Compressive Sensing Image Reconstruction Based On Natural Computation

Compressive sensing (CS) is a novel theory which codes and decodes data signal. The CS points out that, if the signal is compressible or sparse in some linear basis, it can be reconstructed accurately from far fewer measurements than Nyquist sampling rate. Currently, image reconstruction algorithms convert the original multimodal optimization problem to sub-optimization problem. This paper starts from primitive problem and raises an image reconstruction algorithm to solve multimodal optimization problem.To solve the problem of multimodal optimization, we design a reconstructed model, and proposed our algorithm based on this model, which is one of my innovations, besides, we also design two populations, one is designed to study the positions of the wavelet coefficients, and the other is to solve the values of the coefficients. This paper also includes the following contents:(1) Combining the model we design and the natural computation with the advantage of searching the global optimization in multimodal optimization problem, we use the modified genetic algorithm (GA) and design the compressive sensing reconstructed algorithm, called ImSGA for short. This algorithm is designed to study the positions of the wavelet coefficients which need to reconstruct. According to the transmissibility among the different scales in wavelet domain, we use the low scales to conduct the high ones which need to reconstruct, in this way, we complete the initialization of the position coding population.(2) We use the concentration in wavelet domain to design the operators, such as extracting vaccine and injecting vaccine, which are used to adjust the position coding population, and then we combine ImSGA algorithm with the adjusted position coding position to study the positions of wavelet coefficients.(3) We use the GA algorithm to solve wavelet coefficients, and build the relevant algorithm, called ModGA for short. In the process of the algorithm, we use the optimal position coding matrix to construct the coefficients population and the studying process, so that the values of the coefficients can reflect the concentration in wavelet domain.The paper gives the reconstructed results, including the final reconstructed images, the position coding matrix images, and the images to show the coefficients distribution. At last, we analysis the data and visual effect, as well as the effect of the parameters, the results demonstrate that our algorithm has advantages to keep detail information and texture information when the values of parameters are small, and when the values are big, the smooth zone has a better visual effect.