【作者】 邢文浩； 阮秋琦；

【Author】 Xing Wenhao,Ruan Qiuqi (School of Computer and Information Technology,Beijing Jiao tong University,Beijing 100044)

【机构】 北京交通大学计算机与信息技术学院；

【摘要】 在传统的mean-shift人脸跟踪算法中,采用颜色直方图作为跟踪特征,但这并不能充分描述图像的特征,尤其在背景颜色与人脸颜色相似时,导致跟踪受到干扰甚至跟踪错误。本文提出了一种基于边缘方向直方图的mean-shift人脸跟踪算法,用边缘信息和纹理信息作为跟踪特征。实验结果表明,该方法在背景颜色与肤色相似的情况下,跟踪效果明显优于传统的mean-shift跟踪算法。更多还原

【Abstract】 In the traditional Mean-Shift face tracking algorithm,color histogram is used as character,but it is not sufficient for describing the face’s information,especially when the color of the background is similar to the face color,this causes error in face tracing. Here presents a new Mean-Shift algorithm based on edge orientation histograms,using edge information and texture information as characters.Experiments proved that the new algorithm is more accuracy than the traditional ones when the color of the background is similar to the color of the face.更多还原