感兴趣视觉特征耦合种子繁衍的图像检索算法  

Image retrieval algorithm based on interest visual features and seed propagation

作　　者：韩建敏[1] 李国伟[2] 王振飞[3] HAN Jian-min;LI Guo-wei;WANG Zhen-fei(School of Computer Engineering, Henan Institute of Economics and Trade, Zhengzhou 450000, China;School of Computer, Zhongyuan University of Technology, Zhengzhou 450000, China;School of Information Engineering, Zhengzhou University, Zhengzhou 450001, China)

机构地区：[1]河南经贸职业学院计算机工程学院,河南郑州450000 [2]中原工学院计算机学院,河南郑州450000 [3]郑州大学信息工程学院,河南郑州450001

出　　处：《计算机工程与设计》2018年第12期3785-3790,3796,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61172113);河南省科技攻关基金项目(122102210508)

摘　　要：为避免基于内容的图像检索过程中底层特征和高层语义概念的语义鸿沟问题,通过引入主动学习的相关反馈,提出一种感兴趣视觉特征耦合种子繁衍机制的图像检索算法。根据种子核映射,定义自适应约束的种子繁衍机制(adaptive constraints-based seed multiplication,ACSM),利用ACSM提取用户感兴趣的区域(region of interest,ROI)特征;对数据库图像分割,获取分割区域局部特征;利用Euclidean距离测量查询图像与数据库的相似性,根据相似性获得初始检索结果;基于ACSM算子,从ROI与相关反馈中学习图像的低层特征和高层语义之间的关联,对用户的相关反馈进行排序,提高图像检索精度。实验结果表明,与当前图像检索技术相比,所提算法具有更高的检索精度和效率,表现出更好的Precision-Recall曲线。To avoid the semantic gap between low-level features and high-level semantic concepts in content-based image retrie-val,an image retrieval algorithm based on interest visual features seed propagation mechanism was proposed by introducing the relevance feedback of active learning.According to the seed kernel mapping,a constraint based on adaptive seed dispersal(adaptive constraints-based seed multiplication,ACSM)was defined by using ACSM operator to extract the user interested area(region of interest,ROI)features.Segmentation of database image was used to obtain local feature of segmentation region.The Euclidean distance was used to measure the similarity between the query image and the database,and the similarity of the mea-surement was used to obtain the initial retrieval results.Based on ACSM,the association between low-level features and high-level semantics was effectively learnt from ROI and related feedback to improve the accuracy of image retrieval.Experimental results show that the proposed algorithm can significantly improve the effectiveness and efficiency of image retrieval,which has better precision-recall curve.

关 键 词：图像检索 自适应种子繁衍 相关反馈 感兴趣的区域 主动学习 局部特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]