-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I have to develop a GPL project for character recognition. A first draft
mention that a method using "PAKKER & CHEHIKIAN" can find back the
spelling with a ratio of 97%.

When I google for those names, I have 0 ( zero, nada, whaloo ) response.

Can any one give me any details about those people, and some description
of their work ?

On my side, I use the Hough Transform for some personal projects, namely
http://www.demaine.info/projet_self_tracking_webcam/

I know that the HT can be used for character recognition, such as
chisees alphabet .

I have endembed contrainsts about space (physical volume), CPU, RAM, and
energy ( autonomy).

I d like to have any idear if a neural network is easy to implement on
DSP, or if HT is better, and whiche one of those 3 algos is faster.

Learning/teaching will be done at the manifactury. Everything will be
encoded in ROM. But I dont know yet what CPU will be available. All I
know is that I must threat at least 15 char/sec, with picuters between
60x40 and 512x256 ( one to 3 chars per picture), and that I must do my
best to use fewest energy and physical volume as possible. The spec say
that I shall recognise between 40 and 64 characters only ( caps, small
letters, digits, and few punctuation )

Il it now possible to implement Neural Net on FPGA ?
Is it possible to encode HT or "PAKKER & CHEHIKIAN" in FPGA ?

if I have to go for full custom or DSP, which one will be fastest, or
consume less current/power/RAM ?

I know that all that is very fuzzy; but I am asked to make technical
choice when I have not been told about the hardware ...

Thanks for any help.

- --
DEMAINE Beno�t-Pierre http:/www.demaine.info/
\_o< apt-get remove ispell >o_/
There're 10 types of people: those who can count in binary and those who
can't
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBj2aNGWSTLbOSw8IRAu/EAJsG68epCPXZc/DcOkv3uKmEYW50iACeMd26
Cv6fN/wBOjUTDtEP0G/YugE=
=gZOC
-----END PGP SIGNATURE-----

On Mon, 8 Nov 2004, DEMAINE Benoit-Pierre wrote:
>
> I have to develop a GPL project for character recognition.

   Gee, who's forcing you? ;)

> A first draft mention that a method using "PAKKER & CHEHIKIAN" can find 
> back the spelling with a ratio of 97%.
>
> When I google for those names, I have 0 ( zero, nada, whaloo ) response.
>
> Can any one give me any details about those people, and some description
> of their work ?

   "PAKKER" might be a (rather ironic!) misspelling of the more common name 
"Parker," Googling for which probably won't help; or it could be Katerin
Romeo-Pakker, who doesn't seem to have a current web presence except as
"K. Romeo-Pakker" in a lot of online bibliographies and the defunct
www.romeo-pakker.com, which was written in French and doesn't seem to
have contained any academic material, from what little I can read of it
in the Google archive.

   Alain Chehikian is apparently associated with the Institut National
Polytechnique de Grenoble, www.inpg.fr, but I can't find a home page
for him either.

   Both researchers seem to have done things with DSP and character 
recognition, but have not coauthored any papers the Web knows about.
Maybe one of them modified the other's basic algorithm?

   Anyway, how did you find the phrase "PAKKER & CHEHIKIAN" without also 
finding either a reference to literature or a minimal description of the
algorithm itself?

HTH,
-Arthur

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

|   Anyway, how did you find the phrase "PAKKER & CHEHIKIAN" without also
| finding either a reference to literature or a minimal description of the
| algorithm itself?

A company just said me :

we have bees said that the algo from "PAKKER & CHEHIKIAN" is very
efficient; we want you to test the efficiency of the HT, and provide us
a report on your comparison.

They will be able to give me the paper about it within 2 weeks. But I d
like some 'external' informations about it before I read the papers.

And all this do not say me if HT is faster than NN ... has any one done
any comparison ?

- --
DEMAINE Beno�t-Pierre http:/www.demaine.info/
\_o< apt-get remove ispell >o_/
There're 10 types of people: those who can count in binary and those who
can't
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBkChJGWSTLbOSw8IRAj49AJoCN9iXQ3zYC4aLi17QNoB4+57wJQCfZZZC
TT9jP+qAIhUibN/enrMWkfo=
=H6Z0
-----END PGP SIGNATURE-----