----- Forwarded message from "F. James Rohlf" -----
Date: Mon, 30 Jun 2014 13:15:24 -0400
From: "F. James Rohlf"
Reply-To: ro...@life.bio.sunysb.edu
Subject: RE: Image rotation data loss
To: morphmet@morphometrics.org
My tps software explicitly takes images as they are and does not reflect them
(except by a direct transformation in tpsDig2). Some equations for Procrustes
fitting are given on page 42 of Rohlf and Slice (1990, Systematic Zoology).
While straight-forward matrix algebra for a Procrustes fit would automatically
reflect if it would achieve a better fit, we used Gower’s (1971, Biometrika)
formulation that prevents reflections. In recent versions of the software I
use a method based on complex regression because it is somewhat faster and
still prevents reflections.
The reason for not permitting reflections is that it is possible for some
simple shapes to vary in ways that look like reflections although the organism
has not actually been reflected. However, general purpose software has to guard
against cases in which some of the images are reflected relative to the others.
I guess I could/should have added it as an option in the software - but that
has not happened yet. An illustration of some of the statistical problems in
blindly ignoring reflections are described on page 680 of Rohlf (2003, J. of
Human Evolution). In certain cases it can introduce a systematic bias. This is
one of the problems with EDMA and other morphometric methods that ignore
reflections.
A simple example that I often use in class is shown on page 127 of the “white
book”. One could create similar examples for 3D landmarks. This point is
normally not important in practice because more landmarks are usually used and
their arrangement usually does not allow their actual shape variation to be
confused with their reflections.
Hope this helps.
----------------------
F. James Rohlf, Distinguished Professor Emeritus, Stony Brook University
The much revised 4th editions of Biometry and Statistical Tables are now
available:
http://www.whfreeman.com/Catalog/product/biometry-fourthedition-sokal
http://www.whfreeman.com/Catalog/product/statisticaltables-fourthedition-rohlf
P Please consider the environment before printing this email
From: morphmet_modera...@morphometrics.org
[mailto:morphmet_modera...@morphometrics.org]
Sent: Monday, June 30, 2014 1:03 AM
To: morphmet@morphometrics.org
Subject: Re: Image rotation data loss
----- Forwarded message from Joseph Kunkel -----
Date: Sun, 29 Jun 2014 19:47:55 -0400
From: Joseph Kunkel
Reply-To: Joseph Kunkel
Subject: Re: Image rotation data loss
To: morphmet@morphometrics.org
I know that in Rohlf's tpsSuite of software that the geometric equations allow
for accepting left and right wing data without reflecting them. I do the
algebra myself in R and there I also find that the matrix algebra will
automatically fit a left or right wing onto the model. Mirror wings are a
simple rotation onto the model. Being mostly 2-D with wings, I would have to
think it over whether the 3-D landmarks would be treated similarly. Jim needs
to chime in here about the 3-D symmetry and how automatic the matrix algebra
and rotation matrices deal with symmetry.
Joe
On Jun 29, 2014, at 4:44 AM, morphmet wrote:
-------- Original Message --------
Subject: Re: Fwd: Re: Image rotation data loss
Date: Sun, 29 Jun 2014 04:43:20 -0400
From: Dennis E. Slice
To: morphmet@morphometrics.org
Also, your message suggests all images were in the same orientation. If some
were reflected and others not, I don't think tps allows reflection by default.
-ds
On 6/29/14, 3:53 AM, morphmet wrote:
-------- Original Message --------
Subject: Re: Image rotation data loss
Date: Sun, 29 Jun 2014 03:38:32 -0400
From: Dennis E. Slice
To: morphmet@morphometrics.org
First guess would be your landmarks are out of order on one or more
specimens. You can check this by creating a set of links in tpsUtil and
plotting the data showing the links. And/or, in tps, plot the data
showing points and vectors. -ds
On 6/29/14, 12:27 AM, morphmet_modera...@morphometrics.org wrote:
Hello all,
I'm trying to digitize photos in tpsDig for a geometric morphometrics
study on Notropis fishes. In order to test for digitization error I've
copied the first fish photo I took 10 times and digitized each one onto
the same TPS file (made in tpsUtil). I then performed a relative warps
analysis in tpsRelw and performed a Procruste's superimposition in R
(package geomorph) to glance and see if any landmarks varied
significantly. My results show very drastic variation in landmark
placement, with dots all over the place in the GPA plot. I know my
landmark placement isn't that bad, so I think it might have something to
do with data loss. When I photographed the specimen the fish was
flipped, so I had to rotate and flip it to turn it rightside up before
digitization. Is it possible that rotating the photo resulted in data
loss and therefore digitization error? The TPS file had the landmark
coordinates for each specimen pretty close together (about normal
landmark placement error), but the GPA and relative warps plots showed
drastic variance. I've attached the GPA plot to show what I mean. The
landmarks are supposed to go around the entire fish. Also let me know if
you have any other ideas for why there might be crazy variance.
Best,
Connor
connorfre...@utexas.edu
----- End forwarded message -----
----- End forwarded message -----
----- End forwarded message -----
-·. .· ·. .>·. .· ·. .>·. .· ·. .> .··.· >=- =º}}}}}><
Joseph G. Kunkel, Research Professor
112A Marine Science Center
University of New England
Biddeford ME 04005
http://www.bio.umass.edu/biology/kunkel/
----- End forwarded message -----
----- End forwarded message -----
