Research On Single Image Super-Resolution Reconstruction Method

Image is an important carrier for the transitional information of people,However,due to the inherent physical limitations of the imaging equipments and the unpredictable environment constraints,the obtained images are usually the degraded images of the original high-resolution(HR)images,also known as low resolution(LR)images.Image super-resolution(SR)attempts to reconstruct the desired HR images from the observed LR images by software algorithm.Due to the advantage of low cost,convenient replacement and reconstruction quality,SR has attracted a lot of attention and is widely applied to various fields such as medicine,public safety,aerospace,national defense and etc.In this paper,we mainly focuse on restoring a visually pleasing HR image from a single LR image generated by the low-cost imaging system and the limited environment condition.Since the observed LR images to be a non-invertible low-pass filtering,down-sampling and noise version of HR image,the reconstruction of HR images is a serious ill-posed problem.To handle this ill-posed problem,a variety of methods have been developed in recent years.Thus,we first review the present development of SR reconstruction and analyze the existing problems and difficulties.To resolve the problems,we propose three novel SR methods.The major contributions of these proposed methods are as follows:(1)Considering that the trained dictionary pairs by sparse coding based super-resolution(SR)methods have difficulty capturing the complicated nonlinear relationships between the LR and HR feature spaces,we propose a new single image SR method by combining sparse coding with the improved structured output regression machine(SORM).In the proposed method,the dictionary pairs are firstly learned by joint sparse coding to characterize the structural domain of each feature space and add more consistency between the sparse codes of two feature spaces.Then,since the classical SORM does not give sufficient weight to the independence of different output components,we improve the SORM by considering the correlation and independence between different output components to establish a set of mapping functions for tying the sparse code of two feature spaces.With this,the more precise mapping relationships between two feature spaces are obtained by the trained dictionary pairs and mapping functions.Moreover,we proposed a new global and nonlocal optimization for further enhancing the quality of the restored HR images.(2)With exploiting contextual information over large image regions in an efficient way,the deep convolutional neural network has shown an impressive performance for single image SR.Thus,we propose a new deep convolutional network by cascading the well-designed inception-residual blocks within the deep Laplacian pyramid framework to progressively restore the missing high-frequency details in the low-resolution images.By optimizing our network structure,the trainable depth of the proposed network gains a significant improvement,which in turn improves super-resolving accuracy.With our network depth increasing,however,the saturation and degradation of training accuracy continues to be a critical problem.As regard to this,we propose an effective two-stage training strategy,in which we firstly use images downsampled from the ground-truth HR images as the optimal objective to train the inception-residual blocks in each pyramid level with an extremely high learning rate enabled by gradient clipping,and then the ground-truth HR images are used to fine-tune all the pre-trained inception-residual blocks for obtaining the final SR model.Furthermore,we present a new loss function operating in both image space and local rank space to optimize our network for exploiting the contextual information among different output components.(3)Recently,deep neural networks have achieved impressive performance in terms of both reconstruction accuracy and efficiency for single image SR.However,the network model of these methods is a fully convolutional neural network,which is limit to exploit contextual information over the global region of the input image.Accordingly,we discuss a new SR architecture where features are extracted in the LR space,and then we use a fully connected layer which learns an array of upsampling weights to reconstruct the desired HR image from the final LR features.By doing so,we effectively exploit global context information over the input image region,whilst maintaining the low computational complexity for the overall SR operation.In addition,we introduce an edge difference constraint into our loss function to preserve edges and texture structures.Extensive experiments on benchmark datasets validate that our proposed three SR methods outperforms existing state-of-the-art SR methods in terms of the reconstruction quality and computational cost,which is beneficial to the application of image SR reconstruction.