Statistical Methods for Fingerprint Image Analysis

生物识别，或称生物识别，是指根据一个人的解剖或行为特征自动识别他。在各种生物特征(如人脸、虹膜、指纹、语音)中，基于指纹的身份认证历史最长，在法医和民用领域都得到了成功的应用。然而，目前的指纹识别系统在存在噪声和变形图像的情况下，特别是在指纹数据库非常大的情况下，性能是不够的。统计研究的三个领域受到影响，即函数数据的分析、多变量相关性以及空间和一般点过程。拟议的研究将开发和利用贝叶斯框架和相关的计算方案进行推理。该框架将用于解决以下具体问题：指纹特征检测、噪声和变形图像建模、指纹个性(即唯一性)评估和用于信息融合(多生物识别)的有效分布表示。指纹捕获技术的进步导致了新的大规模民用应用，如US-Access计划。然而，由于操作环境中存在的生物测定可变性的影响，以及在每个识别任务中必须执行大量的比较，这些系统仍然遇到困难。所提出的基于模型的方法可能会提高各种指纹处理任务的效率，最终将在实际操作环境中产生更好的识别性能。此外，这项研究将影响指纹证据的报告和用于识别嫌疑人的方式。拟议的研究增加了统计学在重要的计算机科学和工程应用中的作用，并为跨学科研究和协同活动提供了动力。从事建议课题的本科生和研究生都将培养进行科学研究所需的分析和计算技能。通过这种方式，拟议的研究有助于培养未来的科学家，在新兴和关键的生物识别领域工作。

Biometric recognition, or biometrics, refers to the automatic recognition of a person based on his anatomical or behavioral characteristics. Among the various biometric traits (e.g., face, iris, fingerprint, voice), fingerprint-based authentication has the longest history, and it has been successfully adopted in both forensic and civilian applications. However, the performance of current fingerprint recognition systems is inadequate in the presence of noisy and deformed images, especially when the fingerprint database is very large. Three areas of statistical research are impacted, namely, the analysis of functional data, multivariate dependence, and spatial and general point processes. The proposed research will develop and utilize a Bayesian framework and related computational schemes for inference. This framework will be used to address following specific problems: fingerprint feature detection, modeling noisy and deformed images, fingerprint individuality (i.e., uniqueness) assessment, and effective distributional representations for information fusion (multi-biometrics).Advances in fingerprint capture technology have resulted in new large scale civilian applications such as the US-VISIT program. However, these systems still encounter difficulties due to the effects of biometric variability present in operating environments and the massive number of comparisons that have to be executed in each identification task. The proposed model-based methods are likely to improve the effectiveness of various fingerprint processing tasks which will eventually yield improved identification performance in real operating environments. Further, this research will impact how fingerprint evidence is reported and used for the identification of suspects. The proposed research increases the role of statistics in important computer science and engineering applications, and provides an impetus for inter-disciplinary research and synergistic activities. Both undergraduate and graduate students working on the proposed topics will develop the analytical and computing skills required to perform scientific research. In this way, the proposed research helps in the creation of future scientists to work in the emerging and critical field of biometric recognition.