Research On Single Frame Image Super-Resolution Algorithm Based On End-to-end Deep Learning

Image super-resolution technology(super-resolution,SR)is a very important part of image processing,super-resolution technology is used in various fields.With SR technology,some low-resolution images can be reconstructed to obtain high-resolution images with more details and sharper texture.High-resolution images are conducive to the next analysis and image processing.Super-resolution technology is a technology that uses software to improve image quality and it is an economically feasible way to improve image resolution.With the development of artificial intelligence and deep learning,the image super-resolution technology based on deep learning is a research hotspot.The main content of this article include:Firstly,we introduce the basic theories of image super-resolution and the common algorithms.Two kinds of super-resolution algorithms based on deep learning are introduced in detail.The two algorithms are convolutional neural networks and very deep networks respectively.Then the content of self-similarity in image is introduced,which is divided into the similarity of similar blocks in scales and cross scales.And we also introduce how to apply the similarity of images to the super-resolution preprocessing.Finally,we propose an improved deep learning network model based on very deep network.And we elaborate on the residual network and batch normalization theory and network structure used in our method.The whole network structure model and derivation of related theories are described.In the last section,the experimental results and the comparison between the proposed algorithm and the related algorithms are also presented.It is shown that the algorithm proposed in this paper has more subjective and objective effects than the common super-resolution algorithms.