W3C home > Mailing lists > Public > public-html@w3.org > February 2009

ISSUE-66 image analysis heuristics

From: Robert J Burns <rob@robburns.com>
Date: Thu, 5 Feb 2009 21:57:07 -0600
Message-Id: <35B9C5C1-61D5-480F-930B-20080F2C8B5A@robburns.com>
To: HTML WG <public-html@w3.org>

Regarding this issue, I agree with the thrust of the sentiment  
expressed by Matthew May here. I'm glad to see it raised.

I'd just like to add that even if we hold out hope for some great  
advances in image analysis and heuristics, the appropriate thing for  
HTML5 to do would be to encourage those features in HTML authoring  
tools as aids to authors of HTML. Perhaps after seeing great advances  
there, HTML6 might call for such tools to be deployed on the consumer  
side of HTML content.

Too often I think we (and HTML5 does this quite often) want computers  
to be better at what humans can do and we neglect to deploy computers  
for what they're already quite good at. We leave the low hanging fruit  
just hanging there. So rather than providing hooks for authors to  
include metadata and providing algorithms that present glossaries and  
indexes and reference lists for data already in the document (all  
things that software can do quite well), we look for software to  
describe an image, surmise an author's intentions (like that the  
author must not have really meant to place a caption element inside a  
table cell), or summarize a table (all things software does quite  
poorly). So I think HTML5 should focus on the low hanging fruit and if  
we say anything about heuristics leave that as a recommendation for  
authoring tools where authors can confirm the results of the  
heuristics rather than relying on consumer UAs where it is then too  
late to confirm or correct the heuristic results.

Take care,
Received on Friday, 6 February 2009 03:57:47 UTC

This archive was generated by hypermail 2.3.1 : Thursday, 29 October 2015 10:15:42 UTC