【作者】 季长有；

【导师】 段其昌；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 人脸检测就是判断被检图像中的人脸,并确定其大小及位置。人脸检测是人脸识别过程中的必要前期工作,其主要作用是将人脸部分从背景图像中分离出来。一个对一般环境图像具有一定适应能力的人脸识别系统,需要一个鲁棒的、高效的、实时的人脸检测系统。近年来,随着计算机视觉技术和模式识别技术的发展,以及硬件速度的提高和成本的降低,基于生物特征的身份认证技术近年来发展迅速。与其它生物认证技术相比,人脸识别具有更直接、友好、方便的优点,是当前热门研究领域之一。本文以完成人脸识别前期准备为目标,利用肤色的聚类特性,设计并实现了一个复杂背景下的人脸检测算法。论文包括四方面的内容:源图像的采集、图像预处理、人脸肤色分割、人脸检测。论文首先应用海康威视的图像采集卡,完成了图像采集功能和云台控制功能;然后对实时获取的图像进行图像预处理,在图像预处理阶段依次对图像进行中值滤波、光线补偿、彩色图像的灰度化,并在Rein-Lien Hsu等人的光照补偿算法基础上提出了通过设定门限值来提高光照补偿效果。在肤色分割阶段,本文提出一种非线性变换的肤色聚类和边缘检测相结合的肤色分割算法,该分割算法首先利用非线性变换的肤色聚类性分割出图像中的肤色区域,然后根据各个区域的面积决定是否利用局部边缘信息来继续分割该区域。最后应用人脸检测算法对分割后的候选人脸区域进行检测和验证,本文提出基于知识方法的验证和多种形式的自适应模板检测方法相结合的人脸检测算法,并且对于人脸特征不明显以及部分遮挡的候选人脸区域,应用半人脸模板检测来避免漏检人脸。更多还原

【Abstract】 Face detection is to determine the size and location of a human face in an image which is being detected. Face detection is the necessary earlier task in face recognition process, aiming at separating the face area from the background image. Face recognition system adapted to some general environmental image to some extent needs a Robust、high efficient and real-time face detection system. In recent years, with the technological development of computer vision and pattern recognition, as well as the improving of hardware speed and cost reducing, the identity authentication technology based on biometrics has developed rapidly .Compared with other biometrics technology, face detection is more direct、much friendlier、and more convenient and it is one of the hot research fields, drawing more and more attention.This paper sets completing earlier preparation for face recognition as its objection, using skin color clustering, and it has designed and realized a human face detection algorithm under a complex background. This paper includes four aspects: image capture、image preprocessing、segmentation of face skin color and face detection. First, image capture and pan-tilt control are firstly achieved based on HiKivision’s image capture board. And then it comes to preprocessing the real-time images, during which median filter, lighting compensation and image gray processing are to be realized successively, what is more, a threshold value is presented to improve compensation effect on the base of light compensation algorithm of Rein-Lien Hsu etc. During the phase of skin segmentation after pretreatment, a non-linearity transform method of skin color segmentation is proposed, combining skin color clustering and edge detection. It first makes use of the skin color clustering characteristic of non-linearity transform to segment different skin color regions. Then it needs to decide whether local non-linearity information should be applied to go on segmenting the regions depending on their areas. Finally it detects and validates the candidate face region that has been segmented by human face detection algorithm. This paper puts forward a face detection algorithm which combines the detection method based on knowledge and self-adaptive template matching methods. As for the face region where the features are not obvious or some of which is covered, half-face template should be used to avoid missing detection of face region.更多还原