基于卷积神经网络的掌纹识别方法

为避免在处理掌纹识别时人工提取掌纹特征,提出使用卷积神经网络(CNN)来处理掌纹识别问题。首先根据掌纹的几何形状特点进行预处理,切割出掌纹的感兴趣区域(ROI);然后将感兴趣区域进行归一化并组成一个二维矩阵作为卷积神经网络的输入;再使用批量随机梯度下降算法对网络进行训练,得到最优的网络参数;最后对测试掌纹进行分类识别,分类器使用Softmax。应用于香港理工大学掌纹数据库(v2)的掌纹识别率达到99.15%,单张掌纹的识别时间小于0.01 s,验证了方法的有效性。

Convolutional neural network for palmprint recognition

To avoid extracting palmprint features when solved the palmprint recognition problem, this paper attempted to use Convolution Neural Network(CNN) to deal with it. First of all, according to the geometric features of palmprint to preprocess palmprint image, so that extracted region of interest(ROI). And then normalized ROI to form a two-dimensional matrix as the input of CNN. Secondly, mini_batchSstochastic gradient descent algorithm was used to train the network to get the optimal network parameters. Finally, the Softmax classifier is used to classify the palmprint. The results of experiments show that proposed network achieves 99.15% recognition accuracy on PolyU Palmprint Database(2nd Version), and single palmprint image recognition time is in less than 0.01s. The results demonstrate that proposed algorithm can be used to improve the recognition accuracy.