摘要
针对基于传统Gabor小波变换的人脸特征提取存在维数高的不足,提出了一种基于改进的Gabor特征融合和SVM的人脸识别算法,并用二维傅里叶变换进行加速求解,提高了特征提取的速率。提取了人脸图像的Gabor多方向和多尺度特征,然后对同一方向上不同尺度的特征进行融合,再采用fast PCA算法对融合后的特征进行降维,最后用改进的SVM分类器即混合核函数分类器进行分类识别,并利用两种处理模式对分类器进行融合。在FERET和ORL人脸库上进行了实验,结果表明该算法能有效地表征人脸,具有较高的识别率。
Considering the high dimensional deficiencies of face feature extraction based on traditional Gabor wavelet transform, a face recognition algorithm based on improved Gabor feature fusion and SVM is proposed,and it accelerates the solving process with the two- dimensional Fourier transform and improves the rate of feature extraction. The Gabor multi-directional and multi-scale features of face image are extracted, then fusing of the features in the same direction at different scales, after that, the dimension of fused feature by fastP- CA algorithm is reduced. Finally the face images are recognized with the improved SVM classifier based on mixed kernel function, and the classifier is fused by using two kinds of processing pattern. The experimental result is conducted on face database like FERET and ORL, which shows that the algorithm can effectively characterize face and improve recognition rate.
出处
《计算机技术与发展》
2017年第1期65-69,共5页
Computer Technology and Development
基金
国家自然科学基金资助项目(61373055)
江苏省产学研联合创新资金项目(BY2013015-35)
