Christina's LIS Rant
Monday, September 18, 2006
Other issues with gifs and science...
Previously, in my now somewhat dead other blog, I'd spent a bit of time discussing math on the web.
Actually, I think the html-ized articles with the tiny jpeg images of tables and graphs are absolutely the worst.
It seems like we should be farther along at this point. PDFs are really just like having paper -- shouldn't we actually be able to work with the image and search the equations? ¶ 1:37 PM| |cites (technorati) |
- If you need show your math work on the web... (9/04)
- Searching for Math Equations and Symbols on the Web Part 1 (9/04)
- Searching for Math Equations and Symbols on the Web Part 2
- Searching for Math Equations and Symbols on the Web Part 3 (12/04)
- Notices of the AMS: The Importance of MathML to Mathematics Communication (4/05)
Actually, I think the html-ized articles with the tiny jpeg images of tables and graphs are absolutely the worst.
It seems like we should be farther along at this point. PDFs are really just like having paper -- shouldn't we actually be able to work with the image and search the equations? ¶ 1:37 PM| |cites (technorati) |
Comments:
Thanks Christina,
You suggest that images might be turned back into semantic information. I don't know how well this works for maths but it's pretty hairy for chemistry. I've spent some time hacking a prototype and there are several problems:
* the dynamic range is often not as good as it appears to the eye. glyphlike components run into each other even after image processing
* the semantics of components is undefined. A '2' could be a label, an element multiplier, a label, a page number, etc. Humans are good at managinig this; machines aren't.
* components can overlap. Bonds cross and this can be a real pain
* some components can be very small. A double bond (=) sometimes comes out as (:)
P.
Post a Comment
You suggest that images might be turned back into semantic information. I don't know how well this works for maths but it's pretty hairy for chemistry. I've spent some time hacking a prototype and there are several problems:
* the dynamic range is often not as good as it appears to the eye. glyphlike components run into each other even after image processing
* the semantics of components is undefined. A '2' could be a label, an element multiplier, a label, a page number, etc. Humans are good at managinig this; machines aren't.
* components can overlap. Bonds cross and this can be a real pain
* some components can be very small. A double bond (=) sometimes comes out as (:)
P.
# posted by 
: 9/20/2006 11:36 AM
: 9/20/2006 11:36 AM
