形变虹膜图像的鲁棒特征表达与匹配
文献类型：学位论文
| 作者 | 张曼
|
| 学位类别 | 工学博士 |
| 答辩日期 | 2013-06-01 |
| 授予单位 | 中国科学院大学 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 谭铁牛 |
| 关键词 | 生物特征识别 虹膜识别 形变 特征提取 特征匹配 Biometrics Iris recognition Deformation Feature representation Feature matching |
| 其他题名 | Robust Feature Representation and Matching for Deformed Iris Recognition |
| 学位专业 | 计算机应用技术 |
| 中文摘要 | 随着人们对公共安全问题日益重视，生物特征识别技术也越来越受到关注。虹膜是人眼中圆环状的纹理区域，由于其唯一性、稳定性、非侵犯性和高防伪性，成为最适合用于身份识别的生物特征之一。作为一个实用性很强的课题，虹膜识别正在慢慢地从实验室走向社会应用。 在实际应用中由于噪声、光照变化、遮挡、形变等，采集到的虹膜图像通常质量较低，这给我们的识别系统带来了巨大的挑战。其中，虹膜形变会引起虹膜纹理的变化，产生较大的类内差异，是导致识别错误的主要原因之一。但是，目前并没有专门针对此问题的完善解决方案。本文将以现有的虹膜识别系统为框架，分析在实际应用中可能遇到的情况，从图像数据、特征抽取、特征融合和匹配策略四个不同的角度解决形变虹膜识别问题。本文的主要贡献如下： (1)从图像数据层面入手，将图像扰动方法引入虹膜识别。该方法以数据库中已有注册模板为基础，人为扰动生成大量新的图像，模拟实际采集过程中可能遇到的低质量图像，快速有效地扩大注册数据库规模，主要用于形变、旋转和模糊虹膜图像匹配。并且在实际应用过程中，采用了级联匹配方法，在识别正确率和计算效率中取得了平衡。 (2)从特征融合层面入手，提出了虹膜局部特征和全局特征融合方法。该方法对虹膜图像的Gabor特征进行相关滤波，充分利用了Gabor和相关滤波器互补的特性，更加全面地描述虹膜信息，将Gabor特征对噪声、对比度变化鲁棒等特点和传统相关滤波器的平移不变性、对局部遮挡鲁棒等特点有效地结合在一起。 (3)从特征表达层面入手，提出了基于多通道特征融合的形变虹膜识别方法。本方法将归一化虹膜图像分解成不同频带的分量，之后根据频带的不同特点抽取不同的虹膜特征并融合。带通分量抽取基于关键点的几何信息，并辅助抽取低通分量上的校正虹膜局部特征。该方法利用虹膜图像不同频带的不同性质，分别抽取不同的虹膜特征并融合，这样可以有效地利用各个频带的优势，从而提升识别正确率。 (4)从匹配策略层面入手，提出了基于DAISY特征的形变虹膜识别方法。本方法根据归一化虹膜图像的密集多维DAISY特征，确定图像关键点，最后利用动态匹配方法快速有效地匹配形变虹膜图像。DAISY特征可以有效描述虹膜图像的细节信息，而且据此特征选出的关键点更稳定。动态匹配方法运算速度快，且不需要任何关于虹膜形变的先验知识或者模型，即使虹膜形变非常严重，本方法也可以很好地解决此问题。 总的说来，本文从多角度分析了形变虹膜图像匹配问题并分别提出了解决方案，提升了形变虹膜识别正确率，在针对形变虹膜识别的鲁棒特征表达与匹配方面做出了有益的研究。 |
| 英文摘要 | With people's increasing attention for public security, biometrics has been subject to growing concern. Iris is the frontal annular part of human eyes and exhibits rich texture information. For its uniqueness, stability, non-intrusiveness and anti-counterfeiting, iris is one of the most suitable modalities for personal recognition. With the rapid development, iris recognition becomes more and more practical. In real world applications, low-quality iris images caused by noise, illumination changes, occlusions, deformation et al. bring huge challenges to iris recognition system. Especially, iris deformation can lead to changes of iris texture and large intra-class difference, which is one of the main reasons that cause recognition errors. However, the perfect solution for this problem does not exist. Our work is based on the existing iris recognition framework to handle low-quality iris image matching from the following aspects: iris image dataset, feature extraction, feature fusion and matching strategy. The main contributions are as follows (1)Image perturbation is introduced into iris recognition system. We artificially generate iris images to enlarge enrollment data set quickly and efficiently. Our approach aims at representing the deviations of input iris images and modeling the variety of iris images, for example, deformed, blurred, misaligned images et al. Meanwhile, two-stage cascaded perturbed image matching strategy is applied to save computational time. (2)Fusion of Gabor features and correlation filters is proposed. Gabor features are robust to noise, illumination changes et al. while correlation filters are not only shift-invariant but also robust to occlusion. We used correlation filtering of Gabor images for robust correlation matching of iris images in Gabor feature domain which can make full use of their complementary characteristics and represent iris information comprehensively. (3)A novel deformed iris image matching method using bandpass geometric information and lowpass Ordinal features is proposed to address deformed iris image matching problem. In our work, normalized iris images are decomposed into multi-scale and multi-direction subbands. Key point based geometric features are extracted in bandpass subbands and aligned local features are extracted in the lowpass subbands. The proposed method extracts and fuses different features in different subbands to take fully advantages of subbands, which can increase reco... |
| 语种 | 中文 |
| 其他标识符 | 201018014629097 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/6552]  |
| 专题 | 毕业生_博士学位论文 |
| 推荐引用方式 GB/T 7714 | 张曼. 形变虹膜图像的鲁棒特征表达与匹配[D]. 中国科学院自动化研究所. 中国科学院大学. 2013. |
入库方式： OAI收割
来源：自动化研究所
浏览0
下载0
收藏0
其他版本
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。