Sparsity-based Video Denoise Algorithm And Application

Denoising is one of the very important research areas for image/video processing. Following the development of sparsity theory, the method designed for denoising develops from simple denoising ?lters to sparsity based methods. Nowadays, cutting edge methods include: methods based on patch matching-these methods base on the self-similar characteristic of images/videos in and between frames; methods developed from compressive sensing theory- these methods are designed according to the design of minimization problems to optimize the sparsity of images.In the process of research of denoising methods based on patch matching, denoising models for different kinds of data are proposed. First of all, for the denoising task of video data, a framework called OL-BM4 D is introduced. This framework varies BM4 D algorithm based on veri?ed robust optical ?ow for it to work on video denoising task. Secondly, for 4D medical data, a framework called BM5 D is introduced. This is the ?rst time patch matching method is designed for 4D medical data. This method makes use of variance stable algorithm to solve the denoising task for 4D medical data mixed with rice noise which is of unstable variance.In the process of research of denoising methods based on compressive sensing,an algorithm based on clustering in three dimensions for OCT data is designed. This algorithm cluster patches based on dictionary algorithm and optical ?ow. Then solve the three-head L1 minimization problem extracted from the algorithm, we proposed surrogation function based interactive shrinkage method and calculated the interactive shrinkage solution for the problem. Upon this method, we designed a shadow compensating method based on noise estimation to solve the shadow compensation problem in OCT image when it is mixed with noise of non-zero mean, this further enhances the image. With this result, we introduced a foreground detecting method into the detection of epichoroidal space in OCT image to test the performance of denoising and compensation methods and show the application of the method on medical data as well.