一种基于数据场和小波包熵的掌纹识别方法

A palmprint recognition method based on data fields and wavelet packet entropy

掌纹的方向是一种十分有效的特征,但如何将纹线特征和方向特征有效地融合仍然是未解决的问题.提出一种基于场特征的掌纹识别方法.该方法利用数据场和小波包熵构建和表征掌纹场特征以实现掌纹识别.首先将数据场理论引入到掌纹识别领域,构建掌纹数据场,并将其分解为绝对数据子场、相对数据子场和方向子场;然后,基于小波包将不同数据子场分解求取各节点相对小波包能量,并计算小波包熵表征各掌纹子场不同频带能量分布特征;最后,将各子场特征拼接整合为掌纹场特征,并使用BP神经网络对其进行分类.实验结果表明,该方法可以获得较高的识别精度.

Palmprint images contain rich unique features for reliable human identification,which makes it a very competitive topic in biometric research.From a low resolution palmprint image,the information of principal lines and wrinkles can be obtained to realize palmprint recognition.The direction feature of palmprint lines is an effective feature.But how to effectively fuse the direction feature and other palmprint line features is an open problem in palmprint recognition.In order to solve the problem,apalmprint recognition algorithm based on palmprint field features is proposed in the paper.In the method,data fields and wavelet packed entropy are used to construct palmprint data field and extract a new palmprint feature,the palmprint field feature.The field feature is the combination of the structural feature and direction feature.Firstly,the data field theory is introduced into palmprint recognition field and each point in the palm lines is seen as a data point with unit mass to map an enhanced palmprint image from gray space to the corresponding potential space.In the space,all points in the palm lines will be affected by other points to form a palmprint image data field.Because the distribution of palmprint data field is affected by the thickness,direc-tion and distribution density of palm-lines,a wealth of the structural and direction information of palm-lines are provided by palmprint data field.For the sake of improving the distinguish ability,the palmprint data field data is decomposed into a relative palmprint data field,an absolute palmprint data field and a direction data field.The absolute data field can make a rough distinction between background and targets,the edge information is highlighted in the relative data field and the direction information of points in each palm-line is obtained in the direction data field.Next,the different sub-data fields are decomposed by wavelet packet transform and the entropies for all nodes for wavelet packet are calculated.These wavelet packed entropies can represent the features of energy distribution of different sub-data fields in different nodes.Finally,all of features of each sub-field are joined into one palmprint field feature,which is fed to backpropagation neural networks for classification.The experimental results illustrate the effectiveness of the method.