Algorithm Research Of Super-resolution Based On Sparse Representation

As a method of improving spatial resolution without hardware equipment, super-resolutionimage recovery technology uses different but similar information in multiple low-resolutionimage,and combines prior knowledge to rebuild a high-resolution image. At present,this resolution-enhancing technology may prove to be essential in criminal technology, medical imaging, high-definition digital TV, image compression,military fields and so on.This thesis mainly research of super-resolution based on sparse representation.Firstlyintroduces the basic theory of super-resolution image recovery. Then explains the basic principle ofsparse representation,commonly method of sparse coding and dictionary learning.The mainresearch work is as follows:1.It analyses super-resolution method based on reconstruction.And it compares thecharacteristics and performances of the two typical airspace super-resolution restorationalgorithms:maximum posterior probability method and convex set reflection method.2.It discusses super-resolution image recovering methods based on sparse representation,appropriate chosen over-complete dictionary.3.A new method based on sparse representation and combines convex set reflection ispropose.The algorithm consists of two phase:training and reconstruction.In the training phase,interpolation subtraction between high-resolution image patch and low-resolution image patch to getimage patch with high-resolution features;introduces feature-sign method for sparse coding;throughjoint training to get over-complete dictionary. In the reconstruction phase, usage of the projectiononto convex sets to get the optimal solution in global optimization. The experiment shows that ouralgorithms for super-resolution image restoration get better visual effect.