MSPA-MCS: Face Recognition Using Integral Invariants and Cryptology

人脸识别的基本技术挑战是处理由于姿势、光照、年龄、面部表情、距离和其他因素而导致的面部外观的极其广泛的变化。在这些变化下，理想的面部特征应该保持不变。虽然目前的人脸识别方法在其工作条件下受到限制，但不变量数学理论的最新进展以及与密码学的最新联系为人脸识别问题提供了新的见解，这可能有助于扩大人脸识别系统的通用性和鲁棒性。在这个项目中，一个跨学科团队将探索一种独特的方法，将积分不变量数学的最新发展、人类面部识别的心理物理实验、最先进的面部识别算法和系统开发结合在一起。具体而言，我们计划解决以下研究目标：在密码学和对象识别之间建立一个共同的数学基础，特别是获取信息-人脸数据压缩的理论限制研究三维几何对象的积分不变量的数学理论；开发高效的不变特征提取和模式识别算法与专业漫画家互动，并使用实验心理物理证据来评估感知意义，并探索人脸识别背景下数学3D积分不变特征的生物学解释开发3D移动人脸识别的原型系统实现。特别是，我们将专注于在半控制环境中具有多个摄像头的场景，用于访问控制和监视名单定位应用。该项目采用的方法是前沿的，非常规的，并且有望在推进面部识别性能的理论和实践方面取得重大突破。广泛影响在这个项目中，我们计划举办一个夏季研究体验项目，为代表性不足的群体提供服务；开设一门新的生物识别信息学课程，让本科生参与开发部分原型人脸识别系统；举办人机面部识别比赛，以提高市民对这方面的认识，并吸引青少年参与科技。具体来说，在实验室和课程中，我们将为来自数学和计算机科学背景的学生营造一个学习和研究互动的环境，激发新的想法，并欣赏数学和计算机科学之间的相互影响和互动。

The fundamental technical challenge of human face recognition is to deal with the extremely broad range of variations in a face's appearance due to pose, lighting, age, facial expression, distance and other factors. Ideal facial features should remain invariant under these variations. While current face recognition methods are limited in the conditions where they work, recent advances in the mathematical theory of invariants and recent connections forged with cryptology provide new insights for the face recognition problem that may help broaden the generality and robustness of face recognition systems. In this project, an interdisciplinary team will explore a unique approach that brings together the latest developments in the mathematics of integral invariants, psychophysical experiments on face recognition by humans, and state-of-the-art face recognition algorithms and system development.Specifically, we plan to address the following research goals:Establish a common mathematical foundation between cryptology and object recognition, in particular obtaining information-theoretical limits to data compression of human facesInvestigate the mathematical theory of integral invariants for 3D geometrical objects, and develop efficient invariant feature extraction and pattern recognition algorithms Interact with a professional cartoonist and use experimental psychophysical evidence to assess perceptual significance and to explore biological interpretations of mathematical 3D integral invariant features in the context of human face recognitionDevelop a prototype system implementation for 3D moving face recognition. In particular, we will focus on a scenario with multiple camera in a semi-controlled environment for access control and watch-list spotting applicationsThe approaches taken in this project are cutting-edge, unconventional, and promise significant breakthrough in advancing both the theory and practice of face recognition performance. Broad ImpactsIn this project we plan to hold a summer research experience program to serve under-represented groups; to develop a new biometric informatics class that involves undergraduate students in developing parts of prototype face recognition systems; to hold a man-machine face recognition contest to garner public awareness of this area and to attract young people to science and technology. Specifically, in both laboratories and courses we will foster an environment for students from both mathematics and computer science backgrounds to interact in learning and research, to stimulate new ideas and to appreciate the mutual impact and interactions between mathematics and computer science.