基于密集连接与激励模块的图像超分辨网络 下载: 931次
ing at the loss of information and edge blurring during texture recovery using super-resolution technology based on convolution neural networks, we combine dense block and squeeze module to learn the mapping from low-resolution to high-resolution in an end-to-end manner. The dense block structure formed by the fusion of dense connection utilizes context information of image region effectively. The squeeze module amplifies valuable global information selectively and suppresses the useless features. The multiple 1×1 convolution layer structures in the image reconstruction section reduce the dimension of the previous layers, and speed up the calculation while reducing the loss of information. Processing the original image directly shortens the training time, and the optimization of convolution layers and filters reduces the computational complexity significantly.
胡诗语, 王国栋, 赵毅, 王岩杰, 潘振宽. 基于密集连接与激励模块的图像超分辨网络[J]. 激光与光电子学进展, 2019, 56(20): 201005. Shiyu Hu, Guodong Wang, Yi Zhao, Yanjie Wang, Zhenkuan Pan. Image Super-Resolution Network Based on Dense Connection and Squeeze Module[J]. Laser & Optoelectronics Progress, 2019, 56(20): 201005.