摘要
随着人耳识别技术的发展,如何寻找不受光照和对比度影响的特征并用于识别成为研究热点之一。本文提出一种利用相位一致性对人耳进行特征提取的方法,该方法首先从8个方向计算人耳图像中每点的相位一致性,然后对经过相位一致性变换的输出图像分块求和,形成特征矢量,最后通过KDDA将特征向量投影到低维高可分空间,并用欧式距离对人耳图像分类。对77个人,共308幅人耳图像进行实验,在等误率为1.7%的情况下,识别率可以达到98.2%左右,验证了该方法所提取的人耳特征不受光照影响。
As ear recognition is to be developed, it is crucial to find an illumination and contrast invariant measure of feature significance. This paper proposed a method of ear Feature Extraction based on Phase Consistency. This approach first calculates the Phase Congruency of every point in the ear image from 8 directions, and then we divide output images of Phase Congruency into equal blocks and sum value of points in each block of the image to obtain feature vector, at last project vector into low-dimensional space with enhanced discriminant power via KDDA,and in the space we calculate the Euclidean distance for matching. Experiments are car- ried on the database of 308 images of 77 ears. When the EER is 1.7%, the CCR could reach about 98.2. And The invariance to illumination is illustrated in ear recognition.
出处
《微计算机信息》
2011年第12期4-6,共3页
Control & Automation
基金
教育部"春晖计划"科研合作项目(Z2005-2-11009)
辽宁省创新团队项目(2006T102)
辽宁省科研项目(L2010436)
