameters is crucial for the performance of the estimator. In this paper we propose a novel technique to find a convergent solution which results in the least mean square errors. Our result shows that about 40 percent of the macroblocks in a picture has benefited by using the affine transformation. In our approach, the additional computational complexity is minimal since a pyramidal scheme is used. In one of our experiments, only four interations are necessary to find a convergent solution. The improvement in prediction gain is found to be around 0.77 dB.