Visual Attention Model Aided Non-Uniform Asymmetric Coding of Stereoscopic Video

摘要

Owing to its wide-interoperability, stereoscopic 3D video format in High Definition (HD) is a popular choice for 3D entertainment media distribution. However, the delivery over bandwidth constrained networks exhibits challenges in terms of intermittent congestions in the network traffic, which enforce the delivery system to perform perception-aware coding to save bandwidth. In the scope of stereoscopic 3D video, asymmetric quality adaptation has proved to be an effective method in terms of maintaining the perceived quality while reducing the required transmission bandwidth. On the other hand, Region-Of-Interest (ROI) based coding in accordance with the visual attention cues, which offers non-uniform quality assignment to regions of different saliency levels has not been widely studied in combination with asymmetric coding of stereoscopic 3D video. In this work, the effectiveness of using visual attention aided non-uniform asymmetric 3D video coding is explored. The importance of incorporating compression artefacts in the formulation of visual attention model is also revealed. The discussions in this paper are based on a comprehensive subjective test with 8 stereoscopic video sequences of different spatial and temporal characteristics at different conditions.