Hi, There are books on font identification algos for people to use - size and bracket on n serifs, are the a or g double story, ratio of ascender to x height in h, ratio of i stem, stuff like this - and PANOSE2 has a load of this stuff. Font classification is a recent interest of mine, need to know more about it...
Regards, Dave On 8 May 2009, 11:00 AM, "Ed Trager" <ed.tra...@gmail.com> wrote: Hi, all, On Fri, May 8, 2009 at 10:36 AM, Ben Weiner <b...@readingtype.org.uk> wrote: > Hi, > > Alexandre Pro... What the Font indeed must work by analyzing bitmaps more or less using the principle I described in my previous posting of subtracting a test bitmap from a known bitmap in the database and looking at the "residue" left over : less "residue" means a better match. It still seems like a hard problem to me: First, in the case of a system like WhatTheFont, you must have a good algorithm for aligning and scaling bitmaps to the right size before trying to subtract one from the other. Secondly, if you have a large database of bitmaps, just using a brute-force approach to match the test glyph bitmap against every bitmap in the database seems inefficient ... Ideally one would want a way to create some sort of digital "fingerprint" from the full bitmap that could be used as an index key for rapid retrieval of closely-matched glyph bitmaps. Of course there have got to be ways to do this. But, as I said, it seems like quite a bit of work to me ... In fact, I wish I knew about some of the ideas for doing such "fingerprinting" of similar images for the purposes of indexing, etc. : Knowing how to do that would also provide a nice way to show a user related fonts. Similar to what web sites like Amazon Books does, but for fonts: "If you like this font, take a look at these similar fonts ..." - Ed > Cheers, > Ben > > -- > Ben Weiner | http://readingtype.org.uk/about/contact.html > >