Research On Super Resolution Reconstruction From Image Sequences

Super resolution reconstruction from images or video sequences is an active topic in image processing. The super resolution reconstruction refers to restoring a high-resolution image from sequences of low-resolution images degraded by warping, blurring and aliasing. Compared to traditional interpolation reconstruction method, super resolution method can reconstruct a more detailed, clear high-resolution image from several low-resolution frames. So it has been applied on many areas, such as satellite imaging, medical imaging, video surveillance etc. In this paper, the super resolution reconstruction algorithm in space domain has been researched. The research include image registration algorithm, POCS super resolution reconstruction algorithm and MAP super resolution reconstruction algorithm.Image registration is the basic of super resolution reconstruction and its purpose is to obtain the motion vector field. Three image registration algorithm has been proposed in this paper, block matching algorithm, image registration based on Harris corner points detection and optical flow algorithm. The results shows that the algorithm can satisfy the requirement of super resolution.On the aspect of POCS reconstruction algorithm, this paper studies the theories of POCS super resolution method and proposes the realization of this algorithm. For the purpose of reducing noise of reconstructed high resolution image, a super resolution reconstruction method for video images based on POCS algorithm has been proposed. First, the original high-resolution image estimation has been instead of bicubic interpolation image. Second, the problem of edge preservation in reconstruction has been solved by edge detection and PSF modification. Experimental results show that the high-resolution image reconstructed not only increases the resolution but also preserves the details. On the aspect of MAP reconstruction algorithm, this paper studies the theories of MAP algorithm and realize the algorithm in MATLAB. In order to reduce the complexity of MAP algorithm and increase the convergence speed, an algorithm of optimizing the iteration step has been proposed. Experimental results shows that this algorithm can improve the reconstructed high resolution image's quality compared to other interpolation method.