Scalable Saliency-Aware Distributed Compressive Video Sensing

Abstract

Distributed compressive video sensing (DCVS) is an emerging low-complexity video coding framework which integrates the merits of distributed video coding (DVC) and compressive sensing (CS). Because the human visual system (HVS) is the ultimate receiver of visual signals, we aim to improve the perceptual rate-distortion performance of DCVS by designing a novel scalable saliency-aware DCVS codec. Firstly, we perform saliency estimation in the the side information (SI) frame generated at the decoder side and adaptively control the size of region-of-interest (ROI) according to the measurements budget by applying a saliency guided foveation model. Subsequently, based on online estimation of the correlation noise between a non-key frame and its SI, we develop a saliency-aware block compressive sensing scheme to more accurately reconstruct the ROI of each non-key frame. The obtained experimental results reveal that our DCVS codec outperforms the legacy DCVS codecs in terms of the perceptual rate-distortion performance.