中文摘要：

为了提高动态图像序列中人脸轮廓跟踪时梯度矢量流（GVF）Snake算法的实时性,同时解决人脸跟踪中的遮挡问题,提出了GVF Snake和单变量一阶灰色模型GM（1,1）相结合的人脸轮廓提取方法。该方法首先利用人脸运动信息和肤色模型粗略地检测出运动人脸轮廓,然后采用GVF Snake算法将人脸轮廓精确地提取出来,从而有效解决GVF Snake算法的初始化问题。根据人脸轮廓运动的整体性,利用GM（1,1）模型预测人脸轮廓质心的位置,并以预测位置作为GVF Snake的迭代依据,同时将GVF Snake提取的人脸轮廓质心位置作为下一帧图像GM（1,1）模型的预测依据。存在遮挡时,则以GM（1,1）模型预测保持跟踪的连续性。实验结果表明,该算法跟踪的平均时间仅为GVF Snake算法的8.0%,平均跟踪误差仅为GVF Snake算法的31.2%,而且能更好地反映人脸轮廓的运动规律,跟踪实时性强,鲁棒性好。

英文摘要：

In order to improve the real-time performance of the Gradient Vector Flow Snake（GVF Snake） algorithm for face contour tracking in a dynamic image sequence and to overcome the occlusion problem in face tracking,a novel image extraction method combining the GVF Snake algorithm and the single variable first-order grey model GM（1,1） is proposed to extract the face contour.In this method,the moving face contour is roughly detected out firstly by using human motion information and the skin-color model,and then the accurate face contour is extracted by using the GVF Snake algorithm,by which the initialization problem of the GVF Snake algorithm is solued.For the integrity feature of face contour motion,the GM（1,1） model is used to predict the centroid position of face contour and then the position is used as the iteration basis of the GVF Snake algorithm.Meanwhile,the centroid position of face contour extracted with GVF Snake is taken as the prediction basis of the GM（1,1） model for the next frame.When the occlusion exists,the continuity of tracking can be held with the prediction of GM（1,1） model.Experimental results show that by proposed method,the average tracking time and the average tracking error are only 8.0% and 31.2% of those of the GVF Snake algorithm respectively.It can be concluded that this method can better reflect the motion law of face contour,and has strong real-time performance and good robustness.