李伟,张旭东.基于卷积神经网络的深度图像超分辨率重建方法[J].电子测量与仪器学报,2017,31(12):1918-1928 |
基于卷积神经网络的深度图像超分辨率重建方法 |
Depth image super resolution reconstruction based on convolution neural network |
|
DOI:10.13382/j.jemi.2017.12.006 |
中文关键词: 深度图像 超分辨率重建 卷积神经网络 残差网络结构 |
英文关键词:depth image super resolution convolution neural network residual network |
基金项目:国家自然科学基金(61403116)、中国博士后基金(2014M560507)资助项目 |
|
|
摘要点击次数: 2760 |
全文下载次数: 8860 |
中文摘要: |
为了更有效地提高深度图像的分辨率,构建了一种更深层次的深度图像超分辨率重建的卷积神经网络。该网络直接将低分辨率深度图像作为网络的初始输入,通过卷积神经网络学习图像的高阶表示,获得更具有表达能力的深层特征,同时在网络的输出层引入亚像素卷积层,针对提取到的特征学习不同上采样滤波器,实现上采样放大操作。为了实现网络更好地收敛,在网络中加入了残差网络结构。在4个常用数据集上的实验结果表明,与其他先进方法相比,该方法网络收敛速度更快,并可以有效地保护图像的边缘结构,解决伪影问题,且在定性和定量两方面均取得了很好的重建效果。 |
英文摘要: |
In order to improve theresolution of depth imagemore effectively, a deeper convolution neural network is constructed in this paper. The network directly adapts the low resolution depth image as the initial input of the network,and learns the high order representation of depth image through the convolution neural network to obtain the features with more expressive ability.At the same time,the sub pixel convolution layer is introduced at the output layer of the network. Based on the extracted features, a set of sampling filter is learned to achieve the amplification operation. For a better performance of the convergence, the residual network is added to our network. The experimentsare conducted on four commonly used datasets, and the results show that our network is faster than other advanced ones at the convergence rate. The proposed method can effectively protect the edge structure of the depth image,solve the artifact problem,and reachesgreat performance both in qualitative and quantitative aspects. |
查看全文 查看/发表评论 下载PDF阅读器 |