Fusion methods for side information generation in multi-view distributed video coding systems

摘要

The generation of the side information is an important part in the design of a distributed video coding (DVC) system as it directly relates to the system's rate distortion performance. In multi-view systems spatial/view inter-camera correlations can be exploited alongside temporal/motion intra camera ones for generating the side information as accurately as possible. In this paper, algorithms for fusing multiple such side information estimates, generated with temporal and view interpolation methods, are proposed. Two algorithms are suggested along with weighted fusion schemes for combining multiple methods. The results presented indicate performance improvements of up to 2 dB compared to exiting fusion approaches.