一种基于标准混合高斯模型的快速人脸检测方法

Method for fast face-detection based on normalization Gaussian mixture model

针对室内移动机器人运动过程提出一种快速而稳定的人脸检测方法.由于室内存在多种物体,背景不断变化,且光照条件可能不断变化,提出采用人脸肤色的标准混合高斯模型与人眼特征相结合的人脸检测法,无需对原始图像进行尺度变换.检测过程首先将经过补光处理及光线增强的人脸库转换到YCbCr空间,求其非线性变换空间YCb′Cr,′求出左右脸标准正态密度函数及混合高斯分布;然后根据人眼颜色特征,分别对人脸肤色候选区域进行人眼候选区域提取,利用人眼Gabor模板的不变Hu矩与人眼候选区域的相关性,找出人眼拟合矩形区域,再综合利用人眼与人脸的特征关系以及人脸候选区域的投影关系检测出人脸区域.大量实验表明,新方法速度快,适应性较好,并可扩展检测到侧面人脸.

An effective and fast method of face detection for a service robot is proposed, which combines a mixture of Gaussian distribution model of skin tone color with eyes features. Candidate skincolor regions over the entire image are extracted based on normal Gaussian distributions models of left and fight faces in a nonlinearly transformed YCb Cr color space, following the light compensation and enhancement procedures. Eyes candidates are located in the candidate skin regions by using different contrasts in color between the eyes and face skins, and Gabor filtered Hu-moments of those are produced as statistical features for further identification of the eyes. Combining the projection histogram of skin-color regions and detected eyes features, the faces regions are accurately located. Experimental results demonstrate that the proposed method is effective and fast to detect both frontal and profile faces with robustness of lighting fluctuations and background clutters.