Compressed Sensing Video Reconstruction Algorithm Based On Autoregressive Model

Compressed sensing(CS)is a novel signal sampling theory with the unique ability of compressing a signal during the process of sampling,which breaks the constraint of Nyquist sampling theorem and reduces the amount of data in the process of signal acquisition,storage and transmission.It has already absorbed the concerns of lots of scholars and there is no doubt that this theory can be applied to the field of video and image,such as scene monitoring and target tracking.The reconstruction algorithm is the most important part in compressed sensing so this thesis puts an emphasis on the research of video reconstruction algorithm based on compressed sensing.Aiming at the long time consumption and barely satisfactory performance of compressed sensing video reconstruction algorithm based on autoregressive,an improved algorithm is proposed in this thesis under the frame combining the prediction and residual compensation.The main contents are as follow:(1)This paper improves the prediction object.Firstly,explore the CS measurements correlation of adjacent video frames;then,reconstruct the CS measurement of current frame based on this theory;finally,introduce the CS measurement of current frame which is regarded as a regularization item into the reconstruction algorithm to obtain the prediction of current frame.This prediction method can clearly reduce the prediction elements and improve the reconstruction time.(2)This paper improves the prediction method.Firstly,the CS measurements of key frame and non-key frames in one GOP(Group of Picture)are regarded as the original dataset and the autoregressive(AR)coefficient can be obtained based on the least square method;secondly,the variance of autoregressive residue and Final Prediction Error(FPE)function are regarded as the criterion to select the AR order;finally,the CS measurements in one GOP which are most related to the CS measurement of current frame are regarded as AR support region.When the AR coefficient,AR order and AR support region are acquired,the estimation of CS measurement of current frame can be finished.The method mostly adopts linear operation and adequately considers the inter-frame correlation,so the reconstruction time and performance are improved.(3)It is affirmed that residual compensation can improve the reconstruction performance effectively.Firstly,explore the difference value between the current frame and the prediction of the current frame to obtain the residual frame;then the prediction of the residual frame is achieved by measuring and reconstructing;finally,the reconstructed of current frame is acquired by adding the prediction of residual frame and current frame.The improved approach can adequately excavate the correlation of video frames so the approximation of frames can be obtained with lesser information and time.More importantly,this algorithm can avoid motion estimation(ME)and motion compensation(MC)which has complex calculations.The experimental results show that this scheme has preferable reconstruction time for both slow and drastic detail variation video sequence while keeping a certain performance.