The Fast Methods On Infrared Image Compressive Sensing

In the multi-sensor network system, due to the big image data collected by infrared optical sensors, the processing tasks, such as transmission and storage, have been becoming more and more difficult. Compressive sensing(CS) techniques are attracting more attention of researchers, since it has the advantages of high-efficient compression and perfect reconstruction of images. However, real-time image CS would be quite difficult. In order to solve this problem, this paper investigates the following researches on infrared images:1) In order to lift the speed of compressive sampling, an approach on fast CS based on sub-sampled IFFT is presented. On one hand, for the fast computation speed of sub-sampled FFT, the paper presents the sub-sampled IFFT; On the other hand, for the existing problems such as long time of reconstruction of high-dimensional observation matrix and slow speed of compressed sampling, the method on sub-sampled inverse fast Fourier transform based on compressive sampling(SSIFFT) is presented, using the CS theory of random filters as a basis. SSIFFT method greatly reduces the computation complexity and improves the processing speed by employing the property that sub-sampled in frequency domain is equivalent to down-sampling in time domain. The theoretical analysis and simulation results show that the SSIFFT method is able to quickly and efficiently obtain the compressive measurements.2) To improve the speed of reconstruct, subspace OMP algorithms with the knowledge of target attribute is presented. In order to take full advantages of the subspace OMP algorithms such as low computation complexity, simple structure, implement and widely application in practical system, combing with the existing problem of low speed caused by the improper subspace selection and setting in this type of algorithms, a fixed subspace OMP algorithm is presented with the knowledge of infrared small target sizes on the one hand; based on that, according to the characteristic of infrared small target imaging and employing the target sparsity in space domain and the knowledge of target size and the coherence of target images, an adaptive subspace OMP algorithm is presented to solve the subspace adaptation problem. Simulation results verify that the two proposed algorithms have the advantages of quicker processing speed and higher reconstruction precision compared to the subspace OMP algorithms.To sum up in conclusion, this paper presents the fast methods on infrared image compressive sensing, these lift the speed of compressive sampling and reconstruct, and provide the technique support for the multi-sensor network system in the practical application.