摘要
针对局部保持投影(LPP)在构造邻接图时,基于欧氏距离的近邻选取方式往往不能很好地反映数据间的几何结构关系问题,提出一种融合相关系数的LPP人耳识别算法。该算法通过融合图像相关系数和欧氏距离来构建邻接图,能更好地揭示出数据间的几何结构关系。同时,在设定权值时,融入了图像间的相关系数,能更好地体现高维数据间的相似关系,提取出更有效的鉴别特征。在USTB3和西班牙人耳库上的实验结果表明,本文算法比传统LPP算法识别率提高了10%以上,验证了本文算法的有效性。
Locality Preserving Projections (LPP) algorithm constructs the neighbor graph using Euclidean distance, which may not reflect the actual distribution relationship of the image data in high dimensional space. In order to solve the problem, a new algorithm of locality preserving projections fused with correlation coefficient is proposed. The algorithm uses Euclidean distance fused with correlation coefficient of ear images to construct the neighbor graph, which can reflect the geometry relationship of the image data better. Meanwhile, we also take correlation coefficient into account when we calculate the weights of the neighbor graph, thus reflect the similarity relationship of data, extract the distinguishing features, and realize the data dimension reduction. The results of the comparative experiment of the two methods on USTB3 and Spain ear database indicate that the highest accuracy of the proposed method is improved over10%, which verifies the efficiency of the proposed algorithm.
出处
《光电工程》
CAS
CSCD
北大核心
2015年第6期1-7,共7页
Opto-Electronic Engineering
基金
中央高校基本科研业务费项目(1061120131207)
关键词
人耳识别
LPP
相关系数
邻接图
ear recognition
Locality Preserving Projections (LPP)
correlation coefficient
neighbor graph