Content-based image retrieval in medical applications: a novel multistep approach

摘要: In the past few years, immense improvement was obtained in field of content-based image retrieval (CBIR). Nevertheless, existing systems still fail when applied to medical databases. Simple feature-extraction algorithms that operate on entire for characterization color, texture, or shape cannot be related descriptive semantics knowledge is extracted from images by human experts. this paper, we present a novel multi-step approach, which specially designed applications (IRMA). contrast common approaches, IRMA-concept based conceptual and algorithmic separation of: (a) categorization using global features, (b) geometry registration with respect prototypes within categories, (c) extraction local (d) category query dependent feature selection, (e) index generation resulting hierarchical multi-scale blob representations, (f) object identification links a-priori content blobs, (g) processed abstract blob-level. The comprises several benefits compared CBIR-systems. categories enable prototypes. Furthermore, each might belong categories. A-priori both adjuncted indexing. Therefore, provides high amount understanding enables intelligent queries an level information. Hence, IRMA promises satisfactory completion applications.