Compressed Sensing Reconstruction Via Curvelet Redundant Dictionary And Immune Clone Optimization

In the past, Nyquist Sampling theorem has been always dominated the field of signal processing. With the increasing demand for information,more and more wide signal bandwidth,the requirements of the sampling rate and processing speed in accessing to information are getting higher and higher,which will undoubtedly bring significant challenge to the Nyquist Sampling theorem. In recent years,a novel signal processing theory named " Compressed Sensing " rises, which breaks through the bottleneck of traditional sampling theorem, leads to changes in the field of signal processing,and also sets off a wave of research enthusiasm.Compressed Sensing is a new theoretical framework which sampling and compression is achieved at the same time,it samples at the rate well below the Nyquist frequency, with non-adaptive linear measurements to maintain the structure of original signal, and then accurately reconstruct the original signal by sloving a nonlinear optimization problem.Compressed Sensing theory contains three elements:sparse, non-adaptive measurements, nonlinear optimization reconstruction,in which sparse or compressible of signal is a premise and important base of compressedsensing.and non-adaptive measurements is the key of compressedsensing, also nonlinear optimization is a means of compressedsensing reconstruction. This papper is mainly dedicated to the study of reconstruction algorithm, directly starting from l0norm optimization, solving image non-convex optimization reconstruction problem via immunity clone optimization method. Innovative work of this papper is as follows:Fisrt, considering the redundancy and similarity of the image block, using unsupervised affine propagation clustering algorithm is to find similar measurements in all of measurements in decoder.For similar measurements of a class, a reasonable affinity function is designed, and optimal common base atoms of sparse representation based on Curvelet redundant dictionary for these similar measurements in a class corresponding to image blocks are learned by immue clone optimization method,and thus image reconstruction in the sense of l0norm is achieved. Simulation results show that compared to learning a single optimal base atom for a single measurement corresponding to an image block by immue clone optimization method, it can greatly reduce the probability of learnning false.Second, to avoid the choose of relaxation factor in l0norm unconstrained problem of compressedsensing, in this papper, l0norm unconstrained problem is divided into reconstruction error and sparsity to alternative learning, thus the choice of relaxation factor is avoided, and also slove the defect of fixed sparsity in orthogonal matching pursuit algorithm.Finally, the study only for the common of measurements is not enough, its own characteristics is also important. Therefore, in this papper, these two advantages are combined by filtering and projection onto convex set operation,then the image operated by filtering and projection onto convex set is considered as a priori knowledge injected into the evolution of populations to speed up the population towards the search direction of the optimal solution, and thus an image with good visual effects is obtained, also the block effect is removed. Simulation results demonstrate the effectiveness of the proposed algorithm in this papper, experimental data show that the proposed algorithm is better than some reconstruction algorithm in visual effects and peak signal to noise ratio.