Research On High Performance Decoding Algorithm Of Compressed Sensing Video Encoding

With the rapid development of fifth-generation mobile communications(5G),Internet of Vehicles,and drone interaction technologies,how to effectively transmit signals in real-time at asymmetric transmission and reception resources has become a research hotspot.As the most intuitive way of information transmission,video images are widely used in various environments,such as the sky,the earth and the sea.The essence of video is high-speed continuous multi-frame images.Continuous sampling will generate huge data,which will bring unbearable load pressure on transmission equipment and channel bandwidth.With the improvement of video resolution requirements,there may even be data overflow.At present,mainstream video encoding standard perform motion estimation on the sampling side,and perform optimal block matching based on the redundancy of video signal.At the same time,a lot of computation is introduced into the encoding side.So these coding standard are limited in scenarios with limited coding processing power.Therefore,designing a video transmission system that meets both low coding complexity and high decoding quality is of great significance to emerging industries,such as the multimedia Internet of Things.For nodes with limited coding and computing capabilities,the distributed compressive video sensing(DCVS)is been proposed.Based on the theory of compressed sensing,the amount of calculation from the traditional encoding end is further transferred to the decoding end.The system overall adopts the scheme of independent encoding and joint decoding.The video sequence is directly divided into GOP groups at encoder.Each frame is sampled independently,eliminating the need for block matching and other steps,simplifying the coding process as much as possible.The core work of the system focuses on the decoding end with sufficient computing power.Based on DCVS system,a high-performance reconstruction scheme is designed in this paper,which mainly improves the low-compression non-key frames that are difficult to directly reconstruct.Based on the latest coding standard VVC/H.266,a side information generation algorithm is proposed.When reconstructing non-key frames,the decoder uses the structural similarity of closed images in the time domain,generate relatively accurate side information to assist joint decoding process.Based on the current compressed sensing l1-norm minimization reconstruction idea,we introduce side information for joint decoding of dual information sources,and dynamically adjusts the reliance coefficient to match the relative accuracy of side information and compression value.Through simulation on multiple videos with different resolutions and varying degrees,the effectiveness of the side information generation algorithm VVC-ME and dynamic reconstruction scheme is verified,bring significantly improvement to non-key frames.In order to further improve the coding and decoding efficiency of video transmission,a high-efficiency DCVS system is constructed.The scene decision formula is introduced into the image grouping process,and the GOP size is dynamically adjusted according to the changes of the video itself.Aiming at the problem of insufficient timeliness of image reconstruction algorithms,neur al networks are integrated to speed up.A BMRCNN network is proposed for key frame images,and a combinatorial reconstruction of CNN network is proposed for non-key frames.The pre-reconstructed images are sending into the low-resolution-high-resolution mapping network,realized improvement in decoding quality and time.In addition,the l1-l1 minimization decoding expansion model for non-keyframes is proposed,the input vector y is rapidly approximated to the original sparse signal s.The simulation results show that the proposed scheme significantly improves the coding and decoding efficiency,and effectively reduces the decoding time under the same decoding quality.The new system meets the video transmission requirements in different scenarios.