Research On Auto-estimating Of Point Spread Function And Its Application In Super Resolution

Super resolution is a method which aims to obtain a high resolution image from arange of low resolution images, meanwhile, to improve the quality of original images. Itis widely used in satellite imaging, medical imaging, and video applications. And it isespecially useful in the situation where low quality imaging device is used.Reconstructing images in super resolution is to mine the complementary information inlow resolution images, and recover high frequency information which is lost duringdegradation. The degradation of image quality can be measured by degradation model.If the model is known, the reconstruction reduces to some simple calculations. However,in most cases, the degradation model is unknown, in which case, it can only rely onblind reconstruction to do image recovery. Therefore, blind reconstruction is importantin super resolution, and point spread function estimation is key procedure of blindreconstruction. Fergus-Miskin algorithm is one of the best algorithms on point spreadfunction estimation, and it can achieve good performance in deblurring images.This paper analyzed point spread function estimation in image recovery and superresolution. With or without standard pictures, we compared vary measures to choose theappropriate one for image recovering and super resolution which used in our improvedalgorithm as well.We researched Fergus-Miskin algorithm. Against the poor robust from theimproper patches for point spread function estimation and the low efficiency fromhuman-computer choosing patches, we posed an algorithm for choosing the best patchesfor estimation, which is fast and accuracy.Against the disadvantage of fixing the best maximum size of point spread functionmanually, which can not insure the quality and cause much time wasting, we posed analgorithm for fixing the size automatically.Based on the research findings above, this paper realized an algorithm onauto-estimation point spread function, and we applied it in super resolution. Theexperimentation indicated that our advanced algorithm improved the performance ofsuper resolution.