Research On The Image Super-resolution Reconstruction

With the progress of the times, science and technology alter from day to day, thus,people are paying much more attention to the image resolution. As a result of the limitednumber of detector array and restrictions of the detector structure in the infrared imagingsystem, spatial sampling frequency is usually unable to meet the sampling theorem. It isprone to mixing phenomenon which will result in blurred images. Due to the physicallimitations and expensive cost of the optical devices, how to improve the image spatialresolution on the basis of existing detector has become a hot spot in recent years. Imagesuper-resolution reconstruction is a technology which uses multiple low resolution imageson the same scene with complementary information to reconstruct a piece or pieces ofhigh-resolution images. The basic idea is that in the context of the existing imagingsystems, we breakthrough the limit of the imaging technology by means of certain softwaretechnology in order to obtain higher spatial resolution images. It can improve the spatialresolution effectively which is widely used in digital TV, medical imaging, military,meteorology and other fields.With the development of the SRR, both domestic and foreign researchers haveproposed a number of super-resolution reconstruction algorithms which can be generallydivided into two categories: frequency-domain algorithm and spatial-domain algorithm.Frequency-domain algorithm is limited to the global shift and the image degradation modelis spatial invariance, there is no major breakthrough on the real significance. Spatial-domain method is based on generic observation model, it contains various of priorknowledge which makes it more flexible and adaptable. Currently, some of therepresentative algorithms include iterative back-projection algorithm (IBP), projection ontoconvex sets algorithm (POCS), maximum a posteriori probability algorithm (MAP) and so on. This article is mainly focused on the MAP and POCS algorithm.In terms of the MAP algorithm, we studied its basic principle, analyzed the imageprior probability model, introduced the matrix MAP algorithm and gradient descentoptimization method, also the implementation of the MAP and its process is given. Finally,the simulation analysis of the MAP image reconstruction effect under different parametersis conducted.In terms of the POCS algorithm, this paper analyzed the concept and principles of thePOCS algorithm as well as its implementation process, and then, after a in-depth study, wemade two improvements to enhance image quality and effectively restrained the oscillationeffect. The proposed algorithm is simulated, and the experimental results show thatcompared to the original POCS algorithm, the edge oscillation effect has been significantlyimproved.At last, a summary of the full text is given, the prospect of image super-resolutionreconstruction techniques are discussed, we also give the direction and ideas of futureresearch.