Content-based image retrieval using fuzzy multiple attribute relational graph

In this paper, the authors present a new CBIR approach which can handle queries involving multiple attributes; not only object label, color and texture but also natural spatial relation. They use fuzzy sets to model the imprecision and vagueness of objects in an image. They use a fuzzy attributed relational graph (FARG) to represent each image in the database. Each object in an image is represented by a node with multiple attributes. The relation between objects is represented by an edge. One can also convert a user query into a FARG. This approach makes the image retrieval problem to a sub-graph matching problem. In the experiment using the synthetic database of 1240 images, the proposed approach shows a good performance, compared with the single attribute approach.