Response to Chandler et al.'s
comment on: "Hillslope topography from
unconstrained photographs" by A.M. Heimsath and H. Farid,
Mathematical Geology, 34(8): 929-952, 2002
We cannot comment on the content or accuracy of press-releases
generated by our paper. We can suggest, however, that the reader of
any mass-media should be discerning enough to understand that such
material is meant as a summary, and not a thorough analysis, of a body
of work. With respect to our paper, we suggest a careful reading of
the introduction and discussion sections, where we place our
methodology within the context of previously existing photogrammetric
techniques.
We will comment here on the technical points raised by Chandler et
al.'s comment.
We do not propose that our technique will supplant existing techniques
for estimating digital elevation models (DEMs). Our technique, like
any other, has benefits and drawbacks, all of which are clearly
described in our paper. One benefit, for example, of our technique is
that we are able to estimate a DEM from three (or more) photographs
taken with a hand-held camera with no constraints imposed on the
camera positions or field survey. One drawback, at this point, is
that our technique might not produce as accurate DEMs as might be
extracted from careful aerial photogrammetry or laser altimetry. We
are working currently on formalizing a comparison between our
techniques and these methodologies for generating DEMs.
With respect to the issue of lens distortion, we stated clearly that a
simplified pinhole camera is assumed (first sentence of the Imaging
Model section). The effects of lens distortion from even a mid-range
quality camera are generally minimal and have little impact on the
reconstructed DEMs (see, for example, H. Farid and A.C. Popescu.
Blind removal of lens distortions. Journal of the Optical Society of
America, 18(9):2072-2078, 2001).
With respect to the paraperspective projection, we wrote (p. 946,
Paragraph 1):
"Computationally, this technique begins with a paraperspective
approximation to the geometry of image formation. This
approximation affords a closed-form analytic solution for surface
topography, and is further refined through successive non-linear
minimizations that assume a more realistic [perspective] imaging
model, and imposes an overall smoothness constraint on the
recovered structure."
The suggestion by Chandler, et al. that we only employ a
paraperspective imaging model is simply incorrect.
With respect to the "subduing" effect, we wrote (p. 945, Paragraph 2):
"As can be seen in Figures 5-8, there is a consistent flattening
of the estimated structure. This is due most likely to the
initial paraperspective approximation..."
Chandler, et al.'s suggestion for the cause of the subduing in the
estimated DEM is a puzzling reiteration of a point we already made.
The flattening effect on the crests of the noses may also be partially
due to low point selection densities in these areas. We are currently
investigating this further.
With respect to the use of our methodology versus standard
photogrammetric techniques, we trust that researchers are sufficiently
discerning to determine when and if a technique will suit their needs.
Finally, we did, in fact, write the paper with freely available software packages, LaTeX (www.latex-project.org) and GNU Emacs (www.gnu.org/software/emacs). We are also making freely available the source code for generating DEMs using our method (www.cs.dartmouth.edu/farid/research/phototop). This code runs under MatLab (R) or the freely available GNU Octave (www.octave.org). We strongly believe in and support the concept of open source code, where a community of like-minded users benefit from and contribute to the advancement of a common computational or scientific goal.