Image Fusion And Super-Resolution Reconstruction Based On Deep Learning

Deep learning which maximizes the use of useful information in images has attracted extensive attention.It is of great significance and value to study effective deep learning algorithms and their applications in image fusion and super resolution.Based on the deep learning theory,this paper deeply studies the application methods of deep learning in image fusion and super resolution tasks,takes the establishment of deep learning model as the research focus,and conducts exploratory research on image fusion and super resolution based on deep learning.The main work includes the following aspects.First,the image fusion problem based on convolutional neural network is studied and solved.In order to avoid the risk of losing detailed information(edge information)in the process of image fusion,the image to be fused is decomposed into basic part and detailed part by image decomposition algorithm,and the basic part is fused by using the maximum rule of absolute value of coefficient.The focus detection of the detail part is realized by the deep learning model,and the fusion is realized by the weighted average rule.The final fusion rendering is obtained according to the fusion diagram of the two parts,so as to improve the clarity of the image and the ability to maintain the detail features.Then,the problem of image fusion based on residual learning is solved.Because the deeper the depth of the deep learning network,the more abstract the learned image features are,the more conducive to the later image fusion task.An image fusion algorithm based on residual learning is proposed to solve the problem that deep networks are difficult to converge due to gradient explosion or diffusion.The algorithm using extracted deep convolution neural network model for deep characteristics of fused images,maximum use of the image information,using the ideas of residual learning to solve the deep web optimization problem,at the same time through calculation for fusion of images with high definition images mutual information to determine the input serial number to ensure that the target residual image sparse.At last,this study solves the problem of image super-resolution reconstruction based on deep learning.Considering the importance of image prior information in super-resolution reconstruction,SRCNN model only extracts the features of the input image without any introduction of image prior information and cannot extract the deep features of the image.This paper proposes an image super-resolution algorithm based on deep learning,which takes the output of SRCNN as the pre-estimation image,we use the residual module to approach the high-resolution image step by step to further improve the image clarity and simulates its function in the form of network model.