利用图像特性的模糊聚类图像检索方法  

Image retrieval method on image features of fuzzy clustering

作　　者：李兰[1] 刘洋[1] 马振[1] 邵明文[1] 

机构地区：[1]青岛理工大学计算机工程学院,青岛266071

出　　处：《清华大学学报（自然科学版）》2014年第7期929-934,共6页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(61173181)

摘　　要：在图像的检索方法中,大多数均根据图像的变换域的特征进行检索,其缺点是没有抓住图像的现实属性,从而检索效率低下,检索精度较低。针对此问题,该文根据内容(形状、颜色、纹理等)的视觉特性的不同,结合局部和全局特征,提出一种基于聚类形状的图像检索方法。首先将对象形状包含图像通过Fourier变换的方法进行描述,其次应用双向经验模式分解检测图像边缘,最后应用模糊聚类检索方式进行图像语义类别检索。其中所采用的模糊聚类算法采用机构监督机制,从而使形状识别类别用一组标记形状代表。根据导出的形状原型检索类似形状。相比于现有的检索方法,对比结果显示该方法在检索精度方面有了显著的改善。Most image retrieval methods are conducted based on the features of the image transform domain of retrieval,with the image of real property not seized with low retrieval efficiencies and low retrieval precisions.Aiming at this problem,this paper presents an image retrieval method based on the clustering shape according to different visual features of the content（shape,color,and texture）,combining the local and global features.Object shapes including images were described using the Fourier transform method,with two-way empirical mode decomposition then used to detect image edges and fuzzy clustering retrieval used to complete image semantic category retrieval.Institution supervision mechanism was applied to the fuzzy clustering algorithm so that the shape identification category was represented by a set of tags on behalf of the shape,with the shape of the export prototype retrieving the similar shape.Compared with the conventional retrieval methods,the results show that the developed method significantly improves the retrieval accuracy.

关 键 词：模糊聚类 形状特性 图像检索 边缘检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]