Query-by-Shape interface for Content Based Image Retrieval

摘要

In this paper we describe Query-by-Shape, a new Content-Based Image Retrieval interface. In this method objects are decomposed into features, each feature may consist of a color, a texture or a shape attributes. The query stage consists of comparing the query objects graph with objects from the database. The main advantage of the proposed solution is the ability to query for an object without having full knowledge about it. The proposed interface allows creating graphically drawn queries using simple, predefined shapes. Moreover, the method is able to automatically detect image features and to create a graph used for the query. Because users sometimes would like to get results representing not only exactly the same class of objects, but also similar objects, thus as a result a set of objects, ordered according to the similarity to the query, are returned. The method was implemented as a prototype web-based application. The initial, experimental results proved that the method is very promising, it identifies objects with a high precision.