M-tree: An Efficient Access Method for Similarity Search in Metric Spaces
VLDB '97 Proceedings of the 23rd International Conference on Very Large Data Bases
| Hi-index | 0.00 |
Advances in data storage and image acquisition technologies haveenabled the creation of large image datasets In order to deal withthese data, appropriate information systems (e.g. image digital libraries) have been developed to efficiently manage such collections. One of the most common retrieval approaches is to employ so--called Content--Based Image Retrieval (CBIR) systems. Basically, these systems try to retrieve images similar to auser--defined pattern (e.g. image example). Their goal is to supportimage retrieval based on content properties (e.g. shape, color, or texture), which are often encoded in terms of imagedescriptors This demonstration presents a new CBIR system based on configurable components. The main novelty resides in its Content--Based ImageSearch Component (CBISC) that supports queries on image collections CBISC is based on the OAIprinciples, and thus provides an easy--to--install search engine tosupport querying images by content As with the OAI protocol, queriesare posed via HTTP requests and the responses are encoded in XML CBISC encapsulates multidimensional indexstructures to speed up the searchprocess. Furthermore, it supports the use of different imagedescriptors (metric and non--metric; color, texture, and shapedescriptors; with 1D or 2D feature vectors), which canbe easily combined to yield improved effectiveness. We will show thatthis search component can be tailored for particular image collectionsby a trained designer, who carries out a clearly defined set of pilot experiments to select the appropriate descriptors. Image descriptors are typically domain and usage--dependent Further, agiven image can be associated with very many descriptors However, standard CBIR methods only support a fixed set of descriptors CBISC, instead, allows progressive extension of thedescriptor base Figure 1 presents a screen shotshowing the CBISC Configuration Tool developed to support CBISC designers in the configuration process Basically, this processconcerns the description/definition of both the image descriptors thatwill be used to retrieve images by content, and the image database to which the CBISC is related The XML file generated in thisprocess is used during CBISC execution.