【作者】 郑祥磊；

【导师】 王宏伟；

【作者基本信息】 大连理工大学 ， 控制理论与控制工程， 2009， 硕士

【摘要】 人脸检测是人脸识别之前的重要和必不可少的步骤,用于确定人脸在视频图像中的位置和大小。随着科学技术的迅猛发展,以及人们对模式识别和计算机交互需求的发展,人脸检测技术也越来越受到人们的重视。本文是对复杂背景下人脸检测进行研究,在现有技术的基础上,提出了肤色模型与AdaBoost方法相结合的检测方法,肤色模型方面主要是提出了模糊自适应光照补偿方法,结合了不同方法改进了肤色分割和唇色检测的效果;之后采用增加矩形特征的AdaBoost方法训练了新分类器用于人脸检测。本文的主要工作包括:肤色是人脸很重要的特征,为了避免光照的影响,首先对图片进行了模糊自适应光照补偿,改进了原有的自适应补偿方法,使补偿之后的效果更易于二值化。之后基于YCbCr模型的基础上进行肤色分割,根据最大类间方差法和阈值调整法进行了肤色块的查找。为了增加肤色检测的正确度,在肤色检测的基础上增加了唇色检测,在唇色检测时,在原有唇色检测理论的基础上,结合了两种方法,并提出自己的唇色检测标准,使唇色检测更加准确。AdaBoost方法进行人脸检测是十分迅速的,本文在原有方法的基础上,增加了三种新的矩形特征,利用新的矩形特征和原矩形特征共同训练了新的分类器。之后将分类器加载到自己的程序当中,进行人脸检测。最后在分析了肤色检测方法和AdaBoost方法的优缺点之后,本文将两种方法有机的结合起来,形成一种混合型的人脸检测方法。实验结果表明,本文的方法有效,具有较高的检测性能。更多还原

【Abstract】 Face detection is a very important step before the face recognition. The size and the location of the face are detected in the process. With the rapid development of science and technology, as well as people’s demand for pattern recognition and computer-based interactive development, human face detection technology has also developed so quickly.The methods of face detection in complicated condition based on the current methods have been done some research in the paper. A new illumination compensation method which be called fuzzy adaptive light compensation has proposed and the effect of the segmentation and lip detection has improved in the paper. And new rectangle features have added to train a new classifier here.The main contributions are as followed:Complexion is an important character of face. Avoiding the effect of the light, a new light compensation method which uses the fuzzy adaptive light compensation has proposed to enhance the quality of the image, so it is easier to do the binarization.Then segmentation has been done by Gauss model in YCbCr space. Then the image has been binarized based on the OUST and threshold adjustment. When detecting the lip, a new detection standard has proposed based on the combination of two current methods.AdaBoost is a very popular method to do the face detection. New three rectangle features has been added and a classifier has been trained by both the old and new rectangle features. Then put the new classifier into the program and detect the human face.At last, a new method based on the current two methods was proposed, through analyzing the advantages and disadvantages of the current methods.The experiment proves that the method is effective and has a good performance.更多还原