摘要:为了改善在光照变化和复杂背景影响下的人脸检测效果，在预处理阶段提出一种基于直接最小二乘拟合的三维肤色聚类模型算法。该算法首先将肤色在CbCrCg空间中的三个平面投影分布作为拟合对象，然后使用中值滤波和Sobel算子获取平滑边缘，最后通过直接最小二乘拟合法获取最佳三维肤色模型。在实验中分别将公共人脸库和户外拍摄的人脸图像作为实验对象，实验结果表明，该算法较传统肤色预处理算法具有更好的肤色分割效果，并且能够有效地提高人脸检测率。
关键词:人脸检测；肤色聚类；最小二乘法；CbCrCg空间； 三维肤色模型�
中图分类号: TP391.41 文献标志码:A
�
Face detection pre.processing method based on three.dimensional skin color model
�
SUN Jin.guang, ZHOU Yu.cheng��*, MENG Xiang.fu, LI Yang
School of Electronic and Information Engineering, Liaoning Technical University, Huludao Liaoning 125105, China
Abstract:
In order to improve the face detection test results under the influence of illumination change and complex background, an algorithm of 3 dimension color clustering model based on direct least squares estimate was proposed during the preprocessing phrase. Firstly, three plane projection distribution of skin color was seen as fitting object in CbCrCg space, and then smooth edge was get by median filter and Sobel operator, at last the best 3 dimension color model was get through direct least squares. In experiment, the public face library and face image which is get by outdoor shooting were seen as objects, and the experiment results show that, this algorithm has better segmentation effects than traditional color preprocessing algorithm, and it has impr-oved the detection rate more effectively.
In order to improve the face detection test results under the influence of illumination change and complex background, an algorithm of 3D color clustering model based on direct least squares estimate was proposed during the preprocessing phrase. Firstly, three plane projection distributions of skin color were seen as fitting objects in CbCrCg space, and then smooth edge was got by median filter and Sobel operator, at last the best 3D color model was got through direct least squares. In experiment, the public face library and face image got by outdoor shooting were seen as objects, and the experimental results show that, this algorithm has better segmentation effects than traditional color preprocessing algorithm, and it has improved the detection rate more effectively.
�Key words:
face detection; skin color clustering; least square; CbCrCg space; 3D skin color model
�
0 引言�
肤色检测是人脸检测预处理过程中的一个重要过程。但由于光照和背景变化等复杂环境的影响，使得各种肤色模型的建立存在较大的局限性，这种局限性会造成肤色的误检或漏检现象的产生，进而直接影响人脸检测的效果�［1］。�
目前人脸检测预处理过程中肤色模型的建立方法有很多，由于肤色在空间表现出非常好的聚类性，传统算法通常采用简单模型来描述这个分布�［2-3］。但由于复杂光照和背景的影响，这些简单模型并不能取得较好的检测效果，基于上述问题，文献［4-5］应用亮度分量与色度分量独立性的特点，采用CbCr，CgCr二维模型以提高肤色模型的光照鲁棒性，但同时造成了大量非肤色信息与肤色信息的重叠。针对该问题，孔潇等�［6］采用CbCgCr三维空间粗检测与模糊聚类二次分割相结合的方法；易轶虎等�［7］通过计算不同色度沿亮度的概率分布，提出了一种基于参数查表的肤色检测方法。上述两种方法虽然能够有效提高肤色检测的自适应性和准确性，但与此同时造成了存储和计算量的增加。�
针对以上问题，本文采用一种三维肤色模型的方式进行肤色模型的建立。该算法将CbCrCg空间三个平面CbCr，CgCr，CbCg的投影分布作为拟合对象，采用中值滤波和Sobel边缘检测获取平滑边缘，进而应用直接最小二乘拟合方法得到各平面准确的拟合结果，实现模型建立。�
1 空间选取及肤色模型建立�
1．1 颜色空间选取�
对于肤色检测，颜色空间的选取是关键的一步�［8］。常见的肤色分割空间有RGB、HIS、HSV、YCbCr等，其中YCbCr空间由于具有亮度与色度分离的特性及较好的肤色聚类性，是最常用的颜色空间。相比YCbCr空间， de Dios提出的YCgCr 空间肤色聚类性更好�［9］，而且充分体现了肤色中蓝色分量相对较少的特征。但由于光照的影响，肤色在以上两个空间并非简单的二维关系，即使同一肤色点，由于光照影响，可能处于三维空间相离很远的两个位置，因此简单地在二维平面上投影处理，容易造成肤色和非肤色点的重叠�［10］。本文采用CbCrCg颜色空间，避开亮度信息，通过三个平面肤色分布拟合情况来研究肤色的三维空间分布模型，提高肤色模型对光照和复杂背景的鲁棒性。由RGB空间转化到CbCrCg颜色空间的公式如下：�
�
Cb�Cr�Cg=128�128�128+�
1256-37.793-74.203112.000�112.000-93.786-18.214�-81.085112.000-30.915R�G�B(1)�
�
为涵盖不同光照、年龄、人种的肤色样本信息，本文通过对互联网及AR人脸库彩色子集进行整理，剪切形成257张肤色图片，然后根据同一张图像中肤色点具有连续性的特点，对剪切样本进行去噪处理形成1�037�568个肤色像素样本。其在CbCrCg颜色空间分布情况如图1所示。�