Blind deconvolution of dynamic-scene image sequences

This work devotes to the image sequences deconvolution problem that restores each clear image from a sequence of blurred and noisy measurements with little blur prior. In this paper, a simple but effective sequences blind deconvolution (SBD) method is developed in a Bayesian framework. The method alternately estimates each image and blur, only using the measurement of current frame and the estimations of the former frame, thus it is very memory-saving. Compared with the relative OBD method and the famous RLBD method, it achieves superior performance in almost all case studies. Experiments performed on both synthetic and actual astronomical images, without and with noise, show that the proposed method yields good results on all test data.