1 Introduction
In the past few years, many content-based image retrieval (CBIR) systems such as QBIC[3], Photobook[4], VisualSEEK[5], VideoQ[6], etc. have been developed to manage the overwhelming accumulation of rich digital images. CBIR systems are designed to search image libraries for specific image features (e.g color, textures, shapes). Querying is performed by comparing a search image with the feature vectors of all images in the database. Most CBIR applications use vectors to store and retrieve images, since comparison of vectors is easy to implement. However, this kind of structure does not make use of the spatial information within the image. Graphs are more efficient for describing objects and relationships between objects. When graphs are used, the comparison amounts to graph matching.