Research For Super-resolution Reconstruction Technologies Of Image Based On POCS Algorithm

In practice, it is very necessary to obtain images with higher resolution, better quality. For example, the military remote-sensing detection, vehicle license plate recognition, lesion examine with medical images. Therefore, improving the image resolution is indispensable in various applications. High-resolution image can be obtained mainly through the following ways: increasing the performance of imaging system(enlarge the focal length of camera, increasing the sampling density of sensor array) to obtain high resolution image directly, but this often brings cost increment and volume enlargement of airborne optical platform which is not suitable for airborne equipment miniaturization; using super-resolution reconstruction to obtain high-resolution image indirectly, which is effective and simple. And it is the most effective way to improve the image resolution.With continuous development of super-resolution reconstruction technology, a number of methods have been proposed. All the methods can be classified into two categories, frequency domain and the space domain. Frequency domain approach is mainly for the global translation between images, but there is no major breakthrough. Space domain approach is based on the general model, involving global and local motion, optical blur, motion blur, point spread function, non-ideal sampling and so on. Space domain approach has better adaptability and reconstruction effect. Space domain approach is the main direction of the super-resolution reconstruction research. In this paper, POCS algorithm of space domain approach is used for super-resolution reconstruction. POCS algorithm is the most promising algorithm for super-resolution reconstruction with its intuition, effectiveness, practical observation model and flexible prior information.In this paper, we study the development of image super-resolution reconstruction methods, and compare the methods in common use. Image registration takes important part in the super-resolution reconstruction. Therefore, the paper elaborate the classification of image registration algorithm and the complex matching algorithm with Fourier_Mellin algorithm, keren algorithm, and matrix DFT to get sub-pixel registration accuracy. Then we test the motion estimation accuracy with experiments.Finally, we study the principle and realization of POCS algorithm. We propose gradient-based interpolation algorithm to obtain the initial estimate of high-resolution image and modify PSF centered at an edge pixel during process of simulation-correction iteration. Experiments show that the method can get better image quality; enhance the edge information of reconstructed high-resolution image significantly.