morph(1) morph(1)
Richard Alan Peters II Richard Alan Peters II
10 May 1993
NAME morph - 2D image morphology
SYNOPSIS
morph <In >Out -m e|d|o|c|r|t|b|l|m|n|p|q [-i g|b] [-s g|b]
[-o s|f] [-t nnn [mmm] ] [-l kkk lll ] [-r med |nnn]
[-z] [-n] [-v]
-k SEFile | 3x3 | 5x5 | plus | auto xxx yyy [zzz]
AVAILABILITY
The software was written by
Richard Alan Peters II, Ph. D.
Department of Electrical Engineering
Vanderbilt University School of Engineering
Nashville, TN 37235
rap2@vuse.vanderbilt.edu
It is available directly from the author.
DESCRIPTION
This program performs many (if not all) possible 2D morphological
operations on grayscale or binary images. This includes hit-or-miss
transforms, order-statistic filters, lower-upper-middle filters,
function-processing or set-processing erosion, dilation, opening and
closing, tophat (image minus opening), bothat (closing minus image),
and isolated delete functions. For an excellent tutorial on image
morphology, see Haralick, R.M, S.R. Sternberg, and X. Zhuang, "Image
analysis using mathematical morphology," IEEE Trans. PAMI, vol.
PAMI-9, No. 4, July 1987, pp. 532-550. Or see Maragos, P. and R.
Shaffer, "Morphological systems for multidimensional Signal
Processing", Proc. IEEE, vol. 78, no. 4, April 1990. For additional
info on the hit-or-miss transform, see J. Serra, Image Analysis and
Mathematical Morphology, Academic Press, London, 1982. For additional
info on the lower-upper-middle filter, see 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. Program morph performs morphological operations on Sun
RasterFiles using the subroutine morph_sub(1). Program morph is a
user interface that decodes the user specified options, sets up the
arguments for morph_sub, and reads and writes rasterfiles. Subroutine
morph_sub actually perfoms all the morphology. Users wanting to do
morphology on image formats other than Sun raster files, are
encouraged to write a program, like morph, that calls morph_sub.
OPTIONS
Options may appear in any order. In is the path name of the Sun
rasterfile input image. (Type <In so the file is read in through
stdin.) Out is the pathname of the Sun rasterfile output image. (Type
- 1 - Formatted: November 6, 2025
morph(1) morph(1)
Richard Alan Peters II Richard Alan Peters II
10 May 1993
>Out so the file is output through stdout.) Program morph extracts and
operates on the luminance component of the image. Hence, the output is
grayscale, even if the input is an 8-bit color rasterfile. morph does
not operate on 1-bit binary nor 24-bit true-color rasterfiles. The
letter following -m indicates the morphological operation:
e - erode
d - dilate
o - open
c - close
r - rank filter
t - top hat transform (image minus opening)
b - bot hat transform (closing minus image)
l - general LUM (lower-upper-middle) filter
m - LUM smoothing filter
n - LUM sharpening filter
p - I && !E (pixelwise) where I = original image;
E = eroded image;
q - D && !I (pixelwise) D = dilated image;
one of these letters must be specified; there is no default.
l is the lower-upper-middle filter as defined by Hardie and
Boncelet, (reference given above). m is the LUM smoothing filter
defined therein, and n 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. Note that we define order statistics
opposite Hardie and Boncelet Whereas OS(1) is the minimum for
them OS(1) is the maximum in these routines. p with a binary
hit-or-miss structuring element (SE) will delete in a binary
image, white features with the shape of the "hit" portion of the
SE. (e.g. one can easily devise a SE to delete isolated pixels).
p with a grayscale SE will delete the "interiors" from sets of
white pixels in a binary image. q with a binary hit-or-miss
structuring element will delete in a binary image, black features
with the shape of the "hit" portion of the SE. (e.g. one can
easily devise a SE to delete isolated pixels). q with a
grayscale SE will delete the "interiors" from sets of black
pixels in a binary image.
Switch -i indicates that the next letter tells the image type: either
g for a grayscale image or b for a binary image. If -i is not
included, the default is grayscale. Switch -s indicates that the next
letter tells the structuring element (SE) type: either b for a binary
SE or g for a grayscale SE. If -s is not included, the default is
grayscale. The letter following -o , either s or f , indicates that
the operation is either a set operation or a function operation. (See
reference.) If -o is not included, the default is set op. -t nnn
[mmm] indicates that a threshold of value nnn from below (and mmm from
above; if unspecified mmm == 255 ) will be used on the input if the
following 2 criteria are true: the input is a grayscale image AND the
- 2 - Formatted: November 6, 2025
morph(1) morph(1)
Richard Alan Peters II Richard Alan Peters II
10 May 1993
operation is a set operation. If the two criteria are true and -t nnn
[mmm] is not included, the operation is treated as a function and set
processing (FSP) operation (See the Maragos paper). If the criteria
are not true and -t nnn is specified anyway, it is ignored. Note that
you can do a simple threshold at 128 of a grayscale image with: morph
< in.ras > out.ras -m e -t 128 -k auto 1 1 Switch -r is meaningful
only if -m r is specified. Then the field following -r indicates the
order of the filter. If the letters " med " are in the field, a median
filter is used. If the field contains a number, nnn , then that value
is used. If -m r is specified and -r med | nnn is not, the rank filter
defaults to a median filter. Switch -l is meaningful only if -m l, -m
m, or -m n is specified. then kkk and lll are integers that
correspond to the the values in Hardie and Boncelet (referenced
above). If -m l is specified, the general LUM filter is selected,
then integers kkk and lll in the statement -l kkk lll must satisfy 1
<= kkk <= lll <= med == int(SESupport / 2) + 1, where SESupport is the
number of active pixels in the structuring element specified by the
argument(s) following the -k flag. If -m m is specified, the LUM
smoothing filter is selected, then integer kkk in the statement -l kkk
lll must satisfy 1 <= kkk <= med == int(SESupport / 2) + 1. The value
of lll is ignored. If -m n is specified, the LUM sharpening filter is
selected, then integer lll in the statement -l kkk lll must satisfy 1
<= lll <= med == int(SESupport / 2) + 1. The value of kkk is ignored.
To have morph tell you the value of int(SESupport / 2) + 1, execute
the program with -l 0 0 as well as the -k SE specification.
The presence of switch -z tells the program NOT to zero-pad the
boundary of the image. With this option, the output image has a border
of zeros inside it the width and height of the SE. That is, the
transformed area of the output image is smaller than the actual image
dimensions. This is, in a sense, a more accurate result than the zero
padded default. To zero pad the input permits the program to transform
the border region, but it does this on the assumption that the
original scene was black outside the image. This, of course, is almost
never true. Thus, the border region is inaccurately transformed. Use
this switch if accuracy is more important than having an image that is
"colored in" out to the boundary. Switch -n tells the program NOT to
scale the output of the operation. Such scaling happens by default
for a function operation (-o f) on a gray-scale image (-i g) with a
gray-scale SE (-s g). This switch is ignored by other operations.
Switch -v tells the program to display some info as it computes. The
parameters following switch -k specify the structuring element (SE) in
one of 5 ways:
(1) 3x3 - specifies the structuring element to be a 3 by 3 square
of pixels. (2) plus - specifies the SE to be a 3 by 3 "+" shaped
set of pixels. (3) 5x5 - specifies the SE to be a 5x5 quasi
disk (square without corners) of pixels. (4) auto x y [z] - the
program makes an SE. The self-made SE is a disk with support
- 3 - Formatted: November 6, 2025
morph(1) morph(1)
Richard Alan Peters II Richard Alan Peters II
10 May 1993
covering x pixels horizontally and y pixels vertically. x and y
may be odd or even. (There is no restriction.) If specified, z is
the gray level of the center pixel. If z is not given, the level
defaults to BLACK (0). If auto is specified the SE type is set
to gray. (5) If the field following -k is not one of the above,
then the string, called SEFILE in the usage example, is taken as
the pathname of a structuring element file. If the user has an
environment variable called SEPATH, the program appends SEFILE to
it for the complete pathname.
If any of the program-generated structuring elments, 3x3, plus, 5x5,
or auto is requested, the structuring element type is set to grayscale
and the -s flag is ignored. Similarly, if 3x3, plus, or 5x5 is
requested the operation type is forced to be set processing and the -o
flag is ignored. If auto is requested, the operation type is taken as
set processing if option z is not specified or specified to be zero,
and the operation type is taken as function processing if a nonzero z
is specified. A structuring element file is an ASCII file of integers
separated by spaces. The first two numbers, x, y, are the horizontal
and vertical dimensions in pixels of the smallest rectangle that will
cover the structuring element. Both x and y must be > 0. The next two
numbers, i, j are the horizontal and vertical coordinates,
respectively, of the SE origin. IMPORTANT: The origin is expected to
be in the covering rectangle. The upper left hand corner of the
rectangle has coordinates (0,0); the lower right is (x-1,y-1).
Following the first four integers are x*y integers separated by spaces
and or returns. These numbers are the SE elements. Their
interpretation depends on the morphological operation being performed.
Negative SE elements are ALWAYS treated as logical DON'T CAREs. That
is, when the operation is in progress, image pixels under negative SE
elements are ignored. Thus, the support of the SE is limited to those
elments that are nonnegative. This permits the creation of odd-shaped
and multiply connected SE's, or an SE whose origin is outside its
active pixel area. If the input image is flagged binary, -i b, (i.e.
pixels grouped as zero and not zero), and the SE is flaged binary, -s
b, then the SE is used to perform a hit-or-miss transform. In this
case, zero SE elements cover the "miss" support and nonzero (positive)
elements cover the "hit" support. The gray-levels other than this are
ignored. If the input image is flagged -i b and the SE is flagged
gray (-s g) then the nonnegative (both zero and greater than zero) SE
elements determine the support of a "hit-only" transform. That is,
the nonnegative suport is used as a standard "set"-type SE for set
(binary) morphology. (Of course, the other gray-level info is
ignored.) Note: If the input is flagged -i b, then the flag [-o s|f]
is ignored. Here are the possible gray-level image flags and the
resulting interpretation of the SE elements:
-i g -s b -o s Function-set morphology on support of strictly
greater than zero SE elements. -i g -s b -o f Same as above. -i
g -s g -o s Function-set morphology on support of nonnegative
(greater than or equal to zero) SE elements. -i g -s g -o f
- 4 - Formatted: November 6, 2025
morph(1) morph(1)
Richard Alan Peters II Richard Alan Peters II
10 May 1993
Function-function morphology on support of nonnegative SE
elements. If one of the canned SE's (3x3, plus, or 5x5) is
chosen, the -s and -o flags are forced to be -s g and -o s. If
-k auto is selected, -s is forced to -s g. If along with -k
auto, the optional z value is set to a number greater than 0, -o
is forced to -o f; if z is either zero or left out, -o is forced
to -o s.
- 5 - Formatted: November 6, 2025
