基于多特征Mean Shift的人脸跟踪算法.pdfVIP

第31 卷第8 期 电 子 与 信 息 学 报 Vol.31No.8 2009 年8 月 Journal of Electronics Information Technology Aug..2009 基于多特征Mean Shift 的人脸跟踪算法 张 涛 蔡灿辉 (华侨大学信息科学与工程学院 泉州 362021) 摘 要：该文把局部三值模式(Local Ternary Patterns, LTP)纹理特征引入Mean Shift 跟踪算法，提出了基于多 特征的Mean Shift 人脸跟踪算法以解决Mean shift 跟踪算法的鲁棒性问题。通过对LTP 纹理特征的分析、研究， 提出了一个 LTP 关键纹理模型，既增强了目标的关键纹理信息，又简化了 LTP 纹理模型。在此基础上，提出一 种基于LTP 关键纹理特征和肤色特征的Mean Shift 人脸跟踪算法，有效地解决了Mean Shift 算法的鲁棒性问题。 为进一步提高对快速运动目标的跟踪速度和跟踪性能，该文引入了卡尔曼滤波器对目标进行预测。实验结果表明， 该文的算法在目标定位的准确性和跟踪性能上比Mean Shift 算法均有明显的提高。 关键词：人脸跟踪；卡尔曼滤波；Mean Shift 算法；局部三值模式 中图分类号：TP391.41 文献标识码：A 文章编号：1009-5896(2009)08-1816-05 A Face Tracking Algorithm Based on Multiple Feature Mean Shift Zhang Tao Cai Can-hui (Institute of Information Science and Technology, Huaqiao University, Quanzhou 362021, China) Abstract: In this paper, an improved Mean Shift face tracking algorithm based on Local Ternary Patterns (LTP) of texture and color features is proposed to improve the robustness of the Mean Shift algorithm. Based on the study of LTP texture features, an LTP key texture pattern is introduced to enhance the important features of an object and reduce the computational complexity of the LPT texture model. A multiple feature Mean Shift face tracking algorithm is then proposed based on the LTP key texture and complexion features, and the robustness of Mean Shift algorithm is significantly enhanced. Furthermore, in order to improve the tracking speed and robustness, the Kalman filter is introduced to predict the position of the object window. Experimental results show that compared with the original Mean Shift algorithm, the proposed multiple feature face trac