基于谷算子的人眼特征点检测方法

考虑带姿势旋转的人脸的特征点检测问题，提出了一种基于谷算子得到眼睛的候选区域，用Gabor滤波和方差投影构造模糊模板检测出眼睛的位置，并利用二值图像数学形态学与投影的方法得到有姿势旋转的人脸特征点检测的算法。该算法利用软数学形态学构造了一种新的谷算子，用一种新的局域区域取二值化的方法对图像进行二值分割，提出了通过谷算子运算得到人脸谷极点的方法，减少了一些不稳定点的影响。实验表明，该算法具有很高的精度和很强的鲁棒性。

Detecting Human Eyes' Character Points Based on Novel Valley Operator

This paper consider detecting eyes character points with different poses. In this algorithm, we firstly construct a new valley operator to obtain the candidates of eyes, and a novel algorithm to get binary image is also defined. Then the eye position can be detected by matching the fuzzy template constructed by combine the gabor filter and variance filter, and we can obtain the 12 character points by binary morphology and the method of projection finally. We put forward a novel valley operator based on soft mathematical morphology, it can reduce the influence of un-stabilization points. Experimental results demonstrate the high efficiency of the algorithm in runtime and correct localization rate.