针对彩色图像提出基于四元数的人脸检测算法,从数学上解决传统方法将肤色检测与人脸最后确定分开进行而导致检测率较低、程序复杂度高的弱点,将人脸肤色信息与反映轮廓形状的灰度信息进行融合同时作计算处理,提高了检测速度,且增强了算法的鲁棒性.首先简单概述作者提出的检测系统,然后按算法流程分块说明系统每部分的主要内容,其中包括:定义于实数域的广义瑞利商以及Fisher鉴别分析法在四元数体上的推广;基于四元数的Fisher鉴别分析法在人脸检测中的应用原理;彩色人脸样本集的收集与预处理;彩色非人脸样本的收集及其分类的原因和分类原理;多尺度检测中的融合问题;多组检测结果的综合判断问题.数学上严密的推导表明,基于四元数的检测算法合理可行,满足实时性要求.大量实验测试结果显示,此算法快速有效,具有较强的鲁棒性.
A novel color face detection algorithm based on quatemion is proposed in this paper. Compared with conventional face detection method, which has relatively low detection speed based on the separation of face color detection and face confirming, our method improves the detection speed and advances the stability through simultaneously making use of face color information and gray information which reflects the contour of face. In this paper, the simple introduction of the detection system is given firstly, then every part of our detection system is presented in detail according to the flow-chart, which includes: the generalization of general Rayleigh-quotien and Fisher discriminant algorithm defined in real field; the application principle of Fisher discriminant algorithm based on quatemion in color face detection method;the collection and pre- processing of color face sample set; the collection of color non-face sample set and the principle and reason of its classifying; the fusing problem of multi-scale detection. The detailed mathematical demonstration indi- cates that our detection algorithm based on quatemion is rational and feasible, and it satisfies real time de- mand. The tests based on a large amount of samples show that our method has high detection speed, high correct detection rate, and is very robust.