Research And Application On Image Super-Resolution Based On Deep Learning

With the improvement of computer hardware computing power and the continuous development of artificial intelligence technology,deep learning has achieved outstanding performance in the field of computer vision.The image super-resolution method based on deep learning conducts network training through a large number of samples,and learns the mapping relationship between low-resolution images and high-resolution images directly.The reconstructed images have higher accuracy and better quality.The image super-resolution reconstruction methods based on deep learning gradually become the research hotspots in the field of super-resolution.This thesis has carried out the research of image super-resolution reconstruction method based on deep learning of which the main work includes the following aspects:(1)Using the advantages of convolutional neural networks in the field of image processing,an image super-resolution reconstruction method based on residual learning is proposed.This method uses cascaded deep convolutional networks to extract features from images and introduce residual learning to obtain deep texture detail information.At the end of the network,the feature image is up-sampled through the deconvolution layer to reconstruct a high-resolution image with the same size as the target image.(2)In order to make better use of the high-frequency information of the image,an image super-resolution reconstruction method based on the attention mechanism is proposed.This method introduces an attention mechanism in the residual block,adaptively calibrates the features,adjusts the weight of the feature map of each channel,allocates more computing resources to the core of the task when the computing power is limited,and strengthens the high-frequency information of image features,there by assisting the image super-resolution network to complete the extraction and recovery of image high-frequency information.(3)Considering the actual application value,a deep learning-based image super-resolution reconstruction system is designed and implemented.The system includes user management module,statistical analysis module,model training module,image reconstruction module and data management module.The constructed Web visual interface can more intuitively help users to achieve image super-resolution reconstruction tasks,and provide users with friendly functions and services.This thesis uses Set5 and Set14 for testing.The experimental results show that the image super-resolution method based on residual learning and the image super-resolution reconstruction method based on attention mechanism have improved both in the two evaluation indexes such as the peak signal-to-noise ratio and the structural similarity,and the visual effect of the image can be effectively improved.