摘要
人脸识别是当前模式识别应用的一个重要领域。在理解当前广泛使用的各种人脸识别算法的基础上,提出了一种基于多方法融合的彩色图像纯脸检测与定位的优化算法。该算法首先通过肤色检测及预处理技术缩小彩色图像人脸检测的搜索区域,然后应用基于物体区域方向的检测平面内任意旋转角度人脸计算方法;并通过二次计算旋转角度的方法来准确确定人脸区域的旋转角度;最后利用积分投影函数找到人脸候选区域中双眼的位置,结合人脸特征在人脸中的比例关系准确确定纯脸的位置。该算法还考虑了人脸侧偏时的情况。实验表明,本方法对平面内任意旋转及双眼存在的侧偏人脸有较好的检测效果,对不同光照条件有较好的鲁棒性。
Face recognition has been a complex and difficult problem that is important for surveillance and security, telecommunications, digital libraries, and human-computer intelligent interactions. Based on understanding wide variety algorithms, this paper presented a Multi-way integration optimization algorithm for detection and localization of pure face in color images. First,To an input color image, skin color detection and preprocessing rules was adopted firstly to reduce the face detection searching regions. Secondly a method that can detect arbitrary rotated face within the image plane was proposed based on the computation of the direction of the object. Thirdly, to determine the angle of the rotated face region accurately,a method that compute the rotated angle again was proposed. Finally, the two eyes in the face candidate region was determined by projection funetion, and then the ＂pure face＂ position combining with the proportion of the face features was determined, the side-view face was also considered. The experiments showed that the proposed method provides more better detection results to arbitrary rotated faces within the image plane and side-view faces with two eyes, and more better robust under different illumination.
出处
《计算机科学》
CSCD
北大核心
2009年第7期284-287,共4页
Computer Science
基金
江西省教育厅科技项目(赣教技学[2007]233号)资助
关键词
人脸检测
模板匹配
积分投影
颜色空间
纯脸
Face detection, Template matching, Integral projection, Color space, Pure face
