多模态生物特征识别技术的算法研究

发布时间：2018-10-26 21:35

【摘要】：近年来,生物特征识别技术得到了飞速发展,然而传统的单一模态的身份识别技术存在着一定的局限性,导致了该技术在实际应用中会出现一些不必要的麻烦。伴随数据融合技术的日渐成熟,多模态生物特征识别这一利用多种生物特征进行数据融合识别的身份识别技术获得了很大的技术支持,也促使该技术能够更快的进入我们日常生活中。本文通过对常见的单模态生物特征、传统的多模态生物特征的融合策略等方面的研究,最终采用指纹和虹膜两种单模态生物特征,在多模态生物特征识别中的特征层这一层次进行了融合识别的实验。主要工作总结如下:1.深入了解并研究多模态生物特征识别在各个层次进行数据融合的相关方法,包括串并联和基于典型相关分析(CCA)等在特征层融合、基于最小二乘法和Fisher判别等在分数层融合、加权法和多数投票法等在决策层融合。2.在深入研究多模态生物特征融合识别的基础上,提出基于指纹和虹膜的特征层融合模型,与现有的特征层融合的策略进行了对比和分析,通过实验论证了多模态融合识别的识别率相比较单模态识别更高一些这一论点,并验证了基于典型相关分析的融合算法在多模态生物特征识别中的有效性。3.针对基于典型相关分析的融合算法的不足,提出一种基于矩阵变换的判别典型相关性分析(MDCCA)的多模态生物特征识别算法,在同一环境下对两种算法进行了实验,实验验证了该算法的有效性。4.在本文提出的新算法的基础上,完成了比较完整的多模态生物特征识别过程。本文研究的算法内容在数据融合和多模态生物特征识别算法的领域都有着一定的参考价值。
[Abstract]:In recent years, biometric identification technology has been developed rapidly. However, the traditional single mode identification technology has some limitations, which leads to some unnecessary problems in practical application. With the maturation of data fusion technology, multi-modal biometric recognition, which uses a variety of biometric features for data fusion recognition, has received a lot of technical support. It also enables the technology to enter our daily lives more quickly. In this paper, the common single-mode biometrics and the traditional multi-modal biometric fusion strategies are studied. Finally, two kinds of single-mode biometrics, fingerprint and iris, are adopted in this paper. Experiments of fusion recognition are carried out at the level of feature layer in multimodal biometric recognition. The main work is summarized as follows: 1. The related methods of multimodal biometric recognition data fusion at various levels are deeply understood and studied, including series-parallel fusion and (CCA) fusion based on canonical correlation analysis, fractional fusion based on least square method and Fisher discriminant, etc. The weighted method and the majority voting method are merged at the decision-making level. 2. On the basis of in-depth research on multi-modal biometric fusion, a feature layer fusion model based on fingerprint and iris is proposed, which is compared with the existing feature layer fusion strategy. The conclusion that the recognition rate of multimodal fusion recognition is higher than that of single mode recognition is proved by experiments, and the validity of fusion algorithm based on canonical correlation analysis in multi-modal biometric recognition is verified. Aiming at the shortcomings of the fusion algorithm based on canonical correlation analysis, a multi-modal biometric recognition algorithm based on matrix transform for discriminating canonical correlation analysis (MDCCA) is proposed, and the two algorithms are tested in the same environment. Experimental results show that the algorithm is effective. 4. 4. On the basis of the new algorithm proposed in this paper, a complete multimodal biometric recognition process is completed. The algorithm studied in this paper has some reference value in the field of data fusion and multi-modal biometric recognition.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】