Smoothing methods
Eric Martz (emartz@microbio.umass.edu)
Tue, 27 Jun 1995 12:17:14 -0500 (EST)
Can anyone out there help me find a histogram smoothing algorithm which
doesn't generate artifactually bimodal histograms from unimodal ones?
A couple of years ago, I incorporated Savitzky-Golay smoothing into
my program MFI. Then I found this artifact and built-in lots of checks
to avoid misleading people who enable smoothing. Recently, I noticed
a different algorithm in Tom Bakker Schut's dissertation, which he used
in his program ANALYSE. I coded it into MFI and found the same type of
problem occurred but with different histograms. I wrote the C code for the
latter, but am using the Quinn-Curtis commercial library implementation for
Savitzky-Golay (I have the full source code). Since Savitzky-Golay leaves
up to 6 points at each end undefined, I used a linear interpolation on
them. This may be part (or all?) of the problem. But Schut's method
defines all points. I can FAX sample plots (unsmoothed and smoothed by
each algorithm) and more details if you think you can help. I haven't dug
up a text which describes smoothing methods and their limitations, so any
help is welcome. Please include your FAX number if you want to see the
plots.
/*- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Eric Martz, Professor of Immunology emartz@microbio.umass.edu
Dept Microbiology Voice: 413-545-2325 FAX: 413-545-1578
Morrill IVN 203, Box 35720, Univ Massachusetts, Amherst MA 01003-5720
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -*/