基于mean shift的头发自动检测

Hair Detection with Mean Shift

对于头发的自动分割进行试验研究,并提出自己的算法框架。算法利用mean shift和混合高斯模型,结合颜色,纹理,头发位置特征来检测头发。本方法分为三步。首先利用adaboost算法检测人脸及人眼,人脸及人眼的位置确定可进行人脸大小归一化及得到头发位置模板。其次,抽取头发特征向量并利用mean shift对头部区域所有像素进行聚类,得到聚类区域。最后,利用混合高斯模型判定聚类区域是否为头发区域,进而检测出头发。试验证明本文方法运行速度快,并可有效检测头发在简单及复杂背景下。

This paper present an algorithm for hair segmentation automatically. Our approach uses mean shift and Gaussian mixture model to detect hair combining color, texture and location feature. The approach is divided into three steps. Firstly, face and eye are detected. Face and eye detection allow us to normalize the face sizes so hair location mask can be used. Secondly, this paper extract hair feature and use mean shift to cluster pixels in order to get some regions. Finally, we use Gaussian mixture model to determine the region whether it’s hair region or not. This article demonstrates that our method can precisely detect the hair in different background including varying illumination.