Research On Image Super-resolution And Deblurring Algorithm Based On Deep Learnin

Images are the foundation of human vision,reflecting things in nature,and are also an important source of human cognition and self-awareness.However,images are easily influenced by external factors,leading to the loss of some information,causing image distortion and affecting people’s perception of the world.Therefore,solving the problem of image resolution is a challenging task that requires a deep understanding of the mechanisms and characteristics of people’s perception systems,and the establishment of more comprehensive mathematical models in order to restore high-resolution images from low resolution images.In this case,a deep learning method is adopted without relying on any external conditions to improve the resolution,contrast,and exposure time of the image.Therefore,this article focuses on studying algorithms in the field of image super-resolution and deblurring,with the main research results in the following two aspects:(1)This paper studies the application of deep learning in the super resolution field of reference images.Aiming at the problems of channel learning ability and texture information conversion efficiency in fusion module,an enhanced texture Transformer network based on deep learning is proposed.Reference images,reference image interpolation and low resolution images are used as input images,and high-quality features of reference images are transferred to low resolution images by transforming features and other modules.In addition,detailed features of backbone network features are further extracted,and then texture conversion Transformer module is combined with output features of backbone network and input into MSFI module for feature fusion to transform low-resolution images into higher-definition images.Experiments show that the algorithm achieves remarkable results in different scene images such as nature and architecture.Compared with the latest algorithm,the signal-to-noise ratio and structural similarity indexes in the three data sets are improved by 0.28 d B and 0.02 on average,indicating that the proposed algorithm can improve the texture information of the image to a certain extent.(2)Then This paper studies the application of deep learning in the field of moving image deblurring.In order to solve the problem of too many parameters and high computational complexity in residual networks,a lightweight multilevel asymmetric network for moving image defuzzification is proposed.The network can conduct model training efficiently,which is conducive to improving the convergence speed of the network,and effectively reducing the number of parameters and computational complexity,so that the execution time of the algorithm is increased by 80 ms.In order to make full use of the high frequency signal extracted from the image,it can get better transmission in the network.In this paper,the feature fusion method is adopted.By combining the features extracted from the shallow convolutional module with the asymmetric residual module,the extracted high frequency signal and low frequency information can be fused,and the characterization ability of the network can be significantly improved.In addition,the fusion strategy can further enhance the effect of image recovery and improve the performance of the network.Experiments show that the algorithm has a good performance on object motion image,real city image and other deblurring data sets.The performance indexes of PNSR and SSIM of the proposed algorithm in the four data sets are improved by 0.3d B and 0.06 on average compared with the latest algorithm,indicating that the proposed algorithm has obvious effect in fuzzy images.In summary,this article aims to delve into two important sub areas in the field of image restoration,namely image super-resolution and motion image deblurring.In order to solve the problems of texture transformation and large model parameters,two different algorithms are adopted,including image super-resolution texture Transformer network and lightweight multilevel residual network,for motion image deblurring.The experimental results show that both algorithms have achieved significant results and effectively solved the problem of image quality,further proving the effectiveness of deep learning algorithms in the field of image restoration.In addition,the design of these algorithms is also closer to the characteristics of the human visual nervous system,making the image restoration effect more satisfactory.