Video Compressive Sensing Using Residual Spatial Domain Based on Mutual Information for Vehicle Video Compressive Algorithm

Abstract

AbstractVehicle video terminal devices have weak computing and storage ability. A simple, reliable, and effective data sampling and compression algorithm is urgently needed in the mobile environment with the harsh environment. According to compressive sensing (CS) theory, when image data had highly sparse representation, image data would have a highly compressive rate after sampling using a measurement matrix. Image data sampling and compression can be completed simultaneously. Meanwhile, the image that is divided to blocks can improve to process speed. Therefore, in this paper, an improved block-based CS (BCS) is used for vehicle video residual compression. Each vehicle video frame is divided into blocks and residual is sparse in the spatial domain, but we observed residual images have a large amount of redundant data because of illumination and bumps during vehicle motion, and two blocks of adjacent frames have similar texture structure. In order to expand the range of sparse representation of the residual spatial domain, we propose MIBCS method that uses residual energy and mutual information (MI) to judge and process texture structure for two adjacent non-key frames’ blocks. Finally, we use SPL to reconstruct residual in the spatial domain. MIBCS has been tested in a bus video. The experimental results show this method can reduce the residual data volume by nearly half without significantly reducing the video quality. This method is strong and robust because reconstruction quality is weakly affected by block size.KeywordsVideo Compressed SensingMutual InformationResidual ReconstructionSparse RepresentationResidual Spatial Domain