基于人脸识别的身份认证方法研究-信号与信息处理专业论文.docxVIP

中国科学技术大学博士学位论文 中国科学技术大学博士学位论文 中文摘要 ·提出了一种在复杂背景下从彩色图像中直接定位人眼的新算法。通过运用肤色信息缩小 检测人眼的搜索区域，进～步提高了人眼定位的准确性和速度。 ·提出了将小波分解后低频子带图像与其灰度积分投影相结合的人脸特征提取算法。该算 法兼有小波分解所提取的人脸特征容量较少、对人脸表情变化不太敏感的优点，以及灰度积分 投影能够较好地反映出人脸器官特征局域性的特点，是一种更有效的人脸特征提取方法。 ·提出了基于活跃集迭代法的C3一SVM(批)增量学习算法。对大规模分类问题，极大提 高了学习速度，克服了基于最优化直接求解方式的常规SVM学习算法速度相当慢的缺点。 ·提出了新的准正面人脸认证与鉴别方法，设计并实现了准正面人脸认证模型系统。实验 测试结果表明所提出的方法及系统在学习速度、识别速度与识别率方面均具有良好的性能。 关键词： 人脸认证，人脸鉴别，人脸识别，人眼检测与定位，支持向量机，人脸特征提取。 Ⅱ 中国科学技术大学博士学位论文 中国科学技术大学博士学位论文 英文摘要 ABSTRACT This dissertation deals with the researches on novel face recognition algorithms for human identification including near-frontal face verification and identification．The algorithms consist of three parts：automatic localization of human eyes in face images with complex backgrounds and normalization of face images，facial feature extraction by wavelet decomposition and integral projections techniques，and classification of facial features by support vector machines．The main work and contributions ofthe dissertation are summarized as follows： Detecting and locating human eyes加grayscale still images．For almost all the existing algorithms for detecting and locating eyes in grayscale still images with complex backgrounds，face location must be known before detecting and locating eyes．In addition，most ofthose algorithms are often associated with problems such as slow convergence and lengthy processing time．A novel algorithm is presented，in this dissertation，to achieve fast，accurate eye detection and localization in grayscale still images with complex backgrounds based on geometrical facial features and image segmentation．There is no need to know the face location before the proposed algorithm can be applied to detecting and locating eyes． Detecting and locating human eyes in color still images Traditional algorithms for face detection using skin color models are not effective where facial color appearance is