基于深度学习的非接触掌纹识别方法

Deep Learning for Contact-less Palmprint Recognition

采用非接触的方式获取到的掌纹图像容易受到手掌摆放姿势、光照条件等因素的影响,造成识别效果欠佳.针对这些问题,提出使用卷积神经网络来处理非接触式掌纹识别问题.首先利用肤色阈值切割出手掌图像,采用一种指根点检测算法确定指根点,使用确定的指根点建立坐标系,分割出ROI图像.其次,使用卷积神经网络自动提取特征,并将提取到的特征送入分类器得出结果.最后,在公开掌纹数据库上进行实验,取得了不错的识别效果,证明了此方法的有效性.

Under the influence of palm pose, illumination variation and complicated backgrounds; contact-less hand image recognition systems have unsatisfied recognition effect. To solve these question, a contact-less hand image recognition method based on convolutional neural network （CNN） was proposed. First the hand image was segmented from the background with a skin-color thresholding method. A finger base point detection method was used to find the finger base points. These points could be used to locate region of interest （ROD of palm. Then CNN was used to extract hand features which could be basic for classificatiorL The presented method was tested on publication hand image database. The results demonstrate that proposed algorithm can be used to improve the recognition accuracy.