Research On The Hypothesis Set Optimization-Based Compressed Video Sensing

Compressed Sensing(CS) does the sampling and compression procedures of data in one step, and the high efficiency compression and very simple operation make it particularly suitable for processing multimedia video data in the encoding resource constrained scene. Multi-hypothesis(MH) prediction technique can obtain a high quality performance. In this paper, we propose a scheme for distributed compressive video sensing(DCVS) based on hypothesis set optimization techniques which aims at achieving high reconstruction quality and fast reconstruction speed at low sampling rate.The proposed scheme enhances the reconstruction quality and reconstruction speed by optimizing the hypothesis set, improving the judgment formula of hybrid hypothesis prediction(HHP) model and the optimization problem. Among them, 1) the optimization of hypothesis set is implemented by techniques of superb hypotheses selection-based HHP(Se-HHP) and hypothesis set update-based HHP(Up-HHP). In low sampling rate, we adopt Se-HHP technique to simplify the set and decrease the computational complexity, then enhance both the reconstruction quality and reconstructed speed; in high sampling rate, the set is optimized and updated through Up-HHP technique and then the reconstruction quality is improved. 2) The improvement of the judgment formula of HHP model is carried out through averaging the Euclidean distances to each measurement, which made the adaptive judgment of the HHP model under the fixed threshold realized. 3) For the optimization problem, we combine the 1 and the 2 norm, weight both of them according to the distance vector to form the adaptive distance-weighted elastic net technique(ADWEN) penalty term, and avoid the shortcomings arising from a single use of the 1 or 2 norm. The simulation results show that our proposal outperforms the start-of-the-art schemes which do not use the hypothesis set optimization techniques in low sampling rate.