A Super-Resolution Attention Feedback Network

Since the Internet and social media grow rapidly in recent years. Image processing is playing an increasingly important role in daily life. A key task for image processing is super-resolution. In a super-resolution task, an image of a high-resolution rate would be built on the basis of a low-resolution picture. Super-resolution technologies could help recover damaged images and recognize the detail of the images. As neutral networks develop quickly, many industries have used deep learning to improve their efficiency. Since neutral networks have a good performance in image processing, they quickly become the dominant methods in super-resolution tasks. Unlike traditional methods, deep learning could work better and more effectively. Based on the super-resolution feedback network (SFBEN), a channel attention mechanism has been added to the feedback block of the network. It is all known that attention mechanisms and feedback are two important ways for a human to deal with an image. This paper combines these mechanisms in the network. The new channel attentional layer uses a convolutional layer to generate a weight matrix and then adjust the input based on the matrix. The application of this channel attentional block reduces the training loss of the whole network and increases the peak signal-to-noise ratio.