基于特征注意力机制的图像超分辨率重建

Image super-resolution reconstruction is a technology that obtains corresponding high-resolution images through software or hardware methods based on existing low-resolution images. This paper is based on the generation of confrontation networks, and on the basis of existing research, in order to solve the problem that the existing methods cannot well recover the low-resolution images with unknown complex degradation factors and highlight the foreground, the pyramid feature attention mechanism is introduced into the pure synthesis In the data-trained image blind super-resolution field, a second-order degradation process is used to simulate actual degradation. Specifically, this paper uses the pyramid feature attention mechanism to enhance high-level context features and low-level spatial structure features. Among them, the context-aware pyramid feature extraction mechanism is used in multi-scale high-level feature maps to capture rich contextual features. The channel attention mechanism and spatial attention mechanism are respectively applied to the pyramid feature extraction map and the low-level feature map. The two are fused to detect salient areas, and then Increasing the edge preservation loss to obtain the accurate boundary between the salient area and the background, that is, guiding the network to learn more detailed information in the boundary positioning, so that the super-resolution image restoration effect is better, and the foreground target is paid more attention to. At the same time, this paper uses the second-order degradation model to better simulate the complex degradation process of the image in the real world, and uses pure synthetic data to train the hyperdivision network model. Extensive experimental results show that the model proposed in this article can restore most real-world images, and has better subjective visual effects than existing models. The objective evaluation indicators also prove the effectiveness and superiority of the method proposed in this article.