光照和尺度自适应的Mean Shift人脸跟踪算法

Face tracking based on illumination and scale adaptive Mean Shift

针对原始Mean Shift算法易受光照强度影响及跟踪窗口不随目标尺度自适应变化的问题,提出了一种光照和尺度自适应的Mean Shift人脸跟踪算法。该算法将颜色特征与光照不变性特征局部二值模式结合起来共同表征人脸,增强了复杂背景下目标的跟踪性能,利用矩特征和巴氏系数估计目标的真实尺度,提高了人脸发生较大形变时的适应能力。实验结果表明,提出的算法比传统的基于颜色直方图的Mean Shift算法具有更准确的跟踪结果。

Because the traditional Mean Shift algorithm is easily influenced by illumination change and the problem that tracing window does not adaptively change as target scale. A face tracking algorithm based on illumination and scale adap-tive Mean Shift is proposed in this paper. This algorithm combines color feature and Local Binary Pattern（LBP）feature which is invariant to illumination when representing face, for enhancing the performance of tracking face under complicated background. Then the torque characteristic and Bhattacharyya coefficient are used to estimate the real scale of target, which strengthens the adaption of the Mean Shift algorithm when large deformation occurs. Compared with the traditional Mean Shift algorithm, the approach shows superior performance in terms of both accuracy and robustness.