基于有效参数初始值估计的改进AAM拟合算法

An Improved AAM Fitting Algorithm with Effective Initial Parameter Estimate

为了高效地解决图像特征点配准特别是人脸特征点配准问题,基于活动表观模型(AAM)的拟合算法得到了广泛研究和应用。然而,当参数初始状态和真实情况相差较大时,AAM拟合算法经常无法得到满意的结果。针对AAM拟合算法对参数初值敏感的问题,在AAM反向组合算法基础上,提出了一种简单有效地进行参数初始值估计的方法,通过提取相关区域的矩特征和线性映射计算出比较精确的参数初始值。实验证明方法通过可靠的初始参数值减少了迭代次数,并提高了拟合准确度。

AAM- based approaches have achieved great success to solve image alignment problem and have a wide variety of applications. However, AAM is sensitive to the initial parameter and often converge at local optimal value if the initial state is far away from the ground truth. This paper proposes a simple and effective approach for initial parameters estimate using invariant moments vector and linear mapping. The proposed approach is applied to the face alignment problem. Experiments show that this approach could significantly improve the accuracy and efficiency of face alignment.