Research On Super Resolution Algorithm Based On Generating Countermeasures Network

Image super-resolution reconstruction(Super Resolution,SR)is an important technology in the field of computer vision,which aims to convert blurred images into clearer and more accurate images.Super-resolution technology plays a vital role in fields that require high image details,such as surveillance and medical imaging.Traditional superresolution methods(such as bicubic interpolation)simply enlarge the image without considering additional information such as the structure and color of the image itself,so the generation effect is mediocre.With the rise of the deep learning neural network,the superresolution network based on deep learning can effectively use the additional information in the image and achieve good results.However,super-resolution algorithms usually require pairs of high-resolution images(High Resolution,HR)and low-resolution images(Low Redolution,LR)to train the network.Most of the existing data sets only have highresolution images,which have not been successful For LR-HR data,it is necessary to manually generate low-resolution images to simulate the degradation process in real scenes.However,the degradation process of images in real scenes is more complicated.The current common methods,such as bicubic downsampling,cannot simulate the complex degradation process in real world scenes well; at the same time,although deep learning has achieved good results in the field of super-resolution However,the low-resolution image contains a lot of low-frequency information,and the neural network treats each channel in the image equally,and does not focus on the key parts of the image,which affects the information extraction ability of the network.Therefore,new methods need to be proposed to improve the quality of the data set and the representation ability of the network.First of all,in view of the general quality of existing datasets,this paper first proposes a Configurable Higher-orfer Degenerate Model(CHDM)that combines multiple image degradation methods.On the basis of the four basic degradation methods,CHDM further supports configurable random use,thus realizing the coexistence of single-order degradation and high-order degradation,better simulating the image degradation process in real scenes,and constructing a higher quality HR-LR image pairs,effectively improving the quality of the dataset.The experimental results show that the improved HR-LR data pair constructed by CHDM can effectively improve the image reconstruction results and simulate real degraded scenes.Subsequently,in view of the problem that the existing network cannot extract important information in the image channel,Based on SRGAN(Super Resolution Generative Adversarial),this paper deeply analyzes each layer of SRGAN network.Focusing on the feature extraction and expression capabilities of SRGAN,we improved the residual network of SRGAN,added a channel attention mechanism,and proposed an improved Generative Adversarial Network Channel Attention Super Resolution based on the channel attention mechanism.GCASR)model.And the L1 loss function is introduced on the loss function of the discriminator,which effectively reduces the artifacts in the reconstructed image.In order to verify the performance improvement of the proposed GCASR model,this paper quantitatively conducts comparative experiments with common superresolution models.The experimental results show that the channel attention mechanism can make the GCASR model pay more attention to the high-frequency details of the picture,and it is beneficial to improve the PSNR and SSIM value of the reconstructed picture.Although the model parameters and reconstruction time of GCASR have increased,the overall effect of its reconstructed pictures has made great progress.Finally,this paper applies the CHDM-based GCASR model to the blurred image reconstruction module,and implements a high-precision blurred image resolution enhancement system on the Web.The system can effectively process large-scale image data and has wide application prospects.In summary,the GCASR model based on CHDM and the research results of the resolution enhancement system may play an important role in actual scenes,and promote the development of image processing and image super-resolution reconstruction technology.