Noise-resilient distributed compressed video sensing using side-information-based belief propagation

With the booming of energy-limited video devices ranging from visual sensors to camera cell phones, distributed compressed video sensing (DCVS) has developed as one of the efficient solutions that guarantee low complexity video encoding as well as acceptable recovery quality. In this paper, a noise-resilient DCVS scheme is proposed using side-information-based belief propagation (SI-BP). The video frames are encoded independently to a small number of measurements and recovered jointly based on the distributed video coding structure. The major contribution is the joint recovery algorithm SI-BP where the side information from the recovered frames is fully exploited to improve the original CSBP algorithm in both initialization and stopping criteria. As the estimated probability distribution of the DCT coefficients is introduced in the soft decision SI-BP iterations, our scheme is noise-resilient especially with low SNR. This advantage is verified by the simulation results on several video sequences.