光学学报, 2019, 39 (2): 0210003, 网络出版: 2019-05-10   

基于深层残差网络的加速图像超分辨率重建 下载: 1400次

Super-Resolution Reconstruction of Accelerated Image Based on Deep Residual Network
作者单位
哈尔滨工程大学信息与通信工程学院, 黑龙江 哈尔滨 150001
摘要
针对目前卷积神经网络的超分辨率算法存在卷积层数少、模型简单、计算量大、收敛速度慢以及图像纹理模糊等问题,提出了一种基于深层残差网络的加速图像超分辨率重建方法,该方法在提高图像分辨率的同时加快收敛速度。设计更深的卷积神经网络模型来提高精确度,通过残差学习并且使用Adam优化方法使网络模型加速收敛。在原始低分辨率图像上直接进行特征映射,只在网络的末端引入子像素卷积层,将像素进行重新排列,得到高分辨率图像。实验结果表明,在set 5,set 14,BSD100测试集上,所提算法的峰值信噪比与结构相似性指数均高于现有的几种算法,能够恢复更多的图像细节,图像边缘也更加完整且收敛速度更快。
Abstract
An accelerated image super-resolution reconstruction algorithm based on deep residual network is proposed to solve some existing problems, such as few convolutional layers, simple model, large amount of calculation, slow convergence speed and fuzzy image texture. This method improves image resolution and accelerates convergence speed at the same time. First, a deep convolutional neural network model is proposed to improve accuracy, and accelerate convergence of network models is achieved by residual learning and Adam optimization method. Second, feature mapping is performed directly on the original low-resolution image, and the sub-pixel convolutional layer is introduced at the end of the network to rearrange the pixels, so a high-resolution image is obtained. Experimental results show that the proposed algorithm has higher peak signal-to-noise ratio and structural similarity index than those of existing algorithms on set 5, set 14 and BSD100 test sets, and can recover more image details. The image edges are complete and the convergence speed is fast.

席志红, 侯彩燕, 袁昆鹏, 薛卓群. 基于深层残差网络的加速图像超分辨率重建[J]. 光学学报, 2019, 39(2): 0210003. Zhihong Xi, Caiyan Hou, Kunpeng Yuan, Zhuoqun Xue. Super-Resolution Reconstruction of Accelerated Image Based on Deep Residual Network[J]. Acta Optica Sinica, 2019, 39(2): 0210003.

本文已被 10 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!