Slice loss concealment using bridge pictures in scalable video coding

Abstract

The scalable extension of H.264/AVC (scalable video coding, SVC) provides spatial, temporal and/ or quality scalability with a coding performance that is comparable to that of H.264/AVC. As in the transmission of all types of compressed video, packet losses or errors may result in the degradation of SVC reconstructed video quality. A concealment method for SVC is proposed that uses the motion vectors (MVs) of higher temporal level pictures (bridge pictures) to estimate the MVs of a lost slice. The proposed scheme is tested on a spatial base layer and experimental results show that up to 2.2 dB performance improvement is obtained compared to the temporal direct method, which is a non-normative error concealment method of SVC.Introduction: The latest video CoDec H.264/AVC presents many new features that make it a very efficient coding standard [1]. scalable video coding (SVC) provides the flexibility needed by heterogeneous networks in addition to coding efficiency comparable to H.264/AVC [2]. However, losses and errors owing to communication channels still require precautions. Four error concealment methods for SVC are presented in [3]. One of the methods used for spatial base layer error concealment is the temporal direct (TD) method. In the TD method, each 4 × 4 sub-block (block) of the lost picture is assumed to be coded using the temporal direct mode of H.264/AVC. An improved TD error concealment method is presented in [4], where the MVs of each block are recovered from the MVs of the co-located blocks in the temporally subsequent and/or previous reference picture (RP).The TD and improved TD methods work assuming the motion among pictures and the RPs they are using is translational, which might be true for consecutive pictures. In hierarchical coding structure, pictures and the RPs they are using are close only for higher temporal levels (TL). Lower TL pictures have larger temporal distances (Tdist) with their RPs. So, a further improvement is added in [4] for the pictures having larger Tdists with their RPs. The MVs of a lost block are calculated according to the MVs of the collocated blocks in the left and right neighbouring B-pictures that belong to a higher TL.