Message-Id: <mailto:199407182134.QAA16120@library.wustl.edu> Date: Mon, 18 Jul 1994 14:38:33 -0400 From: Donna Romer <mailto:romer@KODAK.COM> Subject: Visual Resources User Model To: Multiple recipients of list IMAGELIB <mailto:IMAGELIB@ARIZVM1.BITNET>
Date: 07/18/94 14:33:31 To: mailto:IMAGELIB@ARIZVM1.ccit.arizona.edu cc: LZA200 --KR25 ROMER DONNA M>From: Donna Romer, CD Imaging, 3/12/HE
Subject: Visual Resources User Model A few responses to the list members who commented on my earlier note:* Compositional qualities are MOSTLY objective visual issues. Images have identifiable formats and points of view (such as where a camera might be positioned vertically or by distance from the subject). But there are a range of compositional qualities that we can SEE but we have a hard time articulating. Quite often in the language of the artist there is a technique employed that contributes the unnameable quality. For example, there are techniques in photography called FRAMING, or PATTERN REPETITION. FRAMING is the artist's way of taking subject matter and "framing" it using other elements in the scene. Images where the subject is seen through architectural columns or through a window are examples. PATTERN REPETITION is a common technique that juxtaposes the same KIND of object (more or less) in multiples to fill the image frame. Images of leaves or balls or even a field of flowers can qualify. But the interesting thing I found in interviewing people who search for images, was that they typically did not have the same name ( or a name at all) for the technique in question. The language of the searcher was very distant from the language of the artist was the overall point I came away with. In addition, it was clear that the image search process needs these dimensions to arrive at satisfactory results. I would propose that good cataloging on visual formalisms will allow the kind of BROWSE process that eludes us today. As was mentioned in an earlier note, the image search process does not always begin with words. Again in my research, browsing through catalogs of images was the most common way to start the image search process before calling the stock photo agency with a request. I would like to see that current cataloging methods are augmented in this way with visual cataloging materials attached to existing work flows. In image databases where there are pottentially millions of images, compositional vocabulary will be the "difference that makes the difference" in browsing, searching and refinement processes.
* Subjective characteristics are indeed impossible, because of the great variability of image perception across people and even within the same person at different time periods. But it is still a quality that searchers want in an image database. I think that there are interesting possibilities in the area of knowledgebase and semantic network technology that are applicable here. The composite of Object/Action/Composition and other attributes have always seemed to me to be ripe for something like a "subjective equation." At the moment in the Kodak Picture Exchange the stock agencies are entering conceptual values about their images like, threatening or puffy or all-american. I am sure you all can surmise the trouble that is caused by this kind of scenario. I have several ideas in this area that I would like to see researched - and eventually I hope someone with the research budget will attack the problem.
* Last point: In a previous note it was stated exactly right, that the place to start with an image database model, is with the way that people search for images. Starting with a User Model has provided a wealth of understanding for the Kodak project and I would encourage anyone trying to create a system to spend a lot of time in this preparatory phase.
~ Donna Romer