Research And Implementation On Image Super-Resolution Reconstruction Based On Generative Adversarial Network

With the fast development of computer technology,people have entered a new information age.Improving the resolution of an image is one of the important contents of information processing technology.It is limited to improve the resolution of the image by improving the imaging device,and an effective method is needed to overcome the hardware limitation.Aiming at this problem,based on the research on the existing image super-resolution reconstruction algorithm,this thesis proposes a new image super-resolution reconstruction algorithm based on generative adversarial network.Simultaneously combining the actual application scenarios and this algorithm,an Android application for image super-resolution reconstruction is designed and implemented.Aiming at the problem of poor super-resolution effect of partially restored images due to the feature fusion of foreground and background,according to the features of different levels have different characteristics,pyramid feature extraction mechanism is used to obtain high-level features of multi-scale and multi-receptive fields,and this thesis uses channel attention mechanism to select appropriate scales and receptive fields to better generative salient regions and recover more details and textures.Then this thesis focuses on the boundary between object and background and uses spatial attention mechanism to better focus on effective low-level features and obtain clear saliency boundaries.Simultaneously,aiming at the problem that most degradation methods are difficult to obtain satisfactory results for complex degraded real scenes,this thesis adopts a second-order degradation model to simulate more complex degradation process to synthesize more realistic degradation images,which keeps better balance between simplicity and effectiveness.Further,based on attention mechanism and second-order degradation model,this thesis proposes a new image super-resolution reconstruction model based on generative adversarial network.While restoring clearer reconstructed images,paying attention to the foreground of images makes them more in line with the visual habits of the human eye.Finally,an Android application for image super-resolution reconstruction is designed and implemented.The experimental results show that the image super-resolution reconstruction algorithm model proposed in this thesis has better subjective visual effects compared with the existing models,and the objective evaluation indicators also prove the effectiveness of the method proposed in this thesis.