Compressed Sensing Image Reconstruction Based On Ridgelet Redundant Dictionary And Multi-Objective Genetic Optimization

Compressed Sensing(Compressed Sensing, CS) theory created a new signal processing model, which realized the combination of signal sampling and data compression. That is it can sample signal and compress data at the same time. Far below the Nyquist sampling rate of signal sampling, it saves resources and bandwidth, alleviates the higher pressure that the bandwidth of the signal bring to the acquisition device. In this new signal processing mode,the reconstruction of the signal quality is closely related to observation matrix, sparse representation and reconstruction algorithm.Non-convex compressed sensing image reconstruction algorithm Based on a complete dictionary and evolutionary computation has been proposed, which under a fixed sparsity,use the evolutionary computation method to study the optimal atoms combination of complete dictionary, finally use the optimal atoms combination to reconstruct image.However, this kind of algorithm needs to know sparsity in advance; In fact, the sparsity of a real signal in general is unknown. Thus, the algorithms exists a poor performance of image reconstruction problem due to the inaccurate forecasts of sparsity. According to the above problem, this paper proposes a compressed sensing image reconstruction method based on ridgelet redundant dictionary and multi-objective genetic optimization. In this paper, the main work is as follows:Under the framework of non-convex compression based on the strategy of over complete dictionary and image block strategy, the sparsity combined with atoms combination as two optimization objectives are optimized at the same time. Use the reconstruction model which from the compression observations of image block to estimate the image block structure, the image block to be reconstructed is divided into: smooth image block, single direction image block and multiple directions image block. In constructing based on multi-objective genetic optimization non-convex compressed sensing image reconstruction algorithm, designed a variable length coding with sparsity and atoms combination for the two goals; against the variable length coding of two targets proposed a variable length coding genes crossover operation based on the different gene position, a new individual insert based on the prior information and other multipoint mutation multi-objective evolutionary operation.In order to reduce the time needed for reconstruction, the image block of different structural types is clustered according to the observation vector, so as to obtain the best combination of all the image blocks by finding the optimal combination of atoms at a timed. Experimental results show the feasibility and effectiveness of the proposed algorithm. At the same time the following conclusions are drawn from the experimental results: when the image block with different structures is reconstructed, the sparsity is different and the image block with same structures and different cluster is reconstructed,the sparsity is also different. Finally, some key parameters of the algorithm are discussed and analyzed, and the range of value is given.