Research On Image Super-Resolution Algorithm Based On Convolutional Neural Network

With the development of network technology,people are no longer satisfied with the information brought by the blurred images.Therefore,it gradually attracts people's attention on how to improve the clarity of natural images.The process of improving image clarity is called image super-resolution(SR).However,if we use traditional SR algorithms to extract the image features to complete SR reconstruction,we cannot improve image resolution well,since each image contains of different noises and different vague factors.Also,there still exist some shortcomings in training speed,deep-level feature extraction and texture detail recovery on the existing convolutional neural network(CNN)methods.Therefore,this method needs to be optimized to further enhance the overall restoration effect.The main challenge of image SR is to recover high-frequency details of a low-resolution(LR)image that is important for human perception.To address this essentially ill-posed problem,we introduce an improved shallow CNN method for single image SR to improve the operation efficiency,convergence speed and recovery quality and present a deep edge guided dual-channel CNN SR algorithm to progressively recover the high-frequency details.Firstly,our modified shallow CNN together with dividing and conquering is to split the feature space into numerous subspaces and learn edge texture details for each subspace,thereby creating effective mapping functions and getting a better recovery effect.Secondly,we present a deep edge guided dual-channel convolutional neural network SR reconstruction algorithm integrated with Morphological Component Analysis(MCA).The LR image to be processed is decomposed into texture part and structure part by MCA,and then the texture part and the original LR image form a dual channel together,which is then input into the modified network structure to reconstruct the high-resolution(HR)texture part.The reconstruction loss of both the HR image and HR texture is chosen simultaneously for training.Finally,as for post-processing step,we perform histogram matching between our network output and the original LR input to strengthen the visual effect and apply an iterative back projection refinement to improve PSNR value.We compare our method with the state-of-the-art SR methods on Set5,Set14 and BSD-100.As shown in experiment results,this method can restore texture details of images,specially restore the image with rich texture.