摘要
针对Gabor特征维数高难题,提高光照人脸的识别性能,提出一种基于Gabor特征融合和最小二支持向量机的人脸识别算法(Gabor-LSSVM)。首先采用Gabor滤波器提取人脸图像的多尺度和多方向特征,并将相同尺度不同方向的特征融合,初步降低特征维数;然后采用核主成分分析对融合特征进行选择,进一步降低特征维数;最后采用最小支持向量机建立分类器对人脸进行识别,并采用Yale B和PIE人脸库进行仿真测试。结果表明Gabor-LSSVM的人脸识别正确率和识别效率都得到了提高。
In order to solve the problem of Gabor in its high feature dimensionality and to improve the recognition performance in illumination condition, we propose a new face recognition method which is based on Gabor features fusion and least square support vector machine （Gabor-LSSVM）. First, it uses Gabor filter to extract muhi-scale and multi orientation features of face images, and fuses the features in same scale but different directions to reduce the dimensionality of features; then it uses kernel principal component analysis to select fusion features to further reduce feature dimensionality; finally it uses support vector machine to establish the classifier to recognise faces. The simulation tests are carried out on Yale-B and PIE fac＇e database, results show that the Gabor-LSSVM improves both the accuracy and efficiency of face recognition.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第7期191-194,225,共5页
Computer Applications and Software
关键词
复杂光照
人脸识别
GABOR特征
最二小乘支持向量
特征融合
Complex illumination
Face recognition
Gabor features
Least square support vector machine
Features fusion
