【作者】 庄连生；

【导师】 庄镇泉； 郭立；

【作者基本信息】 中国科学技术大学 ， 电路与系统， 2006， 博士

【摘要】 人脸识别是图像处理和模式识别领域的一个研究热点，它涉及到模式识别、图像处理、计算机视觉、机器学习等多个学科领域。人脸识别和认证技术在国家安全、公安系统和城市的公共安全等方面有着十分重要的应用前景。虽然经过了近40年的发展，但是，由于人脸是非刚性物体，并容易受到光照、姿态、表情等因素的影响，要实现一个高识别率、高鲁棒性的全自动人脸识别系统仍然是一个极具挑战性的课题。 本论文的研究围绕着光照问题展开，以提高人脸识别系统对光照的鲁棒性和识别率为主要目标，对人脸图像预处理、特征点定位、特征提取等关键环节展开研究，探讨光照问题的解决方法。本文的主要工作和创新点包括如下几个方面： 1、研究了光照归一化问题，提出了基于全变分模型的光照预处理算法 光照是影响人脸识别的主要因素之一，人脸图像因为光照条件的变化而产生的改变甚至超过由于不同个体之间差异而产生的变化。本文在对图像成像过程进行分析的基础上，根据辐照度光照模型，提出了基于全变分模型的光照预处理算法。实验表明，该算法可以有效地减少光照对人脸识别算法的影响，提高人脸识别系统的识别率和光照鲁棒性。 2、研究了人脸的特征点自动定位问题，提出了复杂光照条件下鲁棒的特征点定位算法 人脸特征点定位是人脸识别系统的一个关键环节，特征点定位的精度与鲁棒性直接影响着人脸识别系统的识别率和鲁棒性。主动形状模型(ASM)是一个重要特征点人脸对齐方法，但是其定位结果比较粗糙。而弹性图匹配算法则能够实现特征点的精确定位，但是其计算复杂度高，定位速度慢。本文在分析两者优缺点的基础上，提出了一种利用ASM和弹性图匹配算法实现人脸特征点的自动精确定位的方法。同时，为了克服ASM算法对光照和模型初始位置敏感的缺点，本文还提出了一种改进型的ASM算法。该算法引入了两种特征，即相位一致性特征和反射系数特征。在搜索过程中，首先利用相位一致性特征实现特征点粗定位，然后在更多还原

【Abstract】 Face recognition is an active research topic in image processing and computer vision, which refers to pattern recognition, machine learning and so on. Face recognition has promising applications in public security, intelligent surveillance etc. Though face recognition has achieved great progress in the last 40 years, it is still a great challenge to build an automatic, high performance, high robust system for face recognition, due to the influence of illumination, pose and expression etc.One important purpose of this dissertation is to improve the recognition accuracy and robustness of face recognition system under various lighting conditions. To achieve this purpose, the focus of our work is on the image preprocessing, feature point location and feature extraction. The main work and innovation of this dissertation includes:1. The illumination normalization problem is studied, and an illumination preprocess algorithm based on total variation model is proposed.Illumination is one of the factors that reduce the recognition accuracy of face recognition system. In most cases, the difference between two images caused by illumination change is greater than that caused by individual difference. According to the irradiance lighting model, we propose a preprocess algorithm of illumination based on total variation model. Experimental results showed that our method could reduce the effect of illumination, and improve the recognition accuracy and robustness of face recognition system.2. The facial feature location problem is studied, and a robust facial feature point location algorithm under variable illumination is proposed.Facial feature location algorithms are important for an automatic face recognition system. It affects distinctly the recognition accuracy and robustness of the face recognition system. Active Shape Model (ASM) is a population method for face alignment, but its location accuracy is rough. Elastic bunch graph matching (EBGM) is another import location algorithm. It can achieve a fine accuracy when locating facial feature points, while its computational complexity is high and convergence is slow. By analyzing the merit and demerit of ASM and EBGM, we proposed a更多还原