Interframe Coding Based On Distributed Compressed Sensing

During the past few years, video codec technologies have made the rapiddevelopment. The international standard algorithm H.264/AVC with theexcellent performance in coding efficiency, Specifically, takes a significantposition on video applications, such as video surveillance、video sharing sites、video conference、digital TV. All that provide people with a clear and fluentpicture, also extreme to meet the people’s sensory experience. However, withhigh-definition and networking, we poses two requirements about video codecsystem: on the one hand, the video codec system can support the collection、storage and transmission of huge amounts of data; on the other hand, the videostream which video codec system outputs must be more suitable for the networkenvironment. H.264/AVC is a hybrid coding framework based on prediction.That largely reduces the inter-frame redundancy and obtain the high codingefficiency, but it is vulnerable to channels errors due to the excessiveexploitation of dependences between the adjacent frames at the encoder side.Based on the framework of separate encoding and joint decoding,distributed compressive sensing (DCS) reduces the dependences on interframecorrelation and curb the error propagation in a certain extent. While it has bettererror-resilience because of the unique collection method and randonmeasurement. Therefore, we can integrate the respective characteristics of DCSand H.264/AVC to improve the error resilience of the inter-frame coding ofH.264/AVC. In summary, we propose an inter-frame coding based on DCS, andthen complete the work in three aspects:1. We propose an inter-frame coding framework integrating DCS andH.264/AVC: at the encoder, we still remain the motion estimation andsend motion vector to the decoder, and then we replace the residual withthe random measurement and send it to the decoder; At the decoder, wedo motion compensation to get the side information(SI). Then thereceived measurement are employed for joint CS decoding jointly with SI.2. In order to improve the performance, we also propose somemodifications: the measurement algorithm that removes the mean ofimage pixel in order to reduce the energy of measurement; thereasonable packaging method so that each data packet has theindependence which can reduce the dependencies between the datapackets; the residual reconstruction by making full use of SI; thetwo-step residual reconstruction to achieve a progressive improvementof reconstruction quality. Finally, the experimental results show that theproposed inter-frame coding is more robust against packet loss than thatof H.264/AVC in the environment of packet loss channel.3. Build an experimental environment: firstly, we find the program ofmotion estimation and get the predictive image and motion vector, at thesame time, we also get the number of coded bits; secondly, we achievethe program of BCS-SPL with C language; lastly, we modify the projectrtp_loss to do random packet loss experiments.