Compressed Sensing And Its Application In Infrared Image Processing

The acquisition of high resolution infrared image has been a key problem in many military and civil application systems with the growing demands of high quality infrared images. The high resolution infrared imager has been paid much attention and gained many good performances in the past decades, but which can not meet the demands yet. The presentation of the compressed sensing theory makes it possible to get high resolution image with low resolution infrared sensors. As the theory foundation, compressed sensing theory breaks the restriction of the sensing systems based on Shannon sampling theory, which can directly obtain the sparse representation of the targets, then less data obtained can be used to construct high resolution signal. The compressed sensing theory provides the foundation for high-performance, low-cost sensor design, and is a breakthrough in the field of sensors.To improve the performance of infrared imaging system, the theory of compressed sensing is studied in this paper, including the sparse representation method, design methods of sensing matrix and high-resolution signal reconstruction method. An adaptive sparse sampling method is presented to sparse the target signal with unknown sparsity. By evaluating the recovery effect with different sparsity, the best one was selected as the target sparsity. As for the sensing matrix, a sub-Gaussian hybrid random matrix was constructed to project the original signal to a sparse space, and then the compressed sensing was implemented on the target. The target can be adaptively projected by segments with the hybrid matrix, and the structure characteristics can be grasped more sufficiently, which benefits further processing, such as constructing the auto target recognition (ATR) systems. To imply the high resolution target reconstruction, a block-compressed sensing method was presented by dividing the image into several sub-blocks with the same size, by which higher quality target can be recovered, and the process speed can be improved also.The simulation was carried on several typical infrared scenes, such as battle field, ship, airplane and electric power equipments, etc. The results show that one can get sparse representation of the interesting target with optimized sparsity, and the recovery image has high PSNR (peak signal to noise ratio) and higher quality.