手指生物特征变换与超球粒化方法研究

发布时间：2018-01-08 16:15

本文关键词：手指生物特征变换与超球粒化方法研究　出处：《中国民航大学》2017年硕士论文　论文类型：学位论文

【摘要】：传统的单模态生物特征识别稳定性差,不能满足人们的实际需求。为了提高生物特征识别系统的有效性和可靠性,近几年,利用多模态特征进行人的身份识别逐渐成为了研究热点。然而,多模态特征融合识别算法在特征融合时存在着很多亟待解决的问题,例如融合时多模态特征图像尺寸不一,以及如何选取更有效的融合方式等。本文以指纹、指静脉和指节纹为研究对象,尝试通过特征变换和超球粒化的方式解决多模态特征识别现存的一些问题。本文的主要工作如下:(1)针对多模态特征的尺度差异性问题设计了两种特征变换的方法。1)在空域上,通过均匀划分和超像素划分两种思路,将不同模态的图像化为相同的块数,并在每个图像块内提取相同维度的特征;2)提出了邻域特征矩阵生成法,依据图像的Gabor方向特征进行特征变换,实现了将三模态不同维度的特征化为相同维度特征的目的,同时也对数据进行了有效降维,减少了系统的运算代价。(2)针对多模态特征的融合问题构建了一个超球粒聚类的融合模型。利用超球粒之间的合并算子和给定阈值将同一个体的不同模态特征融合成一个超球粒,作为后续识别的依据。同时,采用超球粒的三角形融合模型,进一步验证了利用超球粒的思想解决三模态特征融合识别问题的可行性。本文通过大量实验验证了提出的方法,实验结果表明,利用提出的邻域特征矩阵生成法,结合超球粒聚类融合模型,较传统的生物特征识别方法可以达到更高的识别精度。
[Abstract]:Single modal biometric stability of the traditional difference, can not meet the actual needs of the people. In order to improve the effectiveness of biometric identification systems and reliability, in recent years, the use of multimodal features for human identification has become a hot topic. However, multi-modal feature fusion recognition algorithm has some problems with in feature fusion, such as fusion of multimodal feature image sizes, and how to select more effective fusion methods. Based on fingerprint, finger vein and knuckleprint as the research object, attempts to change the characteristics of the chondrules and super multi-modal feature recognition of some existing problems. The main work of this paper are as follows: (1) in order to solve the problem of the scale difference of multi modal features designed two methods of feature transform.1) in space, through the uniform partition and super pixels divided two kinds of ideas, will The image of different modes for the same number of blocks, and the same dimension feature extraction in each image block; 2) proposed the neighborhood feature matrix generation method based on Gabor transform, the feature directions of the image feature, realize the purpose of the three mode characteristics of different dimensions into the same dimensions, but also effectively to reduce the dimensionality of the data, the system to reduce the cost of computation. (2) for the fusion of multi modal features constructed a fusion clustering model. Using the super chondritic merge operator and the given threshold will be different between super chondritic modal characteristics of the same individual into a super chondrules, as subsequent recognition basis. At the same time, the spherical triangle fusion model is further verified by super chondritic thought to solve the feasibility of three modal feature fusion recognition problem. The proposed method was verified by experiments, The experimental results show that the proposed neighborhood feature matrix generation method and the hyper sphere clustering fusion model can achieve higher recognition accuracy than the traditional biometrics method.

【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】