自适应PSO融合的多模态生物特征识别方法

【作者】 方中杰；

【导师】 章毅；

【作者基本信息】 电子科技大学 ， 计算机软件与理论， 2007， 硕士

【摘要】 生物特征识别技术是利用个体所固有的生理和行为特征来进行身份鉴定的技术。与传统的身份验证手段相比，生物特征识别技术具有防伪性好、便于携带、不易丢失、不易遗忘的优点。虽然单一的生物特征识别技术(单模态)具有众多优点，但是每种单模态识别的准确率是有限的，都存在各自的缺点，适合应用于不同的场合。多模态生物特征识别是融合多种生物特征对个体进行身份验证的技术。通过多生物特征融合的方法，可以提高生物特征识别系统的准确率等性能，因此具有广阔的应用前景，是当前生物特征识别领域研究的热点。本文提出了自适应PSO融合算法，在决策层上解决多生物特征融合的问题。在本文提出的方法中，多模态的融合被构造成贝叶斯决策融合的形式，自适应PSO融合算法可以最小化融合系统的贝叶斯风险，搜索得到最优的多模态融合决策规则，从而构造出一个最优的多模态融合系统。为了提高融合算法的性能，本文进一步提出了最小速率限制的二进制PSO算法AIS-MVLBPSO，该算法采用最小速率阈值对PSO颗粒的搜索速率进行限制，从而可以有效改进算法的收敛能力。为了验证自适应PSO融合算法的有效性，本文使用自适应PSO融合算法融合人脸和指纹两种单模态的生物特征识别方法，并在ORL、UMIST人脸数据库以及MCYT指纹数据库上设计实现了对比实验。实验表明，多模态生物特征识别系统的性能要优于人脸、指纹两种单模态识别系统的性能。自适应PSO融合算法可以根据单模态识别方法的性能差异，自动选择最优的融合决策规则来最优化多模态系统的识别结果。根据自适应PSO融合算法，本人设计实现了多模态生物特征识别系统MultiBIS，本文对MultiBIS系统的设计框架、系统实现、系统性能等作了详细介绍。更多还原

【Abstract】 Biometrics is the technology which refers to identifying an individual based on hisor her physiological or behavioral characteristics. Compared to traditional identificationand verification methods, biometrics is more convenient for users, reduces fraud, andcan not be forgotten or replaced. Biometrics has been proven to be successful and hasbeen applied in some fields, however, each single biometric modality (unimodal) has itsadvantages as well as drawbacks, and the error rates associated with unimodal biometricsystem are quite high which makes them unacceptable for deployment in securitycritical applications. Some of the problems that affect unimodal biometric system can bealleviated by using multimodal biometric traits. Systems that fuse multiple cuesobtained from two or more biometric indicators for the purpose of person recognitionare called multimodal biometric systems. Multimodal strategy and fusion scheme cansignificantly improve the overall accuracy of the biometric system. Multimodalbiometrics has been receiving a lot of attention in the recent years.This dissertation presents an Adaptive Particle Swarm Optimization Fusion(APSOF) algorithm which can fuse multiple biometric modalities at the decision level.The fusion problem is designed as a Bayesian decision framework and the APSOFalgorithm can automatically adjust the optimum decision fusion rules to minimize theBayesian error cost for the fusion system. To improve the performance of the fusionalgorithm, this thesis also proposes a novel Minimum Velocity Limited PSO. Theminimum velocity strategy applies a threshold to control the flying velocities of PSOparticles, thus improves the convergence ability and stability of the algorithm.To demonstrate the performance of APSOF algorithm, this dissertation uses thisfusion scheme to fuse two biometric modalities face and fingerprint. The experimentsare designed and carried out on the ORL, UMIST face database and MCYT fingerprintdatabase. Experimental results show that multimodal fusion scheme outperformsunimodal system based on face or fingerprint. It is proven that APSOF algorithm canselect the optimum decision fusion rules adaptively according to the variation of theaccuracy of unimodal system. Based on APSOF algorithm, I developed a Multimodal Biometric Identification System (MultiBIS). This thesis describes the design,implementation and functions of MultiBIS.更多还原