考虑重叠特征分类的图像检索模型仿真  

Simulation of Image Retrieval Model by Considering Overlapping Feature Classification

作　　者：张锦华[1,2] 乔蕊[2] 汤跃明[1] 

机构地区：[1]河南师范大学物理与信息工程学院,河南新乡453007 [2]周口师范学院计算机科学与技术学院,河南周口466001

出　　处：《计算机仿真》2015年第8期444-447,共4页Computer Simulation

基　　金：河南省软科学研究计划项目(132400410934;132400410927)

摘　　要：针对图像进行检索的过程中,由于图像特征受到拍摄分辨率、光线、亮度、拍摄角度的影响,造成图像特征多样化。传统方法忽略了图像的特征多样化对图像检索的干扰,导致无法有效实现图像特征的分类,提出考虑重叠特征分类的图像检索模型方法,在Contourlet变换域中引入Hu不变矩和均值标准差对特征进行提取并分解,分析每个方向子带系数的分布状态。利用非线性映射函数将原始数据映射至高维特征空间中完成分类,对图像进行初步检索,将所获取结果中前N幅图像看作是训练集,对正确样本进行标记,将训练集输入到支持向量机完成训练,获取图像分类器。通过上述分类器对所有图像之间的相对距离进行计算,根据其距离从大到小的顺序进行排列,输出最终结果。仿真结果表明,所提方法具有很高的检索正确率。An image retrieval model method for considering overlapping characteristics classification was pro- posed in the paper. Hu invariant moments and average standard deviation were introduced in the Contourlet transform domain to make feature extraction and decomposition, and analyze the distribution state of sub - band coefficients on each direction. Nonlinear mapping function was used to make the raw data mapping to high - dimensional feature space and complete the classification. The preliminary retrieval was made with the images, and the first N images of the obtained results were taken as the training set. The correct tag samples were signed, and the training set was in- put into the support vector machine （SVM） to complete the training, so as to obtain the image classifier. Through the classifier, the relative distance between all images wasw calculated, and according to their distance, it makes orde- ring from large to small, and to output the final result. The simulation results show that the proposed method has high accuracy of retrieval.

关 键 词：重叠特征 分类 图像检索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]