| The emergence of compressed sensing theory breaks through the limitation of traditional Nyquist sampling law,making the sparse or compressible signals can be accurately reconstructed by collecting a small amount of data,when the sampling rate is less than twice the bandwidth.However,when dealing with two-dimensional image signals,it is necessary to generate and store a large-scale observation matrix to complete the observation and reconstruction of the whole image.The performance and efficiency of the algorithm are greatly reduced.The theory of block compressed sensing overcomes the disadvantages of excessive memory usage and long calculation time.However,in the classical block compressed sensing,the sampling rate of each image block is consistent,and the sampling rate adaptive allocation is not performed according to the image block texture characteristics.Focusing on this problem,through studying and researching compressed sensing theory and existing mature algorithms,this paper has done the following work and improvements on related theories and traditional block compressed sensing:(1)In this paper,the traditional image direct compressed sensing algorithm is studied,and the reason why it is difficult to get practical application is verified by simulation experiments.To overcome the above problems to some extent,this paper studies the influence of block size,sparse basis and reconstruction algorithm on the performance of block compressed sensing algorithm.The influence factor combination of 32×32+DWT+SPL(block size 32×32,Discrete Wavelet Transform,Smooth Projection Landweber reconstruction algorithm)is proposed.The simulation results show that the proposed method has better reconstruction performance in the fixed sampling rate block compressed sensing algorithm.(2)Since the block compressed sensing algorithm with fixed sampling rate uses the same sampling rate for all image blocks,which results in waste of sampling resources and block effect.The quality of reconstructed image needs to be improved.According to the difference of one-dimensional gray-scale entropy information contained in each image block,this paper divides the low-frequency pre-estimated image into blocks and distributes them according to the characteristic information of each block to the corresponding adaptive sampling rate.Then,each image block in the high frequency sub-band is compressed and sampled by using the adaptive sampling rate.Finally,the reconstructed image in the high frequency sub-band is superimposed with the blind deconvolved low frequency sub-band,which realizes the adaptive block compressed sensing algorithm based on one-dimensional gray-scale entropy and blind deconvolution.Compared with the traditional fixed sampling rate block compressed sensing algorithm,the proposed algorithm has better overall visual effect and reconstruction quality.(3)Multi-scale block compressed sensing algorithm assigns a suitable sampling rate to image wavelet coefficients at different scales,which plays a good role in improving the quality of image reconstruction.However,the algorithm still uses uniform sampling rate for all sub-blocks in the same scale,which wastes the prior knowledge contained in low frequency sub-band and does not make full use of the limited sampling resources.Therefore,an adaptive multi-scale block compressed sensing algorithm based on visual saliency and orientation characteristics is proposed.By studying the visual saliency and orientation characteristics of each image block in low frequency sub-band pre-estimation image,the total sampling rate is adaptively allocated to sub-blocks of each sub-band of each layer according to the characteristic information and wavelet structure characteristics for observation,and reconstructed by the improved single-time Wiener filtering SPL algorithm.The simulation results show that the performance of the proposed algorithm is better than the traditional multi-scale block compressed sensing and sampling rate adaptive block compressed sensing. |
PDF Full Text Request