I object.
Can anyone lead me to any topics where I can understand how such a system works? How can an image be scanned by a system to find a font closely relating to it?
Questions and answers come in two flavors on Experts-Exchange. One is the "this doesn't work, what am I doing wrong" or "how do I make this do this" type of question with a concert answer.
The other type is the "I have no idea how to go about this, but I want to land a man on the moon and return him safely to Earth...how do I go about it" kinds of questions.
alleon's question is of the second variety. I was willing to engage him in dialogue as far as he wanted to go and provide any necessary links to support the directions he choose.
As far as only offering suggestions, was he expecting source code?
I think that an equally valid answer, and one worthy of the points offered would be a terse "forget about it"
by: D_BruggePosted on 2009-01-28 at 21:37:24ID: 23494521
This is quite a daunting task, and not one for the weak of heart.
The task can be broken down into two parts. The first is shape recognition. You need your software to be able to distinguish what part of the image that has been uploaded is a letter and what is background. Then determine what the letter is from among thousands of variations of the same letter. For instance, you will need to know a capital T in a serif font as well as a sans serif font and the same thing set in italics. You will need to recognize the shape if it is bold face or as well as narrow.
You can search the web for "image recognition algorithm" to find a number of technical papers describing different approaches to the problem.
The second part is to match the letters that are included in the image with a particular font.
Take a look at this:
http://www.fonts.com/FindF
They have presented the basic steps for visual identification.
As you can see, you have to build yourself a database of all of the fonts that you want to offer as possible solutions to the users query. Fonts.com represents 25 foundries and perhaps 200,000 faces.
For each face you would create a flow chart using something similar to the fonts.com system.
You could build a database using a check off list for each face such as...
Capital letter Q crosses the circle (yes or no),
Capital letter Q touches the circle (yes or no),
Capital letter Q below and separated from the circle (yes or no),
Capital letter Q tail extends or lies inside circle (yes or no),
Capital letter Q circle is open, tail part of same stroke (yes or no).
(Do this for each variation being tested.)
Then it's simply a matter of building a series of queries for the given letters to narrow down your possibilities and present your answer.
(Actually, when you see what all is involved, it's amazing that these programs are offered for free!)
Best of luck to you.