一种改进的基于灰度投影的人眼定位算法

利用灰度投影对人脸图像进行检测和眼睛定位是一种常用方法，但是直接采用该算法进行眼睛定位，容易将鼻子或嘴的水平位置误判为眼睛水平位置，从而导致检测准确率降低。本文在计算图像水平方向灰度投影时，加入像素点的位置方差特征，这样可以准确地找出眼睛的水平位置，从而精确的定位眼睛坐标。算法在FERET人脸数据库上测试，准确率达92.4%。

An Improved Eye Location Algorithm Based on Gray-level Projection

It is a common method to use gray-level projection to detect face image and eye location. But if we use this method alone, we may get the nose or mouth horizontal position as the eye horizontal position, which will reduce the accuracy rate of eye location. In this paper, when we calculate the eye horizontal position, we add a new variance feature. And we find in this method we can get the eye horizontal position more correctly. The improved algorithm was test on the FERET face image database, and the accuracy reach 92.4%.