I am doing a research project of content-based image retrival using oralce92. I am interested in using Oracle's content-based retrieval feature for images.Can anyone tell me more about how the feature vectors or signatures are created from the color, texture, and shape in an image? Is there any way to extract these feature vector.
Also, when two images are compared, the similarity measure is calculated as the score or distance between their attributes. What type of distance is used (e.g. Euclidean, Manhattan)?
I am also considering CBIR for mu thesis work, but as far as I have done the first scan of literature, tools, ongoing open source projects and work of big companies in this area, now I have some impression that is is somehow the theme of past. There was nice Oracle8 extension (and also documentation) from Virage, but not it is hardly to find anything about this in their homepage. From the research point of view this can be nice theme - special algorithm, indices and so on, but - there can be problems with application. However - in thi first half of this year some guys from Google published nice articel of Visual PageRank and I guess - if Google will go to stand for this theme, then there can be some interest from big DBMS vendors as well.