A Novel Framework of CNN for Image Super-Resolution Based on Attention Module
Jiahao Tan, Hiroaki Mukaidani · 2021
Because the convolutional neural network only captures the inherent size feature of a single image in the research of image super-resolution process, a framework based on the attention module and multi-dimension feature merge is proposed. Using the attention module, the network can validly conform non-local information, thus improving the network's feature expression ability. Meanwhile, the convolution kernels of different dimensions are used to extract the multi-dimension intelligence of the image to maintain the intact information of distinguishing feature under the different scales. Experimental results demonstrate that this method is advantageous than some super-resolution reconstruction alagorithms in objective quantitative indicators.