摘要
图像分辨率提高即超分技术是指从低分辨率图像重建相应的高分辨率图像,在医学影像等领域有重要的应用价值。传统的基于插值的方法效果不尽理想,近年来深度学习被应用于该领域。回顾了快速超分辨卷积神经网络(FSRCNN)、深度超分辨率卷积神经网络(VDSR)、超分辨率生成对抗网络(SRGAN)3种神经网络在图像超分中的应用原理,设计实验测试网络结构的效果,使用Set4、Set14、Urban100等数据集进行峰值信噪比、结构相似性等指标的测试,VDSR效果较好,改进VDSR网络结构,由原来的Y通道扩展为三通道(VDSR-RGB),进一步提升了超分效果。
Image super-resolution algorithm refers to the reconstruction of corresponding high-resolution images from low-resolution images,which has important application value in such field as medical imaging.In recent years,deep learning has been applied in this field.The application of three types of neural networks,FSRCNN,VDSR and SRGAN in image super-resolution is reviewed.Experiments are designed to test the effect of these network structures.The PSNR and SSIM are tested by using Set4,Set14 and Urban100 data sets.The performance of VDSR is relatively the best.The VDSR model from the original Y channel to three channels(VDSR-RGB)is improved and the effect of the SR is further enhanced.
作者
韩伟娟
董新捷
董文龙
HAN Weijuan;DONG Xinjie;DONG Wenlong(Zhongyuan Institute of Science and Technology, Zhengzhou 450000, China;Henan Provincial Public Security Department, Zhengzhou 450003, China)
出处
《信息工程大学学报》
2021年第2期159-163,共5页
Journal of Information Engineering University
作者简介
韩伟娟(1986-),女,工程师,硕士,主要研究方向为无线传感网络、图像处理。
