Research On Image Reconstruction Based On Compression Sensing

Because of the explosive growth of demand for high-definition images,compressed sensing(CS)theory has been widely used,which can effectively reduce the requirements of sampling rate and communication bandwidth.The reconstruction algorithm is the core part of CS theory to solve the sparsest representation of under-sampled signals under given sparse basis.The greedy algorithm in CS reconstruction algorithm is simple to implement and easy to customize hardware.However,in some complex scenes,it exposes the shortcomings of slow reconstruction speed and low accuracy.Given these problems,the present research studies CS reconstruction algorithms and improves them.The results are as follows:When dealing with large-scale matrix generated from high-definition images,the least square method used in subspace pursuit(SP)has high computational complexity,which seriously limits the speed of reconstruction algorithm.An improved subspace pursuit via conjugate gradient method reconstruction algorithm(SPCGM)is designed to solve these problems.Among the classical iterative least square methods,conjugate gradient method shows obvious speed advantage.Considering that each iteration of SP algorithm can calculate a complete estimate independently,the improved algorithm uses the least square approach of iterative solution.It uses the signal estimate of the previous round as the initial value to accelerate the current round.Experimental results show that the improved algorithm can achieve faster computing speed than SP algorithm without reducing the reconstruction accuracy.To solve the problem that the traditional greedy reconstruction algorithm of compressed sensing has poor reconstruction accuracy in processing CT images with high contrast and rich texture,an insatiable pursuit via rectification and conjugate gradient method(GPRCGM)based on reshaping and conjugate gradient iteration is designed.The algorithm is divided into pre-screening part and reshaping part.In the pre-screening part,more candidate atoms are added by threshold to enhance the hit rate of atoms.In the reshaping part,candidate atoms are screened and the conjugate gradient method is used to accelerate the iteration.The latter can make up for the adverse effects of the former in extreme cases.Experimental results show that GPRCGM algorithm's success rate and peak signal-to-noise ratio of the algorithm is improved,especially when the compression rate is low.