Image Retrieval Using Structured Logical Shape Feature

Most visual information retrieval systems on the web treat image as an object that is annotated with some keywords. However many semantic contents that cannot be expressed with some keywords are included in one image. It is the reason why most commercial and research systems utilize shape as the most cognitive information to represent contents of an image and to differentiate an image from others. In this paper, we propose an extraction method of logical shape feature to reflect structure of shape and coarse-fine matching method using it. For similar retrieval, we generate pattern segment matrix that is composed of curves’s type in order to find the most similar curve sequence. We use it for coarse-fine matching because our shape features have global characteristic as a structural feature and local characteristic as an adaptive feature of shape. A pattern segment matrix has the advantage to search with only a small quantity of structural features. Our experiments show that structural-adaptive features through logical analysis result in effectively classifying shapes according to their cognitive characteristics. Various experiments show that our approach reduces computational complexity and retrieval cost.