Research On Super-resolution Reconstruction Algorithms Based On Learning

Super-resolution reconstruction technology(SR)has been applied to various fields at present.It is a very important role in image processing.With the SR technology,some low resolution(LR)images can be reconstructed to obtain high resolution(HR)images with richer high frequency information and sharper texture.In this way,it is conducive to the next analysis and processing.SR reconstruction technology in software the way of improve the quality of the images has become a kind of economical and effective means to improve the image resolution.At the same time,with the advent of the era of big data,learning-based algorithm SR reconstruction algorithm is popular in the research of algorithm now.The main content of this paper include:First,the basic knowledge of SR is introduced,including the basic model of SR reconstruction technology and the classification of SR reconstruction algorithm.Some of SR reconstruction algorithms based on interpolation and learning are introduced in detail.Quality evaluation of reconstruction image has also been made a detailed study.Next,on the basis of understanding and implementing several typical image SR reconstruction algorithms,SR algorithm based on sparse representation is analyzed in detail.This paper studies a kind of improved SR algorithm based on centralized sparse representation.The K-means algorithm is added in the training and reconstruction phase,it is used to cluster the similar image patches.Then these patches select dictionary adaptively to reconstruct image.By this way,the quality of reconstructed image is better than the classical method.On the basis of this,the method of obtaining PCA is improved,and the original eigenvalue decomposition is replaced by singular value decomposition,which makes the obtained sub-dictionary more accurate,so as to achieve better reconstruction effect.At last,a super-resolution reconstruction(SRCNN)algorithm for remote sensing images based on depth convolution neural network is studied in this paper.First of all,this paper constructs the training set of remote sensing image and constructs a model suitable for super-resolution reconstruction of remote sensing image.Experimental results show that this constructed training set has a good effect on the reconstruction of remote sensing image.And the quality of the reconstructed image is much better than that of the traditional SR reconstruction algorithm in the subjective vision and objective evaluation of indicators.