Tuesday, June 23, 2009

DHO9: 3 presentations on the Archimedes Palimpsest, Language and Image: T3=Text, Tags and Trust, and Mining Texts for Image Terms: the CLiMB Project

1) 'Integrating Images and Text with Common Data and Metadata Standards in the Archimedes Palimpsest' by Doug Emery and Michael Toth of Toth Associates and Emery IT.

a) Image registration
Reconstructed underwriting was begun with multiple overlapping photographs were taken to get 'one' image. The technique chosen so that two images were combined to create a level of contrast between the two different writings. LED illustration Panel was used in 2007 to capture entire leaf in one shot. The result was a series of shots along the visible light wavelength. In some cases, the text was simply gone and using rigging lights, one was able to see the indentation in the palimpsest. There was six additional texts in addition to Archimedes. One author has 30% of his known writings are found in this palimpsest. Question: Ink to Digits to ??? Once this knowledge is captured, how do you perserve it for prosperity given the issues of digital preservation? Goals of the program were to set up a standards for authoritative digital data, provide derived information, etc. 2.4 TBs of raw and processed data were created once the manuscript was digitized. Every image is registered with 8160 x 10880 pixels, TIFF 6.0, eight bits per sample, resolution is 33 per pixel(?).

b) Project Metadata Standards
Contents standard for Digital Geospatial Data were used. Desired long term data set viability beyond current technologies. 6 types of metadata elements used. Metadata stored in TIFF description standards. Uses OAIS. Working on a hyperspectral imaging of a Syriac Palimpsest of Galen. Also, see Walters Digital Tool.

2) Language and Image: T3=Text, Tags and Trust by Judith Klavans, Susan Chun, Jennifer Golbeck, Carolyn Sheffield, Dagobert Soergel, Robert Stein. Practical problem addressed: the limited subject description of an image of Nefertiti as an example. Viewers tag 'one eye, bust of a woman, Egypt, etc.' T3 Project is driven byt this philosophical challenge to show the image and the words to describe it. Text: Computational Linguistics for Metatdata Building (CLiMB), Tags: Steve. Museum collects tags, and Trust: a method of computing interpersonal trust. Each 'T' had a different goal: CLiMB was used for text while working with catalgoers, Steve (Art Museum Social Tagging Project) was working with the public to find the images they wanted through tagging; and Trust, amiguity and synonymy are the biggest hurdles. Trust (http://trust.mindswap.org) is useful as a weight in recommender systems. Some research questions posed: How to compute truste between people from their tags? Can clustering trust scores lead to an understanding of users? What types of guidance are acceptable? Does 'guidance' lead to 'better' tagging?
Current work of CLiMB:
1. analyze tags from steve with CL tools (50,000 tags): morphology and ambiguity issues (e.g. gold as metal or color?)
2. experiments on trust in the T3 setting: how to relate trust in social networks
3. consideration of tag cloud for ambiguity: blue (the color or your mood?)
4. review of a guided tagging environment

3) Mining Texts for Image Terms: the CLiMB Project by Judith Klavans, Eileen Abels, Jimmy Lin, Rebecca Passonneau, Carolyn Sheffield, Dagobert Soergel. Judith Klavens of the University of Maryland also gave this presentation. Image catalogers-->catalog records-->image searchers. Terms are informed by art historical critieria, ability to find related images and leverage meaning from the Art and Architecture Thesaurus (AAT). One technique to solve disambiguation is to use SenseRelate to find correct meaning of the head noun, compare the correct meaning's definition to the definition of all the AAT and compare the results.

No comments: