Using graph representation in content-based image retrieval

Summary form only given. We propose a content-based image retrieval application using an inexact graph matching algorithm. The matching algorithm decomposes the matching process into K phases, each exploiting a different part of solution space. With most plausible parts being searched first, only a small number of phases is required in order to produce very good matching (most of them optimal). A content-based image retrieval application using the new matching algorithm and a fuzzy clustering is also described.