Abstract: For multi-frame super resolution reconstruction technology, the perfect result is combing all the information contained in low resolution sequence.So, the difference information between the frames, namely, the non-redundant information, is very important.However, the frames of sequence has strong correlation in the spatial-temporal domain.The non-redundant information is very dim compared with the redundant information, and it is difficult to separate.This paper proposes an improved super-resolution algorithm to separate and process the non-redundant information.Many experiments reveals that the performance of this algorithm is excellent.