Researches On Compressed Sensing-based Multimedia Communications

With the development of mobile wireless communications and multimedia signal sensor techniques,mobile intelligent devices and multimedia sensor networks become popular in our social life.New multimedia application scenarios with resource-constrained encoder are emerging due to the development of mobile devices-based multimedia services.The traditional multimedia encoding standard is not conducive to uplink stream services in the new multimedia application scenarios due to its complexity encoding processing.Compressed sensing as a new signal sampling and compression technology provides the possibility to reduce the complexity and improve the coding efficiency of encoders.When a signal is sparse in some domains,compressed sensing can recover the signal from a small number of measurements.Signal sampling and compression are realized simultaneously.It reduces the sampling rate and the pressure of information acquisition.The research of image and video codec systems based on compressed sensing is of great significance to promote the further development of multimedia communication service in the new scene.However,compressed sensing-based image and video coding schemes are unsatisfactory in recovery quality compared to the traditional standards.Moreover,combined with the specific application scenarios,the coding efficiency of the encoder has room for further improvement.This paper make a deep research in the spatial and temporal features of multimedia signals.A series of improvements in compressed sensing based video encoding scheme,image recovery algorithm,and video decoding system are proposed.The main contributions of this paper are as follows.1)This paper proposes a residual transformation-based multi-level compressed sensing image reconstruction algorithm.For image recovery,lack of prior knowledge and the varying content characteristics of input images make the recovery quality unsatisfactory.The applicability of current image compressed sensing recovery algorithms is bad.To solve the above problems,a new multi-level image reconstruction system is proposed.The image reconstruction process is divided into multiple levels.Based on the changes in the content and structural characteristics of the input images in different recovery levels,a new compressed sensing reconstruction model with adaptive constraints is proposed.According to the image features in different recovery levels,different constraint terms are used.Both the algorithm applicability and the recovery quality are improved effectively in the new proposal.2)A new reconstruction scheme for distributed compressed video sensing system is proposed.In current distributed compressed video sensing systems,the recovery quality is low.Significant fluctuation of recovery quality occurs in the recovery of non-key frames.To solve these problems,the improved recovery schemes for key frames and non-key frames are given,respectively.For key frames,a novel secondary reconstruction scheme is proposed by combining the total variation and multihypothesis algorithms.The adjacent non-key frames is removed from the recovery of key frames.It effectively improves the decoding efficiency and the recovery quality of key frames.For non-key frames,the adjacent non-key frames are first introduced into the recovery of non-key frames as the auxiliary reference frames by using cross recovery method.It effectively reduces the fluctuation of the recovery quality of non-key frames.Moreover,by adaptively selecting the weight prediction models,A good balance between the prediction accuracy and computational complexity is obtained.The proposed reconstruction system greatly improves the recovery performance of distributed compressed video sensing systems.3)A high-efficiency terminal-to-cloud uplink compressed sensing video codec system is proposed.This paper proposes a series improvements in compressed sensing based video coding scheme,channel transmission,and video decoding scheme.For video coding scheme,a new skip block-based compressed sensing residual encoding scheme is proposed.It effectively removes the redundant image blocks and avoids unnecessary encoding computation.For time-varying channel states,this paper uses a channel state-based sampling rate adaptive selection scheme.The transmission rate is adaptively adjusted by matching suitable sampling rates for different channel states.The link congestion and transmission interruption at the low channel quality are avoided effectively.For the cloud decoding scheme,considering the strong computing capabilities of the cloud server,a new reconstruction scheme based on multiple reference frames and local secondary recovery is proposed for the reconstruction of non-key frames.The recovery quality of non-key frames is improved significantly.The quality fluctuation of non-key frames is effectively suppressed.