Image Restoration Via Adaptive Filtering And Sparse Regularization

Image as the key method for the human to capture vision information, it provides abundant information for human cognitive activity, and it becomes the important way of human recognition and perception of the objective word. The image we got is often with degraded quality because of the constraint of the imaging environment, imaging equipment, transmission, store and so on. In order to overcome the above shortcomings, it often use the image restoration algorithm to recovery the high quality image from the degraded observation.After analyzed the original reconstruction based model for image recovery and the image structure information, this thesis introduced a novel image reconstruction model which can recover the high quality of high frequency on the image and get more high quality detail information. Not only the framework of the image recover mode introduced, but also a detailed solution for image super-resolution provided in the thesis.In the classic single image recovery models includes two terms ––– data term and regularization term, the data term is often to keep the recovered image accord with the observation model and the regularization term is the prior knowledge of the image to keep the reconstruction object function can be solved.The key of our model from the original one is that an adaptive filter is used to remove the spatial relevance among pixels first and then only the high frequency of the image is added in the regularization term. Moreover, the wavelet transform coefficients of the residual term is smaller which agree with the minimization function. In the object function, we use the union regularization term of total variation term and the sparse wavelet transform to make full use of the advantages of the two.To solve the union regularization object function, the thesis introduced a detailed reference solution of image super-resolution problem by the alternate direction method. In the experimental part, a set of remote sensing images and depth images are tested by non-local mean adaptive filter and guided image adaptive filter to demonstrate the effectiveness of the proposed framework. Experimental results show the outstandingperformance of the proposed method in quantitative evaluation and visual fidelity compared with the state-of-the-art methods.