基于adaboost的人脸检测改进算法-应用数学专业论文.docxVIP

本文共采用了来自Internet和现实生活中的400张彩色人脸图片，背景各异，人 本文共采用了来自Internet和现实生活中的400张彩色人脸图片，背景各异，人 脸姿态方面存在一定的平面内旋转和平面外旋转，另外有一定的光照影响。试验结果表 明，本文改进的AdaBoost人脸检测算法能达到90％以上的检测率，而且对各种图片的 自适应性得到了提高。 关键词：人脸检测，肤色分割，AdaBoost，小波变换 Il Face Face Detection Based on Improved AdaBoost Algorithm Major：Applied Mathematics Name：Liu Chang Supervisor：Lai Jian-Huang Abstract Face detection is a necessary first—step in face recognition systems，with the purpose of localizing and extracting the face regions from the complex background．Because of its great application value in surveillance systems，content—based image detection systems，intelligent human．computer interfaces and so on，it has been approved to be an independent subject and received wide attention from a great many of researchers in these years．In 2001，Viola and Jones introduced a framework to elegantly combine AdaBoost with a cascade scheme to detect faces in real time．But there ale some problems when we use the detector generated by AdaBoost，because the computation complexity will be greatly influenced by the choice of searching areas and moving steps of the rectangle window,and also by the resolution of the image． This paper mainly aims at improving the flexibility of AdaBoost algorithm in face detection，and make this face detector work more efficiently and accurately．There are four main original ideas in this paper： (1) A detailed survey on AdaBoost face detection algorithms and comparation among different improved AdaBoost algorithms． (2) A new method of skin color segmentation is introduced before the AdaBoost cascade structure，which greatly reduces negative samples that cascade stmcture will face，and also work for the sub-window location,move strategy and combination process． (3) Bi-orthogonal wavelet high frequency features are introduced after the AdaBoost cascade structure，which can avoid information lost problems that the I／／ simple simple haaplike features have． (4) A new strategy of the strong classifier threshold generation proce