基于分形残差网络的单幅图像超分辨率重建

近年来,各种基于卷积神经网络的单幅图像超分辨率方法取得了优异的性能提升.现有的超分辨率网络大多数都是使用单种尺度的卷积核来提取低分辨率图像的特征信息,这样很容易造成细节信息的遗漏,也无法很好地利用低分辨率图像的多尺度特征来提高图像的表达能力.为了解决超分辨率重建中存在的问题,提出了一种新的超分辨重建方法称为分型残差网络...

Single image super-resolution based on fractal residual network

In recent years, various single-image super-resolution methods based on convolutional neural networks have achieved excellent performance improvements. However, most of the existing super-resolution networks use single-scale convolution kernel to extract the feature information of low-resolution images, which can easily cause the omission of detailed information, and cannot make good use of the multi-scale features of low-resolution images to improve the expressive ability of images. In order to solve the problems in super-resolution reconstruction, a new super-resolution reconstruction method called fractal residual network (FRN) is proposed. The network uses fractal residual attention blocks to make full use of different hierarchical features to generate more refined features. At the same time, the channel attention mechanism is introduced to adaptively rescale the characteristics of each channel to increase the ability of network discrimination learning. In addition, the algorithm combines local residual learning with global residual learning to compensate for the loss of information and reduce the difficulty of learning. Experimental results show that this method is better than many other algorithms in reconstruction performance.