基于矩形代数和公共模式方法的相似图像检索

An improved similarity retrieval of images based on CPM and rectangle algebra

指出了图像检索中公共模式方法(common pattern method,CPM)所建立的type-i公共子图无法精确描述区域间的空间拓扑关系.研究采用矩形代数表示CPM中区域间的空间拓扑关系,得到了拓扑表达更精确的相似性图像检索算法(SRRA).该算法将对象抽象为最小边界矩形,采用矩形代数描述对象间的二维空间关系,构建基于矩形代数的相似图,并从中寻找最大相似对象集合.实验结果表明,SRRA不仅在效率上优于基于CPM的算法,且检索效果更符合用户要求.

The common pattern method （CPM） is one of the excellent algorithms among state oi the art similarity image retrieval methods. However, the type-/rule using in CPM is unable to exactly distinguish the topological re- lationships between areas. By applying rectangle algebra to CPM, a novel similarity retrieval by rectangle algebra （SRRA） was proposed. SRRA abstracts an object into a minimum bounding rectangle, uses rectangle algebra to express the 2D space relationship between objects, constructs similarity graphs based on rectangle algebra, and obtains a maximum similar objects set. The experimental results show that SRRA performs better than CPM with respect to the time consumed and the precision of retrieval results.