摘要
对形状轮廓相似目标进行识别时,应用全局特征很难得到有效的鉴别结果,针对这一问题,提出了一种基于Con-tourlet、核主成分分析+Fisher线性辨别(KPCA+FLD)的特征提取方法。选取Contourlet分解后提取出来的多尺度局部特征,以加权求和的方式进行融合处理,选用KFD(KPCA+FLD)对融合后的特征进行降维,选择鉴别力强的特征。最后通过一系列的仿真实验,包括选用不同的特征提取方法、分解层次、核函数、融合权重,验证了该特征提取方法的有效性。
It is difficult to effectively identify the targets of similar shape and contour,when applying the global characters.A feature extraction method is proposed based on contourlet transform and KFD（kernel principal component analysis and fisher linear discriminant）. All sample images are decomposed using contourlet transform firstly,the extracted multi-scale local characters are integrated with different weights.Then the dimensionality of the fused feature vector is reduced using KFD algorithm,realizing the choice of characters which have strong identifying ability.Finally,a series of simulation experiments are finished,including using different method for feature extraction decomposition levels,nuclear function,weight of integration.Experimental results show the validity of this feature extraction method.
出处
《计算机工程与设计》
CSCD
北大核心
2011年第1期240-243,共4页
Computer Engineering and Design
基金
江苏省高校自然科学研究基金项目(09KJB510002)
南京工业大学青年学术基金项目(39710006)
