摘要
针对人耳识别中人耳的角度变化这个难点问题,提出一种结合Gabor小波和监督保局投影的人耳识别算法.由于Gabor特征维数高、冗余大,首先通过统计样本的边缘点再采样的方法对人耳进行稀疏的描述,然后利用类别可分离性判据评价Gabor展开系数的分类能力,选择最有利于识别的Gabor展开系数构造新的Gabor特征.在人耳库中的实验结果表明,采用文中算法提取的Gabor特征维数少、鉴别能力强,结合监督保局投影进行识别取得了很高的识别率,对于人耳角度的变化具有良好的鲁棒性.
A new ear recognition algorithm based on Gabor wavelet and supervised locality preserving projection （SLPP） is presented in this paper to solve the difficult problem of ear recognition with ear pose variation. Considering the redundancy in the high dimensional Gabor feature vectors, ear images are first described sparsely by statistical edge points sampling. Then a criterion with discriminating power is employed to evaluate the classification ability of the Gabor coefficients. The Gabor coefficients most favorable to the recognition are selected to construct new Gabor features. Our experiment results on ear database show that the proposed method produces less number of Gabor features and achieves high recognition rate with supervised locality preserving projection. The method is also robust to the ear pose variation .
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2010年第8期1259-1265,共7页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(60903127)
关键词
人耳识别
GABOR小波
流形学习
监督保局投影
human ear recognition
Gabor wavelet
manifold learning
supervised locality preserving projection
