Research On Light Field Acquisition And Reconstruction Based On Compressive Sensing

Compressive sensing(CS)was used to improve the efficiency of image sequence acquisition and the accuracy of reconstruction.Due to the reconstruction of the 3D model by using the light field requires capture and reconstruct massive images.Existed methods have some disadvantages such as low acquisition efficiency,insufficient reconstruction accuracy and long transmission delay.Using compressive sensing can effectively reduce the dimensionality of the signal and restore the original signal with high probability.Therefore in this paper we proposed block compressive method based on structured random matrix and distributed compressive sensing method based on joint sparse model capture and reconstructe the light field.Block compressive sensing based on structured random matrix(BCS-SPL-SRM)was used to segment large size light field images into small blocks for sparsity,transmission and reconstruction.In this method,the measurement matrix was improved on the basis of the BCS-SPL method.The SRM which is suitable for the light field was used as the measurement matrix to measure the image.In our study,three groups of light field image sequences were reconstructed.Reconstructed images gotten by BCS-SPL-SRM algorithm had higher PSNR value than other algorithms.The time of image reconstruction was shortened by 3.61 s over other algorithms.In addition the algorithm had faster convergence speed than that of other algorithms.Distributed compressive sensing method based on joint sparse model take advantage of a large number of overlapping information between light field image sequences and focus on the differences between images.In this study,two kinds of compressive sensing algorithms which are suitable for light field images were proposed: DCS-VS and DCS-SI.DCS-VS algorithm was used to compress and reconstruct vertical slices of 3D image matrix constituted by image sequences.It was proved that the sparsity of vertical slice is lower than that of original images,and it would have higher efficiency in signal acquisition,transmission and reconstruction.The DCS-SI algorithm was used to divided the image sequences into the center view image and the edge view image.And the edge view image was fused with the corresponding side information during reconstruction.It reduced the transmission of the original signal.In the experiment,DCS-VS algorithm was used to reconstruct the image sequences.Compared with other algorithms,the reconstructed image had high PSNR value of 2.01 d B.DCS-SI algorithm was used to reconstruct the image sequence,and the reconstruction result had better subjective effect.Combining the above two algorithms,array light field acquisition and reconstruction based on compressive sensing(ALF-CS)was proposed.This algorithm was used in capturing and reconstructing the light field.And block compressive sensing was applied to single image acquisition and reconstruction.Distributed compressed sensing was applied to image sequences acquisition and reconstruction.In the experiment,a complete 3D model reconstruction process was carried out by using the image sequence captured and reconstructed by ALF-CS.The results showed that the image sequences reconstructed by ALF-CS could improve the acquisition efficiency of the light field,and the reconstructed image sequences could be used to reconstruct the complete 3D model.