References about i.smap

[James Darrell McCauley]

O. Hellwich (olaf at photo.verm.tu-muenchen.de) writes on 17 May 94:
>Does anybody know whether the article
>
>Bouman, Shapiro, "A Multiscale Random Field Model for Bayesian Image
>Segmentation" submitted to IEEE Transactions on Image Processing,
>
>as quoted in the Grass User's Manual, was actually published?

yes, it is in the most recent issue and the title is the same (sorry,
I left the exact ref at home. I'll send it to you (and anyone else who
personally requests it) directly).

(sorry ((about the double parens) (I (used to) program in LISP)) :)

BTW, I've used i.smap for some recent work and it performed very well.
I didn't utilize a mutimodal mixture model for training data (because I
was trying to do an apples-to-apples comparison with ML and another
algorithm), but I suspect that SMAP would have done even better if I
had.

I'm considering presenting/submitting some of my results with SMAP
somewhere, but in the mean time, anyone interested may follow the
"recent work" link at the bottom of my home page (see below) for an
abstract which leads to a first rough draft reporting the work (beware
it's 7.7MB uncompressed and includes color postscript)

I recall that, since this draft was originally written for a
"non-believer," there's a detailed procedure of how this multispectral
classification was/is done in GRASS included as an appendix, which may
be of general interest.
(i.tape.other/r.in.poly/i.gensigset/i.smap/r.kappa/m.ipf/i.texture/r.clump),

[Re: why (not) ARC (for me): ...just another example of how GRASS/OGIS
is on the cutting edge.]

--Darrell

James Darrell McCauley, Purdue Univ, West Lafayette, IN 47907-1146, USA
mccauley at ecn.purdue.edu, mccauley%ecn at purccvm.bitnet, pur-ee!mccauley
** will finish PhD/engr in 9/94 - need job. inquiries welcome (no hh, plz) **
** Personal WWW page URL ftp://pasture.ecn.purdue.edu/pub/mccauley/me.html **

P.S. for those without net.access, I've included the abstract (minus
the pretty picture and paper) below.

<H1> Comparison of Scene Segmentations and Classifications:
SMAP, ECHO, and Maximum Likelihood</H1>
<H2> James Darrell McCauley </H2>

Sequential maximum <I>a posteriori</I> (SMAP) segmentation, the
Extraction and Classification of Homogeneous Objects (ECHO)
segmentation, and maximum likelihood (ML) estimation were compared in
a supervised classification of multispectral data collected from an
airborne scanner.  The generalization of SMAP provided larger and
fewer contiguous areas and a more visually acceptable <A
HREF="images/smapecho.gif">map</A> than ML or ECHO.  Iterative
proportional fitting was used to to normalize error matrices and a
Tukey multiple comparison test was used to examine differences in
classification results.  At a risk level of $\alpha=0.001$,
significant differences were found in all mean class classification
accuracies: SMAP &gt ECHO &gt ML (in order of decreasing accuracy).
However, evaluation through pairwise analyses of $\kappa\/$ statistics
showed no significant difference ($\alpha=0.1$) between any two
classifiers. Iterative proportional fittings and a Tukey multiple
comparison test were found more sensitive and preferred over pairwise
$\kappa\/$ analyses.<P>

A report for David Landrebe, submitted 20 Apr 1994.<P>

<A HREF="ftp://pasture.ecn.purdue.edu/pub/mccauley/papers/smap-echo-ml-compare.ps.gz">
A gzipped PostScript copy of the paper is available.</A>
<P>

<ADDRESS> James Darrell McCauley (mccauley at ecn.purdue.edu) </ADDRESS>