%0 Journal Article %T 一种判别极端学习的相关反馈图像检索方法<br>A Retrieval Method of Relevance Feedback Images Based on Discriminative Extreme Learning %A 黄晓冬 %A 孙亮 %A 刘胜蓝 %J 西安交通大学学报 %D 2016 %R 10.7652/xjtuxb201608016 %X 针对基于支持向量机(SVM)的相关反馈图像检索方法计算复杂度高、缺乏判别能力以及图像特征提取不充分的问题,提出一种基于判别极端学习的相关反馈图像检索(DELM)方法。在图像特征提取阶段,通过连接图像的颜色、纹理及边缘直方图实现图像的特征提取,解决了以往多数检索方法仅使用单一图像特征造成的图像描述不充分的问题;在检索的反馈阶段,将最大边际准则(MMC)引入到极端学习机中,通过分析极端学习机隐层空间的类内离散度和类间离散度得到包含判别信息的分类模型,并给出降维和不降维两种形式,以提高相关反馈图像检索系统的检索能力。DELM方法能有效应用于基于内容的图像检索中,并显著提高图像检索的性能。实验结果表明,DELM方法和采用SVM、ELM和最小类别方差ELM的方法相比,在Corel？？1K数据集下检索平均准确率分别提高了11.06％、5.28％和6.40％。<br>A novel retrieval method of relevance feedback images based on discriminative extreme learning, named DELM, is proposed to concern the high computational complexity, low discriminant ability and insufficient image feature extraction of content？？based image retrieval (CBIR) with the relevance feedback (RF) methods based on support vector machine (SVM). The proposed method extracts image features in the phase of image feature extraction through color, texture and edge histogram of the image to solve the problem that image feature extraction of the existing methods that based on single feature is insufficient. A maximum margin criterion (MMC) is introduced to extreme learning machine (ELM) in the phase of retrieval feedback. A classification model including discriminative information is obtained through analyzing discrete degrees within and between scatters of feature space in ELM hidden layer, and two versions of DELM are proposed to improve the retrieval performance of RF based image retrieval system, that is, a dimension reduction free based version and a dimension reduction based version. The DELM method is effectively applied to CBIR, and significantly improves the quality of retrieval performance. Experimental results on Corel？？1K dataset and comparisons with the methods using SVM, ELM, and minimum class variance ELM (MCVELM) show that the average retrieval precision of the DELM method increases by about 11.06%, 5.28% and 6.40%, respectively %K 图像检索 %K 相关反馈 %K 支持向量机 %K 极端学习机 %K 判别信息< %K br> %K image retrieval %K relevance feedback %K support vector machine %K extreme learning machine %K discriminative information %U http://zkxb.xjtu.edu.cn/oa/DArticle.aspx?type=view&id=201608016