基于卷积神经网络的掌纹识别

Non-Contact Palmprint Recognition Based on Convolutional Neural Network

随着社会的发展,身份信息的安全问题日益凸显。为解决用户身份识别过程中受环境影响较大,以及掌纹识别时提取掌纹特征复杂的问题,本文进行了“基于卷积神经网络(CNN)的掌纹识别”的研究。运用该算法的优势在于简化了掌纹识别的前期预处理,可以直接将采集的原始图像进行输入,然后识别。通过卷积操作和最大池化操作,减少了训练参数量,大大节约了时间。最后使用Softmax分类器对结果进行分类。实验结果显示,该方法对不同人的掌纹有较高的识别率,克服了传统掌纹识别精度差,识别时间长,人工提取特征困难的缺点。

With the progress of society,the security of identity information has become increasingly prominent.In order to solve the problem of complex environmental impact and complex extraction of palmprint features in the process of user identification,the paper studies palmprint recognition based on convolution neural network.The advantage of this algorithm is simplifying the pre-processing of palmprint recognition,by directly inputting and recognizing the collected original image.Through convolution operation and maximum pooling operation,the method reduces the amount of training parameters and saves time greatly.Finally,the results are classified by using Softmax classifier.The experimental results show that this method has high recognition rate for different palmprints,overcoming the shortcomings of traditional palmprint recognition,such as low recognition accuracy,long recognition time and difficulties in manual feature extraction.