基于差分投影和梯度向量流场的眼睛特征提取

提出了差分投影的方法确定眼睛区域，并根据人脸图像的边缘梯度图提出了一种新的基于梯度向量流场的眼睛特征提取方法。该算法改进原有的梯度向量流迭代方程求解梯度向量流场，以梯度向量流场中的汇点作为候选点，通过人脸器官的几何位置关系等方法评价候选点以定位眼球。算法能较好地容忍一定的光照变化、人脸的小角度倾斜和旋转、闭眼和眼镜等干扰。在具有以上干扰的ORL人脸库(400幅图像)和JAFFE人脸表情库(213幅图像)上的实验证明该算法具有较好的眼睛特征抽取能力。

Eyes Feature Extraction Based on Difference Projection and Gradient Vector Flow Field

A new difference projection method is proposed to extract eye candidate area, and then, according to features of edge gradient maps of face images, a new eye feature extraction algorithm based on gradient vector flow field is proposed to determine eyeball candidate points. A gradient vector flow field is computed iteratively by improving the iterative formula of gradient vector flow algorithm. Sink points in the gradient vector flow field will be candidate points of eyes, and then some other methods are used to score these candidate eye pairs. The proposed algorithm can tolerate such disturbances as slightly varying illumination, face rotation, closed eyes and eyes with glasses. It is proved that the new algorithm has the good ability to extract eye feature by experiments on the ORL face database with 400 face images and the JAFFE facial expression database with 213 face images.