用于人脸识别的人脸图像标准化算法

人脸图像的标准化，作为人脸图像预处理过程，是计算机人脸识别中很重要的一个环节。论文给出了一种改进的人脸图像标准化算法。该算法首先利用人脸图像的垂直灰度投影曲线确定脸的左右边界；接着根据人脸图像的水平灰度投影曲线来确定眼、嘴的位置；然后依据确定的眼、嘴以及脸的左右边界这些参数对图像进行剪裁和缩放；最后获得人脸的标准图像。采用ORL人脸数据库进行仿真实验，结果表明，该算法基本消除了背景、头发、服装等干扰，保留了人脸识别所需的主要信息，更利于后续的特征提取与识别，解决了常规方法不适合侧面人脸图像或者闭上双眼的人脸图像的问题，是一种有效的方法。

A Method of Face Image Standardization for Face Recognition

Face image standardization, as preprocessing of face image, is a very important loop in computer face recognition. In this paper, an improved method of face image standardization for face recognition is presented, in which right and left borderline is determined according to vertical gray-level projection of face image, and the positions of eyes and mouth are determined according to horizontal gray-level projection of face image, and cropping and zooming of face image are realized according to parameters chosen. Finally a standardizing face image is obtained. Experimental results based on ORL face database demonstrate that interferences such as backgrounds, hair and costumes are almost eliminated, and the main information required in face recognition is preserved. The method is much beneficial for the follow-up feature extraction and recognition, and valid in side-face and two-eye-closed face image for which traditional methods are not proper.