From here on..

HSB lectures Wed Sept 27 / Recognition / Clustering

Are you going to thoroughly
discuss Lam/Huang/Suen   on multiple classifier combination?

Oct 2  Recognition (RJF away)
Oct 4  Recognition (RJF away)

.........

Oct 9-on  RJF should lecture for a while.. but on what?

What we had in the outline was more on
matching, parameter estimation, properties of
k-nearest neighbors... but this has been previously mentioned, 
and I suspect you may wish to elaborate on this
topic  and/or cover some of it already mentioned,
in more detail on Oct 2-4.

Arguments against RJF doing yet more of
this include: 
a. It's in Jordan's course; 
b. RJF is not an expert on it. 
............

Perhaps useful would be chapter 8 in Bunke&Wang
(Dengel, Hoch..)  Techniques for Improving OCR.
problems with this paper:
 a.  it is a hodge-podge of stuff;
 b.  the most principled is 
     section 8.2 on voting, which is like chapter 3 (Lam/Huang/Suen)

.........
Another possibility is RJF talking about more
ad hoc stuff, like Chapter 21 or 22 in B&W on
math and music respectively.

Or talking about ad hoc attempts to separate characters
or patch up broken ones. (from printed text).
There are ICDAR and 

There is a survey in B&W on removing form artifacts
(Chapter 18), but it seems to be mostly skeleton rather
than meat. References rather than summary. The paper in
O&K by Casey etal "Intelligent Forms PRocessing" and
another Taylor/Fritzson/Pastor  might
be better. These two papers are "here's my system; see it
work" ones.

Another possibility is for me to discuss what Textbridge
as an API looks like, perhaps to give an inkling of what
parts of the OCR processing parameters stick out of
the bundle.  It gives some hints about what must be
inside the box.


I'm off to a meeting, but will write some more later.