关键词: 人脸跟踪, Mean Shift算法, 局部二值模式, 尺度自适应

Abstract: 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 adaptive 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.

Key words: face tracking, Mean Shift algorithm, Local Binary Pattern（LBP）, scale adaptive