基于K-L变换的人脸自动识别方法

研究了利用计算机实现的人脸自动识别方法。在传统的“特征脸”方法基础上，提出了一种改进的人脸自动识别方法。该方法对于经过预处理的标准人脸图像，以类间散布矩阵为产生矩阵，通过Ｋ－Ｌ变换降维并提取人脸图像的代数特征。同时，利用遗传算法进行特征选择，以构成有利于分类的自适应子空间。在此子空间内，将图像进行正交分解，然后分别对各类训练样本进行二次Ｋ－Ｌ变换，进一步构成其旋转子空间，从而最终实现了一个分层次的最小距离分类器。实验表明，本方法识别率较高，且对于人脸的姿态、表情及光照条件均具有一定的不敏感性。

Research of automated face recogni tion based on K L transform

Automated Face Recognition (AFR) is one of the most challenging tasks for computer vision and pattern recognition. Based on the traditional eigenfaces method, this paper presents an improving approach to AFR. For those standard face images, regarding the between class scatter matrix as generating matrix, it extracts the algebraic features of face images through K L transform. Then, using Genetic Algorithm (GA) for feature selection, it constructs an adaptive subspace where the training samples of each subject are carried out orthonormal expansion and constitute their rotation subspaces using the second K L transform . Finally, a layered minimum distance classifier is realized . Experimental results show the effectiveness of the approach and its insensitivity to the face posture, expression and illumination conditions.