Lee Yooho, Jun Dongsan, Kim Byung-Gyu, Lee Hunjoo
Department of Convergence IT Engineering, Kyungnam University, Changwon 51767, Korea.
Department of IT Engineering, Sookmyung Women's University, Seoul 04310, Korea.
Sensors (Basel). 2021 May 12;21(10):3351. doi: 10.3390/s21103351.
Super resolution (SR) enables to generate a high-resolution (HR) image from one or more low-resolution (LR) images. Since a variety of CNN models have been recently studied in the areas of computer vision, these approaches have been combined with SR in order to provide higher image restoration. In this paper, we propose a lightweight CNN-based SR method, named multi-scale channel dense network (MCDN). In order to design the proposed network, we extracted the training images from the DIVerse 2K (DIV2K) dataset and investigated the trade-off between the SR accuracy and the network complexity. The experimental results show that the proposed method can significantly reduce the network complexity, such as the number of network parameters and total memory capacity, while maintaining slightly better or similar perceptual quality compared to the previous methods.
超分辨率（SR）能够从一张或多张低分辨率（LR）图像生成高分辨率（HR）图像。由于最近在计算机视觉领域研究了各种卷积神经网络（CNN）模型，这些方法已与超分辨率相结合，以提供更高的图像恢复效果。在本文中，我们提出了一种基于轻量级CNN的超分辨率方法，称为多尺度通道密集网络（MCDN）。为了设计所提出的网络，我们从DIVerse 2K（DIV2K）数据集中提取了训练图像，并研究了超分辨率精度与网络复杂度之间的权衡。实验结果表明，所提出的方法可以显著降低网络复杂度，如网络参数数量和总内存容量，同时与以前的方法相比，保持略好或相似的感知质量。
