GF2ARG - generates an ARG from a gf line set
NAME
gf2arg
The (edited) synopsis generated by the optionsGM library is:
Usage: gf2arg [options]
arg_1 (null) [null] file prefix
-a [0] ASCII o/p format
-p [0] Uses polar relations
-g [-1 -1] image size
-R [-1.000] region-of-interest size
-n [-1.000 -1.000] x & y s.d.s for
feature measurements
-x [-1] standard deviation of error in unknown x position
-y [-1] standard deviation of error in unknown y position
-r [-1] standard deviation of error in unknown rotation
-s [0] standard deviation of error in unknown scale
-help (1) [0] print out usage info
gf2arg
is a filter that takes as its input a set of points or line segments in gf
format (gf(5)), and generates an Attributed Relational Graph (ARG). Support of
the gf format is still limited - it can only handle points and lines, and does
not understand the GeomStruct statement. However it does distinguish between
oriented and non-oriented line segments (gf "orline" and "line" types
respectively). The (binary) format of the ARG file is no longer documented
because it keeps on changing. The functions that read and write it are put
together in "arg.cc", so it is relatively easy to maintain them in step with
each other.
If a file prefix is specified, the gf set is read from <prefix>.gf and the ARG
written to <prefix>.arg.
If no file prefix is specified,
stdin
and
stdout
are used for i/o.
<prefix>
- input file is <prefix>.gf; output file is <prefix>.arg. Default is to use
standard input and output.
-a
- generates an ASCII representation of the ARG, for diagnostic purposes. This format CANNOT be read by
the matcher program.
-c
- position of centre of node
j
with respect to node
i
is included in binary relations in cartesian form; default is to include them
in polar form. Cartesian for is recommended where line segments have bad
longditudinal alignment (due say to line breakage). Polar form is recommended
if allowance is being made for scaling errors.
-e
- creates angle values modulo 360 deg (default is 180 deg). for example, use
-e for edge detector o/p, default for ridge detector o/p.
-g <x_size> <y_size>
- allows the user to include the size of the original image in the ARG that the
features were extracted from. This information does not have to be that
accurate; an adequate guess can usually be found on the basis of the maximum
feature separation (which is used for the default).
-R <ROI_size>
- specifies feature separation beyond which relations will not be used. (-ve
value indicates no limit.)
-n <x_sd> <y_sd>
- causes covariance matrices to be generated for all possible relations.
<x_sd>, <y_sd>
specify standard deviations for feature measurements. In the case of line
segments, the x axis is assumed to be along the segment and the y axis
perpendicular to it.
These parameters are the standard deviations of the errors in the parameters
for the transformation between the scene and model spaces. Thus they indicate
how much is known about this transformation. The transformation may consists
of a translation, rotation and scaling. Each of these can be one of: known
(s.d. = 0), part-known (s.d. indicates range of likely values) or not known at
all (-ve value). A side effect of using these parameters (apart from -s) is
that they often significantly reduce the amount of memory and computation time
needed.
These parameters
must
correspond between scene and model args. I.e. if a parameter is part-known in
one it must be part-known in the other. Also, when the parameter(s) are
present, it is assumed that the scene and model have been registered to within
the available accuracy. A -ve value indicates that the parameter is not used.
-x <posn_range>, -y <posn_range>
- s.d. of relative translation error; default: not used
-r <rot_range>
- s.d. of relative orientation error (in degrees); default: not used for
segment features, = 0 for point features
-s <scale_range>
- s.d. of relative scale error (expressed in terms of scale factor); default: =
0
matcher(1), arg(5), gf(5).
Bill Christmas, University of Surrey, October 1994.
06-May-97. Automatically converted by man2html,
written by G.Matas (g.matas@ee.surrey.ac.uk)