Tetrolet Sparse Regularization Learning Sample Image Super-resolution Algorithm

Using signal processing technology to make image super-resolution reconstruction has become one of the hotspots in the field of image processing. This article systematically reviewed the modern research status about image super-resolution, analyzed the confidential probability distribution, the joint probability distribution and the Hidden Markov tree statistics model by putting emphasis on the theory and experiments. And then combined the related theory and many analysis on the experiment, explored the application of Tetrolet transform in the algorithm of de-noising and image super-resolution reconstruction which is based on Tetrolet sparse constrain regular term and sample learning.The main work in this article is shown as follows:Firstly, on the basis of introducing the theory and construction method of the Tetrolet transform build the fitted model of the generalized Gaussian distribution, confidential probability distribution, the joint probability distribution about the Transform efficient. The experiment showed that the Tetrolet transform have the better sparse.Secondly, according to the sparse representation characteristics of the Tetrolet transform, threshold shrinkage de-noising is applied on the coefficient in the transform domain. And then on the basis of combining the Anisotropic Total Variation(ATV) characteristics designed an image de-noising algorithm unit the Tetrolet transform and ATV. The experiment of the algorithm showed that it can effectively suppress the Pseudo-Gibbs phenomenon result from the process of threshold shrinkage.Thirdly, the image super-resolution reconstruction model which is based on the TV regularization is researched, and then according to the sparse representation characteristics of the Tetrolet transform, build the regularized image super-resolution reconstruction model which combined the sparse representation characteristics of the Tetrolet transform and the TV and designed the forward-backward splitting algorithm according to the optimized resolution to the model. The experiment showed that the regularized model can maintain the edge and texture of the image effectively.Last, a two-step image super-resolution algorithm which is based on the prior optimized Tetrolet sparse regulation and the sample learning. The algorithm estimate the initial high resolution image according to the Tetrolet sparse regulation and then use the image database-based learning method further integrate the detail information and reduce the complexity of the Tetrolet coefficient study. The fast strategy of the coefficient learning was proposed in this article. The experiment showed that the efftiveness of the two-step algorithm.