Image Super Resolution Reconstruction Based On Convolution Sparse Coding

Image super-resolution reconstruction is an important branch of computer vision.Because it can effectively enhance the quality of the image without changing the condition of imaging equipments,it has been widely applied in the fields of satellite image,medical image,video codec,pattern recognition and so on.This paper mainly develops image superresolution reconstruction algorithms on the convolutional sparse coding by means of deep learning theory and sparse representation theory.The specific contents are as follows:1.Image super-resolution based on adaptive convolutional sparse coding and convolutional neural networks.Firstly,a low-pass filter is used to decompose the low resolution image into low-frequency part and high-frequency part.Secondly,a convolution neural network is applied to construct the mapping relation from the low resolution image of the lowfrequency part to its high resolution image to reconstruct the high resolution image.In the meanwhile,an adaptive convolution sparse coding is proposed to reconstruct the high resolution image for the high frequency part.Experimental results show that our proposed adaptive convolutional sparse coding algorithm has a better reconstruction effect than the convolutional sparse coding algorithm,convolutional neural network and some other image super-resolution reconstruction algorithms.2.Image super-resolution based on four-channels convolutional sparse coding.In the proposed method,a test image is put in four channels via rotating image ninety degree in four times.A low-pass filter and a gradient operator are used to decompose the input image into high-frequency part and low-frequency part.Then the high-frequent part and the lowfrequent part in each channel are reconstructed by means of convolutional sparse coding and cubic interpolation method,respectively.Finally,the reconstructed high resolution image is obtained via the process of weighting on the output images from four channels.The proposed method not only overcomes the problem of the consistency for the overlapping patches,but also improves the detail contour for the reconstructed image and enhances its stability.Experimental results show that the proposed method has a better peak signal to noise ratio(PSNR),structural similarity index(SSIM),and noise immunity than some other classical super-resolution reconstruction methods.3.Image super-resolution reconstruction based on the adaptive sparse domain select and convolution sparse coding.Firstly,a low-pass filter is used to decompose the low resolution image into low-frequency part and high-frequency part.Secondly,an adaptive sparse domain selection algorithm is used to reconstruct the low-frequency part of the low resolution image.In the meanwhile,an adaptive convolutional sparse coding is used to reconstruct the high resolution image.Experimental results show that our proposed adaptive convolutional sparse coding has a better reconstruction effect than the adaptive sparse domain selection method,convolutional sparse coding,convolutional neural network,and some other image super-resolution algorithms.