Space-time Super-resolution Algorithm From A Single Medical Video

Space-time super-resolution algorithm is a kind of method which can break the limitations of the spatial resolution and temporal resolution of image sequences. It has a wide application prospect, since this method does not require high hardware condition and only needs low cost to implement. This thesis introduces its basic principle and algorithm foundation, makes an analysis about the main problems of the current, and proposes the improved method.First of all, the paper introduces the main algorithms involving space-time super-resolution, including the super-resolution algorithm based on multi-scale structural self-similarity from a single image, the space-time super-resolution algorithm from multiple videos and the space-time super-resolution from a single video. It provides a theoretical basis and experimental foundation for the subsequent algorithm improvements. Because the space-time super-resolution can effectively reduce the phenomenon of motion blur and motion-based aliasing, the space-time super-resolution is applied in medical video for the first time in this paper and the result of experiment shows that the quality of the original video sequences is significantly improved.Secondly, a space-time super-resolution algorithm which based on Gaussian mixture model is proposed on the base of the space-time super-resolution algorithm from a single video. Because medical image has some characteristics, such as the distribution of gray-scale is determined by the different parameters of human tissue characteristic parameters, the difference between neighboring gray-scale is too small that the naked eye can hardly tell the difference and so on. The traditional matching of image patch or time-space patch contains a lot of errors, these errors will affect the super-resolution reconstruction effect. An improved matching algorithm is proposed in this thesis. Gaussian mixture model based clustering method is referenced here to establish constraint conditions,so that analyzing the similarity of image patch can give a more accurate matching and the error of original matching algorithm is corrected, the better space-time super-resolution effect can be produced.Then, analyzing the search algorithm of time-space similar patches, this paper proposes a new and improved method. Due to the optimization and efficiency of the algorithm, the existing reconstruction algorithms whether the multi-scale single image super-resolution algorithm, the space-time super-resolution algorithm from multiple videos or single video, spend too much time on running, which limits the use and promotion of the algorithm. This paper introduces average hash search algorithm on the base of original algorithm. Average hash search algorithm can keep the characteristic of image patch while quickly screening out the unqualified space-time patches. The experimental result shows that the method can improve the search efficiency.