Similarity Searching in Large Image DataBases

Loading...
Thumbnail Image
Files
TR_94-88.pdf(1.53 MB)
No. of downloads: 380
Publication or External Link
Date
1994
Authors
Petrakis, E.G.M.
Faloutsos, Christos
Advisor
Citation
DRUM DOI
Abstract
We propose a method to handle approximate searching by image content in large image databases. Image content is represented by attributed relational graphs holding features of objects and relationships between objects. The method relies on the assumption that a fixed number of ﲬabeled or ﲥxpected objects (e.g. ﲨeart lungs etc.) are common in all images of a given application domain in addition to a variable number of ﲵnexpected or ﲵnlabeled objects (e.g. ﲴumor , hematoma etc.). The method can answer queries by example such as ﲦind all X-rays that are similar to Smith's X-ray . The stored images are mapped to points in a multidimentional space ad are indexed using state- of-the-art database methods (R-trees). The proposed method has several desirable desirable properties: (a) Database search is approximate so that all images up to a pre-specified degree of similarity (tolerance) are retrieved, (b) it has no ﲦalse dismissals (i.e., all images qualifying query selection criteria are retrieved) and (c) it scales-up well as the database grows. We implemented the method and ran experiments on a database of synthetic (but realistic) medical images. The experiments showed that our method significantly outperforms sequential scanning by up to an order of magnitude.
Notes
Rights