【作者】 熊飞；

【导师】 张桂林；

【作者基本信息】 华中科技大学 ， 模式识别与智能系统， 2007， 硕士

【摘要】 人脸检测识别技术是图像处理及模式识别领域的重要课题之一。在身份确认应用方面有重要的作用。本文通过对人脸检测识别系统的分析,在立足于算法的角度上,将整个系统分为,人脸检测、人脸校正、特征提取、分类识别四个部分。通过研究生阶段的研究以及与他人的合作,在人脸检测模块中,本文分析了传统的人脸检测算法,并采用了基于Adaboost的改进人脸检测算法检测出视频序列中的人脸图象;在人脸校正模块中,本文根据检测到的人脸图像中眼睛的位置校正人脸图象,针对传统眼睛定位算法的优势和缺陷,提出了基于Gabor滤波多频段融合技术的人脸校正算法,并以此为基础校正人脸图象;然后,本文应用40个Gabor滤波器与人脸图像卷积得到的Gabor特征为分类识别的特征向量;最后采用R-LDA作为系统的分类确认器,巧妙的解决了Fisher线性判别中遇到的小样本问题。本文最终实现了人脸自动检测确认系统的原型系统的搭建。该系统能够自动检测人脸图像,并对检测到的人脸图像做校正归一化处理,并提取特征向量,最后交给分类器进行身份的确认。通过实验证明,本文采用和提出的各个模块的算法是具有良好的鲁棒性,较其他一些传统的算法有明显的优势。以此为基础搭建的软件系统,也能够符合人脸自动检测确认任务的需要。更多还原

【Abstract】 Face recognition and verification is an important field in image processing and pattern recognition research. It is valuable on the identity verification aspect.This paper divides the system into four parts via the detailed analysis of face verification systems, which are face detection, face image alignment, feature extraction, classification. Through the research in graduate study and the collaboration with others, the traditional face detection algorithm is analyzed and the Adaboost based algorithm is applied to detect the face images in the video; in the face image alignment model, the face image is aligned based on the eye position, and aiming at the advantages and disadvantages the Multi-band Gabor filtering fusion is presented for localizing eyes and aligning the face images; then a set of 40 Gabor filters is applied to extract the Gabor feature of the aligned face images; lastly the R-LDA is applied as the classifier to solve the small sample size problem in the Fisher criterion and finish the verification task. At last, a prototype of face verification system is finished. This system can automatically detect the face images, and align them based on the eyes’position, and verify the people based on the Gabor feature extracted from the aligned face images.Through the experiments, it is proved that the algorithms presented and applied in this paper are robust, and has obvious advantages compared with the traditional algorithms. And the software based on those algorithm can complete the automatic face detection and verification task.更多还原