[sci.image.processing] Mathematical Morphology Software Vers. 4.0 available for ftp.
Alan Peters
Version 4.0 of my mathematical morphology software is now available
for anonymous ftp at 129.59.100.16.
Here is a description. It is followed by instructions and a
description of the 4.0 changes.
Software for 2D and 3D Mathematical Morphology
written by
Richard Alan Peters II
Assistant Professor of Electrical Engineering
Vanderbilt University School of Engineering
Nashville, TN 37235
(615) 322-7924
ra...@vuse.vanderbilt.edu
Mathematical morphology is a powerful tool for image analysis and enhancement.
Morphological operators are shape-dependent, nonlinear image transforms such
as erosion, dilation, opening, closing, and rank filters.
The mathematical operators are defined in $n$ dimensions so it is possible to
create programs that will operate on 1D signals, 2D images, or 3D datasets
using exactly the same concepts.
As operators on 2D binary images, mathematical morphology is well known;
much software is available which performs binary morphology.
However, the true power in morphology lies in its ability to transform grayscale
2D pixel images and 3D voxel datasets.
Software for these operations is less common.
This is a morphological software package written under the
auspices of the US Airforce Office of Scientific Research at Arnold Airforce
Base, Tullahoma, TN during the 1991 Summer Faculty Research program.
Subsequently, it has been in continual use and has undergone numerous
enhancements.
The software includes a 2D image morphology program, a 3D voxel image
morphology program, a program for enhancement and noise reduction of 2D
images, and related support routines for image arithmetic and logical
operations. There are standard unix man pages for all programs.
These programs will perform many -- if not all -- of the possible
morphological operations on both binary and grayscale images.
They include erosion, dilation, opening, closing, rank filters, LUM
filters, etc. The 3D image program is ideally suited for the processing
of image time-sequences as well as true 3D data.
The programs provide many of the commonly used features automatically
(such as disk shaped or ellipsoidal structuring elements) or permit
the user to customize the operation to fit an application.
The user programs operate on Sun rasterfiles.
However, the main morphological routines are c-code subroutines.
These are independent of specific file formats, so a users can easily
write an interface program to operate on other file formats.
The software was written on Sun Sparcstations under SunOS Release 4.1.3.
However, I used no indigenous libraries other than libm.a.
The user interface is of the standard csh command-line variety.
This should make porting to other unix platforms trivial (Yeah, right!).
This software is freely redistributable, providing I and my school get
appropriate credit where due.
I am not in a position to actively support development of this software.
Nor do I have a lot of time to answer questions about it, although I'll
make every attempt to do so.
The software was used by a dozen grad students in a seminar class this
spring; so I'm reasonably sure the most egregious bugs have been squashed.
Please do report any bugs to me, though. I'll do what I can to fix them.
Enjoy!
-Alan.
The following are the programs included in version 4.0 of the
mathematical morphology software:
Morphology programs:
morph - 2D mathematical morphology. It performs erosion, dilation
opening, closing, rank, tophat, bothat, LUM, LUM smoothing,
LUM sharpening, and binary shape removal. It does both
binary and grayscale operations as either set or function
operations.
morph3d - 3D mathematical morphology. It performs erosion, dilation
opening, closing, rank, maxmin, minmax, LUM, LUM smoothing,
LUM sharpening, and binary shape removal. It does both
binary and grayscale operations as either set or function
operations.
mclean - 2D morphological noise reduction program that uses
morphological size distributions for grayscale image noise
reduction while preserving small features.
Arithmetic programs:
andimg - Perform a pixelwise logical AND of two images.
avg3d - Performs a pixelwise running nth order combination on an
image time-sequence. Combination types are average, minimum,
maximum, median, rank, ordinary linear combination, and
rank-ordered linear combination.
comb3d - Performs a pixelwise frame-by-frame combination of n
image time-sequences or 3D voxel images. Combination types are
average, minimum, maximum, median, rank, ordinary linear
combination, and rank-ordered linear combination.
combimg - Performs a pixelwise frame-by-frame combination of n
images. Combination types are average, minimum, maximum, median,
rank, ordinary linear combination, and rank-ordered linear
combination.
linco - Performs a linear combination of two byte images. Performs
6 different kinds of output scaling: (1) clip if out of
bounds, (2) linear compress if out-of-bounds, (3) linear
scale of min,max to black,white, (4) scale up if max < white,
(5) scale down if min > black, (6) affine scaling using user
parameters. It can also be used to scale single images.
maximg - Perform a pixelwise maximum of two images.
minimg - Perform a pixelwise minimum of two images.
orimg - Perform a pixelwise logical OR of two images.
Rasterfile programs:
rasremap - Creates a grayscale rasterfile from an arbitrary 8-bit
rasterfile. Causes pixels to at once reference grayscale
colormap and reflect true grayscale value.
ras2raw - Creates a byte array from an 8-bit rasterfile.
raw2ras - Creates an 8-bit rasterfile from a byte array.
Script files (using morph):
bclose binary closing
bdilate binary dilation
berode binary erosion
bopen binary opening
brank binary rank filter
cleanup example of image noise reduction using rolling ball
gfclose grayscale function-function closing
gfdilate grayscale function-function dilation
gferode grayscale function-function erosion
gfocco grayscale function-function openclose - closeopen
gfopen grayscale function-function opening
grank grayscale rank filter
gsclose grayscale function-set closing
gsdilate grayscale function-set dilation
gserode grayscale function-set erosion
gsocco grayscale function-set openclose - closeopen
gsopen grayscale function-set opening
hitormiss binary hit-or-miss transform
invroll inverted rolling ball transform
invtop inverted tophat transform
isodel binary isolated delete
medge morphological edge detector
rollball rolling ball transform
thresh threshold a grayscale image
tophat tophat transform
Installation instructions:
1. cd to whatever directory in which you want to have the installation
procedure create and fill a directory called "morph".
2. ftp to image.vanderbilt.edu and get morph.tar.Z. A transcription of
a typical session follows:
% ftp 129.59.100.16
Connected to 129.59.100.16.
220 image FTP server (SunOS 4.1) ready.
Name (nnn.nn.nnn.nn:xxx): anonymous
331 Guest login ok, send ident as password.
Password: [your email address]
230 Guest login ok, access restrictions apply
ftp> cd /pub
250 CWD command successful.
ftp> bin
200 Type set to I.
ftp> get morph.tar.Z
200 PORT command successful.
150 Binary data connection for morph.tar.Z (n.n.n.n,xxxx) (243551 bytes).
226 Binary Transfer complete.
local: morph.tar.Z remote: morph.tar.Z
243551 bytes received in 2.9 seconds (81 Kbytes/s)
ftp> quit
221 Goodbye.
3. dearchive the morph directory and its contents:
zcat morph.tar.Z | tar xvf -
x morph/Makefile, 1593 bytes, 4 tape blocks
x morph/arith/maximg.c, 2437 bytes, 5 tape blocks
x morph/arith/linco.c, 7716 bytes, 16 tape blocks
...
x morph/.nse_depinfo, 201 bytes, 1 tape blocks
x morph/man/man1/morph.1, 9134 bytes, 18 tape blocks
x morph/man/man1/linco.1, 3107 bytes, 7 tape blocks
4. change access permission of directory "morph" (if you want to)
chmod 755 morph
5. cd to morph and make either all of morph
cd morph
make
or just the libraries
cd morph
make libraries
version 4.0 changes
(1) The 'auto' SE flag now makes SE's with even dimensions. e.g. 2x2
(2) The rasterio utilities (in morph/utils) were rewritten to properly
process images with an odd number of pixels per row.
Also, OpenFile now decompresses unix-compressed (*.Z) rasterfiles.
(3) Three new rank-order dependent filters were added,
LUMFilt, LUMSmooth, and LUMSharp.
LUMfilt is the lower-upper-middle filter as defined by
Hardie, R. C., and C. G. Boncelet, "LUM filters: A class of
rank-order-based filters for smoothing and sharpening,"
IEEE Trans. Signal Processing, vol. SP-41, No. 3, March 1993.
LUMSMO is the LUM smoothing filter defined therein, and LUMSHA
is the LUM sharpening filter.
These filters compare the center pixel in a neighborhood defined
by an SE to upper and lower order statistics in the neighborhood.
Depending on the ordering either the center pixel or one of
the order stats is output.
(4) The 3D rankfilter now agrees with the 2D that rank == 1 == max
(or dilate) and rank == support-of-structuring-elment == min (erode).
--