一种基于中值流的Mean-shift改进方法  

Animproved algorithm of Mean-shift based on the median flow

作　　者：李增前 张奇志[1] 周亚丽[1] 

机构地区：[1]北京信息科技大学自动化学院,北京100192

出　　处：《北京信息科技大学学报（自然科学版）》2014年第3期38-42,48,共6页Journal of Beijing Information Science and Technology University

基　　金：国家自然科学基金资助项目(11172047;11072038);北京市属高等学校人才强教深化计划资助项目(PHR201106131)

摘　　要：通过引进中值流的方法,以中值流算法作为运动模型,利用中值流的预测信息指导Mean-shift算法对目标进行搜索。在Mean-shift对目标进行搜索的时候,通过引进目标的纹理特征——局部二值模式(local binary patterns,LBP)使得在目标和背景颜色相似、光照变化强烈的情况下也能对目标和背景进行区分。仿真实验表明,该算法比Mean-shift算法在准确性和跟踪性能上均有明显的提高。The median flow algorithm is introduced as a movement model. The predict information of the median flow is used to guide the Mean-shift algorithm to search the target which will solve the scale zooming problem. When the Mean-shift search the target,the texture feature of LBP（ local binary patterns） is introduced to tell the difference between target and background even when they have similar color and illumination changes greatly. The experimental results show that the proposed algorithm is obviously improved than the Mean-shift algorithm both in the accuracy and tracking performance.

关 键 词：中值流 纹理特征 MEAN-SHIFT 尺度放缩 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]