Dorsal hand vein recognition based on convolutional neural networks
Dorsal hand vein recognition based on convolutional neural networks
复制标题
DOI：
10.1109/bibm.2017.8217830
复制
发表时间：
2017-11
期刊：
影响因子：
--
通讯作者：
Haipeng Wan;Lei Chen;Hong Song;Jian Yang
中科院分区：
文献类型：
--
作者：
Haipeng Wan;Lei Chen;Hong Song;Jian Yang
本文提出了一种基于卷积神经网络（CNN）的手背静脉识别方法，比较了不同深度CNN模型的识别率，分析了数据集大小对手背静脉识别率的影响。首先提取手背静脉图像的感兴趣区域（ROI），采用对比度有限自适应直方图均衡化（CLAHE）和高斯平滑滤波算法对图像进行预处理；然后训练Reference-CaffeNet AlexNet和VGG深度CNN提取图像特征。最后，应用逻辑回归进行识别。在两种不同大小的数据集上的实验结果表明，网络深度和数据集大小对识别率有不同程度的影响，基于VGG-19的手背静脉识别率达到99.7%。在本文中，我们还探讨了在SqueezeNet上集成学习的可行性。识别率略有下降，为99.52%，但模型尺寸急剧下降。
In this paper, we proposed a dorsal hand vein recognition method based on Convolutional Neural Network (CNN), compared the recognition rate of different depth CNN models and analyzed the influence of dataset size on dorsal hand vein recognition rate. Firstly, the region of interest (ROI) of dorsal hand vein images was extracted, and contrast limited adaptive histogram equalization (CLAHE) and Gaussian smoothing filter algorithm were used to preprocess the images. Then Reference-CaffeNet AlexNet and VGG depth CNN were trained to extract image feature. Finally, logistic regression was applied for identification. The experimental results on two different size of dataset shown that the depth of network and size of data set size have different degree effect on recognition rate, the dorsal hand vein recognition rate based on VGG-19 reaches 99.7%. In this paper, we also explored the feasibility of ensemble learning on SqueezeNet. The recognition rate declined slightly with 99.52%, but the model size has been decreased sharply.