Image Super-resolution Using Convolutional Neural Network Based On Texture

As a carrier of visual information,image is one of the important sources of information.The quality of the image determines the accuracy of the information and the amount of information obtained,therefore how to improve image resolution become an important part in image processing technology.Because of neural network having the ability to fit any complex function,it can solve the ill-posed problem of super-resolution reconstruction to a certain extent.Therefore,it also shows good performance in super-resolution reconstruction.However,some texture information is not reconstructed well enough.To solve this problem,we proposed a super-resolution algorithm using convolution neural network(CNN)based on texture information.The main idea is to extract the texture features for classification,and finally a super-resolution model is built based on convolutional neural network for the same category of image samples.The main research of this paper is as follows:Firstly,introduced the image super-resolution reconstruction technologies,and introduced several typical super-resolution algorithms,include their basic principle and advantages and disadvantages.This paper mainly focused on the super-resolution reconstruction using convolution neural network algorithm,and did researches based on this algorithm.Secondly,introduced the method of image texture extraction and the classification method based on texture feature.First of all,proposed a method of extracting texture,using texture extraction method to get texture mask.Then,extracting the texture features by the gradient operator method.Finally,classifying the texture blocks by the KNN algorithm.Thirdly,introduced the application of image classification based on texture feature in superresolution reconstruction.It is proposed to build a super-resolution model for the same type of texture samples.This method is to improve the super-resolution reconstruction using convolution neural network method.The improvement is to add a texture mask to the loss layer,so as to achieve super-resolution for texture area.It is shown by the experiments that this algorithm can restore some texture information very well,and the results are better than other algorithms.