【作者】 郑文侃；

【导师】 陈辉煌； 洪景新；

【作者基本信息】 厦门大学 ， 计算机应用， 2008， 硕士

【摘要】 在模式识别与图像处理研究领域中,人脸检测是一个备受关注的研究方向,也是人脸识别的关键步骤。随着智能化信息处理技术的发展,人脸检测的应用背景已经远远超出了人脸识别系统的范畴,它在安全访问控制、智能监控、视频会议、基于内容的图像检索等方面有着日益广泛的应用前景。然而,人脸作为人类自身再熟悉不过的一个组成部分,其检测问题却是个极富挑战性的课题。如果能找到高性能的解决方法,将为解决其它类似的复杂模式检测问题提供重要的启示,这使得研究快速而准确的人脸检测技术具有了十分重要的意义。本论文研究的是较复杂场景的灰度图像中竖直正面人脸的检测问题,提出了量化角特征与AdaBoost方法相结合的人脸检测算法。论文首先对人脸检测的一些相关技术做了总结、比较和研究。然后对于基于统计模型的人脸检测方法做了较为详细的分析,重点分析了量化角特征和AdaBoost如何应用于学习分类。在论文的第三章,我们根据量化角特征(Quantized Angle Features)的思想,进行将该原理实用化的数学公式推导,并在此基础上进行统计分析,确认结果可用于模式分类。此后,在第四章中对AdaBoost算法进行改进,实现了在AdaBoost算法中可以使用量化角特征。第五章中给出了人脸检测系统的完整实现,并对检测中的缩放方法以及相应的参数进行了实验比较,同时还给出了系统运行的结果。实验结果表明,本论文的人脸检测算法能够有效地检测出较复杂场景的灰度图像中正面的人脸图像。论文最后对系统的运行效果进行综合分析,总结了本文工作的创新点,并指出进一步的工作。更多还原

【Abstract】 Human face detection is an important problem in pattern recognition and image processing. And it is the key step of face recognition. With the development of intellectualized technology, human face detection has wide applications in many fields such as security access control, intelligent supervision, video conference and content-based image retrieval. Human faces are familiar to us, but theirs detection is a challenge. If we can find a high-performance solution, it will offer important enlightenment for solving other similar complicated pattern recognition problem. It makes studying fast and accurate human face detection technique have important meaning.In this thesis, we present an algorithm based on Quantized Angle features and AdaBoost to detect frontal views of human faces in sequential gray images.We make summaries, comparison, and researches on correlative technologies at the beginning of this thesis. Then we make detail analyse at face detection method base on statistical model, analyzes common face detection methods and how Quantized Angle features and AdaBoost be applied to learning and classifying.According to the concept of Quantized Angle Features in chapter 3, we make the principle to be the practical mathematical formula. Then we apply it to statistical analysis, the result show that features can be applied to pattern recognition. In chapter 4 we improve the algorithm of AdaBoost so that it can be combined with Quantized Angle Features. In chapter 5, we introduce the realization of face detection and show the result, and show the experiment result of scaling method and corresponding parameter in detection system. The experiment result shows that the algorithm works efficiently in the gray images of complex scenes.At last, we carry on comprehensive analysis to the operation result of system and point out the further work.更多还原