全手掌纹5类主线特征选择方法研究

Research on feature selection method based on five classes of palmprint principal lines on the whole palm

为解决非接触式掌纹识别中的光线干扰和算法复杂度问题,提出采用全手掌纹5类主线作为特征的思想,并研究主线特征的选择方法,即根据感情线、理智线、生命线、事业线、成功线的条数、交点数及部分端点信息将掌纹分为9个类别。采用主线特征的优点在于能够选择到不受光线干扰和运算复杂度小的掌纹特征自动提取方法,并为手多模态决策层融合提供数据基础。根据香港科技大学(HKUST)掌纹图库实验表明,与传统的只依据感情线、理智线和生命线3条主线为特征的分类方法相比,类间样本区分度增大,类内相似样本数由传统方法的81.54%降低至35.00%,其余各类样本趋于相对均匀,即说明了该方法不仅具有较高的分辨力,而且具有可行性和优越性,为手多模态特征识别技术中主线自动提取及匹配提供了理论支撑。

To solve the problems of light interference and algorithm complexity,we propose an idea that the principal palm-lines of the palmprint on the whole hand can be used as the feature of the palmprint.We propose a feature selection method of principal palm-lines,and according to the information of the number of emotion lines,intelligence lines,life lines,career lines and success lines,the information of the number of the cross points and the information of some endpoints,the palmprint is divided into 9 patterns.The advantage of this method is that it can adopt an auto extraction method for principal palm-lines that does not suffer from the interference of stray lights and operation complexity;also it can provide a quantity of basic data for decision level multi-model hand recognition.According to the test results of the palmprint data from HKUST,the differentiation of inter-class samples increases and the likelihood number of the within-class samples decreases from 81.45% to 35.00% compared with that of the feature selection method based only on emotion line,intelligence line and life line,which indicates that the proposed method enhances the resolution power,has feasibility and superiority,and provides experiment data and theory support for the research of multi-model hand recognition methods.