CASIA OpenIR  > 毕业生  > 博士学位论文
基于统计学习的多模态生物特征识别
其他题名Multi-Modal Biometric Recognition Based on Statistical Learning
楚汝峰
学位类型工学博士
导师李子青
2007-06-15
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业计算机应用技术
关键词生物特征识别 人脸识别 眼睛状态检测 近红外人脸识别 掌纹识别 多模态生物特征识别 Biometrics Face Recognition Eye-state Detection Near-infrared Face Recognition Palmprint Recognition Multi-modal Biometrics
摘要随着信息化的不断加快,国家公共安全、信息安全等关系国计民生的领域需要高可靠性、高安全性的全新身份识别技术,生物特征识别技术应运而生。但是,仅靠单一生物特征模态(比如人脸、虹膜、指纹、掌纹等)的识别技术,往往由于其自身的一些缺陷和不足,有时难以满足人们对身份识别高准确性和易用性的要求。多种生物特征的有效融合可从多方面改善身份识别系统的性能。 本文以人脸生物特征模态为基础,对基于统计学习的多模态生物特征识别算法与系统进行了研究。论文的主要工作及贡献如下: 1) 研究了生物特征识别相关的一些基础算法。探讨了基于统计学习的眼睛定位、眼睛状态检测和近红外人脸识别算法,在实际应用中取得了很好的性能。提出了一种基于Gabor幅度特征和统计学习的掌纹识别算法,与现有的方法相比,取得了更好的实验结果。 2) 提出了一种基于近红外和可见光图像的多模态人脸识别算法。分析了近红外和可见光模态人脸图像的特点,提出了基于分数层融合的多模态人脸识别算法,并构建了相应的原型系统。在室内场景和室外场景两种典型的应用环境中,对该融合算法与系统进行了深入分析和性能评估。 3) 基于排序测度特征表示的人脸和掌纹识别技术,初步研究了人脸和掌纹多模态分数层融合算法。在大规模数据库上对人脸和掌纹融合算法进行有效性测试。分析了不同的分数归一化方法对识别的影响,并测试比较了不同分数层融合方法的性能,为后续研究提供了有意义的参考。 4) 提出了一种基于统计学习的人脸和掌纹特征层融合算法。通过合理选择有效的局部特征表示和统计学习方法,有效地利用了人脸和掌纹两种模态中最具互补性的局部特征,构建了高效的多模态融合识别分类器。该算法不仅可以有效提高单一模态生物识别系统的性能,而且可以提高系统的运算效率。
其他摘要Currently, most biometric systems deployed in real-world applications are uni-modal, relying on the evidence of a single source of biometric information for authentication. Such systems cannot meet desired performance requirements. To overcome such difficulties, multi-modal biometric systems are developed in which evidences from multiple sources of information are integrated to improve the performance. In this thesis, we investigate the statistical learning-based multi-modal biometric methods and systems, especially methods of fusion with the face modality. The main contributions are as follows: 1) Statistical learning-based eye localization, eye status detection and face recognition methods are proposed. Extensive experiments are provided to evaluate the proposed methods and demonstrate the good performance of proposed methods. In addition, a novel Gabor magnitude feature-based method is presented for palmprint recognition. Based on statistical learning, we propose to use Gabor magnitude features for palmprint representation. Experimental results on three large palmprint databases demonstrate the effectiveness of proposed method. Compared with state-of-the-art Gabor-based methods, the proposed method achieves higher accuracy. 2) By analysis of the characteristics of near-infrared and visual face images, we explore a multi-modal face recognition for indoor and outdoor scenarios, and provide experimental results to demonstrate the effectiveness of such fusion scheme. 3) We investigate a method for face and palmprint multi-modal biometric identification to improve the identification performance. Effective classifiers based on ordinal features are constructed for faces and palmprints, respectively. Then, the matching scores from the two classifiers are combined using several fusion strategies to give a unique matching score. Experimental results on a middle-scale data set have demonstrated the effectiveness and improvements over the unimodal systems. 4) We present a learning approach for fusing face and palmprint at feature level for personal identification. First, we propose to use Gabor features as the unified representation for faces and palmprints. Second, we utilize statistical learning to fuse two modalities by selecting most effective and complementary features from the large number of concatenated features and construct a powerful classifier. Experimental results on two large multi-modal databases demonstrate the effectiveness of the proposed system, which uses fewer features and achieves better performance than that of the uni-modal systems.
馆藏号XWLW1149
其他标识符200418014628011
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/6021
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
楚汝峰. 基于统计学习的多模态生物特征识别[D]. 中国科学院自动化研究所. 中国科学院研究生院,2007.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20041801462801(3665KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[楚汝峰]的文章
百度学术
百度学术中相似的文章
[楚汝峰]的文章
必应学术
必应学术中相似的文章
[楚汝峰]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。