针对训练样本较少的情况,提出了一种新的人脸识别方法。采用Gabor小波变换得到不同的子图信息,从子图中提取特征;对每个滤波器滤波产生的子图分别进行非负矩阵分解以实现数据降维及特征选择;设计两层分类器完成图像的分类识别,采用基于距离的最近邻分类器对图像进行第一层分类识别,通过对第一层分类结果进行统计记票,获得最终的识别结果。在Yale人脸库中进行实验,实验结果表明,给出的方法有效地提高了人脸识别率。
For the case of few training samples, a new face recognition method is proposed. Features are extracted from different sub-images obtained through Gabor wavelet transform. Non-negative matrix factorization is used to accomplish data dimensionality reduction and feature selection for sub-images generated by each filter. Two-layer classification is designed to finish the classification. Nearest neighbor classifier based on distance is adopted to complete the first layer classification.The final classification results are gained by counting the record of the results from the first layer. Experiments are done on Yale face database. The results show that the method presented in this paper can effectively improve the recognition rate.