摘要
现有的性别识别算法往往是基于特定人脸数据库进行识别的,由于人脸的复杂度和多样性,在实际应用场景中存在较大缺陷.为此,提出一种基于Gabor小波加权组合特征的性别识别算法.首先对人脸区域进行预处理,调整光照、姿势等进行去噪,通过Gabor小波变换得到人脸的特征向量;然后利用梯度值构造一个权值矩阵对人脸特征进行组合,在显著降维的同时获取有效的组合特征;再对该特征向量进行主成分分析,进一步降维得到加权组合特征;最后将大量训练样本的加权组合特征用支持向量机进行有监督式学习,得到一个二分类的性别分类器.实验结果表明,针对现实场景中的人脸图片,该算法比现有算法具有更高的识别准确率.
Most existing gender recognition methods were tested on specific face databases. However, they are faced with big challenges in practical applications due to the complexity and diversity of human faces. This paper presents a novel gender recognition method based on weighted combination of Gabor wavelet transform. Firstly, the face image is preprocessed to remove the influence of light and posture. We then utilize Gabor wavelet transform to generate feature vectors of the face image and construct a weight matrix to combine the Gabor features into a robust feature based on gradient direction. It reduces the redundancy greatly and, at the same time, highlights the most valuable components of combined features. Secondly, we use principal component analysis to further reduce the dimension of the obtained feature. Finally, we put a large number of extracted features from training samples into support vector machine to train a binary gender classifier. The experimental results show that the proposed method achieves higher accuracy on both the public face database and the face database collected from practical applications when compared to state-of-the-art methods.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2015年第9期1767-1774,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61100080
61402120)
广东省自然科学基金(2014A030313154)
关键词
性别识别
GABOR小波变换
加权组合
gender recognition
Gabor wavelet transform
weighted combination
