漂移系数优化颜色直方图的Mean Shift算法  

Mean Shift algorithm with shift coefficient refined color histogram

作　　者：宋康康[1] 陈恳[1] 郭运艳[1] 

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211

出　　处：《计算机工程与应用》2014年第13期173-177,181,共6页Computer Engineering and Applications

基　　金：宁波市科技局自然科学基金(No.2010A610109);宁波大学学校科研基金(理)(No.xkl11075)

摘　　要：基于颜色直方图的传统Mean Shift视频目标跟踪算法,容易受到背景信息的干扰而造成定位偏差。针对此问题,提出了一种基于漂移系数的加权颜色直方图算法。根据核窗口内每个像素提供系数的大小,抑制对漂移贡献小的颜色特征,从而增强目标特征的刻画,并通过抑制背景信息和引入新的目标信息来更新模板。实验结果表明,该算法定位跟踪效果优于背景直方图加权算法,并且迭代次数更少。采用模板更新后,跟踪性能更加稳定。The conventional Mean Shift video target tracking algorithm based on color histogram is very sensitive to the interference of background noise, lead to the target location error. To address this problem, a weighted background algo-rithm based on shift coefficient is proposed. According to the shift coefficient given by each pixel in the kernel window, the proposed approach can reduce color feature amount making small contribution to mean shift in the color space, enhance the characteristic of the target, update the template promptly by restraining background interference and introduce new target cues. Experimental results show that the proposed method performs better than background weighted histogram algorithm with less iterations, and even more stable with the template update mechanism.

关 键 词：漂移系数 加权直方图 Mean SHIFT算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]