摘要
指纹分类是自动指纹识别系统中的关键技术,但目前的算法对低质量的指纹图像的分类还存在较大的误差。为了能够对低质量的指纹图像进行准确分类,提出了一种基于频谱能量的指纹分类,首先对分块的指纹图像进行傅立叶变换,然后根据频谱图中能量的分布特点得到指纹图像的方向图,提取core点周围的指纹图像的方向向量作为该指纹图像的特征向量。最后使用K近邻分类器和最小距离分类器对输入指纹进行分类。在NIST-4指纹数据库上的实验结果表明了算法的有效性,分类正确率达到94.1%,且算法速度比同类算法有较大的提高。
Fingerprint classification is the key technology in automatic fingerprint identification system （AFIS）, but most of the present methods：give bad results when the fingerprint image is low quality. For the solution of this problem, a new method based on energy in frequency domain is presented. Firstly, the fingerprint images in spatial domain is transformed to frequency domain and the directional images is constructed according to energy in the frequency domain. Then directional vectors around the core point are got as eigenvector of fingerprint images. At last, the input image is classified by K-nearest neighbor classifier and least distance classifier. Experimental results on NIST-4 database show the validity, the classification accuracy reaches 94.1% and in the speed this method has competitive performance compared with other methods.
出处
《计算机工程与设计》
CSCD
北大核心
2008年第8期2010-2013,共4页
Computer Engineering and Design
基金
山东省科技攻关基金项目(2005GG3201089)
山东省优秀中青年科学家科研奖励基金项目(2006BS01008)
关键词
指纹识别
指纹分类
傅立叶变换
最小距离分类器
K近邻分类器
fingerprint identification
fingerprint classification
Fourier transform
least distance classifier
K-nearest neighbor classifier
