Research On Image Super Resolution Based On Multi-dictionaries Learning

With the rapid development of multi-media technology,video and image are quickly gaining in popularity.As one of the most important way to express visual information,image is also a major carrier of information.High resolution images can provide more detail about the target scene than the low resolution images.In many application of image analysis,such as medicine,remote sensing,detection and meteorology,detailed information is very important.Therefore,how to improve the spatial resolution of image attract more and more people's attention.High resolution image,which meet people's anticipate and the requirements of practical application,is hard to achieve because of the limitation of hardware device and illumination environment during the process of obtaining image.In the case of adopting hardware's way to obtain high resolution images,we not only have technical bottlenecks but also higher cost.Thus,software's way become an alternative method to improve the spatial resolution of the image.To fulfill the task of super-resolution reconstruct,this article study the single image super resolution reconstruction algorithm based on multi-dictionaries learning,the main work includes the following:(1)Describes the research background and significance,then introduces the development and current situation of image super resolution technique,particularly the perception of compressed sensing.Summarizes the research progress of super-resolution algorithm based on dictionary learning technique and analysis its existing problems.(2)Considering that super-resolution reconstruct is largely dependent on the training set and the over complete dictionary is difficult to learn,this paper put forward the framework of multi-dictionaries learning,which combine K-SVD algorithm for each set of specific categories of training sample to get the specific dictionary.This framework would make each specific dictionary perform better than a common characteristic of the class,thus makes the image processing based on multi-dictionaries can code and describe the information more accurately.(3)Put forward a super-resolution algorithm based on multi-dictionaries learning.First classify the input image and choose the specific dictionary.Then solve the optimization problem to get the sparse coefficient.Finally joint the sparse coefficient with the specific dictionary to reconstruct high resolution image.The image super-resolution reconstruction experiments validate our algorithm can effectively reconstruct the detail of the high resolution image and get higher PSNR.