Help for LINEMTCH

PURPOSE
     LINEMTCH performs one-dimensional line-to-line correlation between two
images.  It is helpful for precise comparison or registration of two images.
It allows one of the images to have a different scale than the other or
to be offset from each other, but does not allow one image to be rotated 
with respect to the other.
It is similar in purpose to program PICMATCH, but is more special purpose.
It could have application in geometric calibration of cameras.
LINEMTCH requires that the geometric realtionship between the two images be 
approximately linear.  It computes the variation (OFFSET) between the images
and the linear model.  Thus it can perform precise (subpixel) measurement
of non-linear variations.

TAE COMMAND LINE FORMAT

     LINEMTCH INP=(A,B) INP=INT  PARAMS

     where
     A                   is the first image to be correlated.
     B                   is the second image to be correlated.
     INT                 is  the output IBIS interface  file 
                         that contains the pairs of matching 
                         sample   locations  together   with 
                         editing    information   such    as 
                         correlation value.
OPERATION

     First  the  program sets up a geometric model  relating 
     the two images using ITIE-OTIE parameters.  Since it is 
     assumed  that  the relationship between the geometry of
     the two images is approximately linear,   two 
     tiepoints should be enough to specify the  model.   The 
     model is used to estimate the a search locations and to 
     resample  the input image to match the sampling of  the 
     output  image.   Then  the  FFTs of  input  and  output 
     samplings  are computed and phase correlation  function 
     is obtained as an inverse FFT of the normalized product 
     of those two.   The maximum of the correlation function 
     should  yield the value of offset.   If the search area 
     is  larger than the sampling size,  program extracts  9 
     uniformly  spaced  samplings from the  search  area  of 
     input image and choses the offset value as maximum of 9 
     phase correlation function,  thus ruling out accidental 
     matches.   For each line pair, at the sample locations
     defined by the parameters MINS, MAXS, & SPACING, the offset
     (between the geometric model and the location of the
     cross-correlation maximum) and the correlation value
     are computed. The  final results are placed in columns  of 
     the IBIS interface file as follows:

Column:      1            2         3         4       5
Value:  line pair # input samp  output samp  offs  correlation
                                                        value
EXAMPLE

     linemtch INP=(A,B) OUT=INT ITIE=(1,400) OTIE=(20,730)           LINE=(1,1,5,5) MINS=100 MAXS=800 SPAC=100 POWER=5           AREA=128
RESTRICTIONS
     Input images must have byte format.
     Line size <= 60000 samples
     Power <= 10, area <= 1024
WRITTEN BY:                   B. Gokhman

COGNIZANT PROGRAMMER:         K. F. Evans

REVISION:                     New           March 4, 1983

  7-96  SP   Ported to UNIX.  Unfortunately, the only source file available
             appears as if it was not the final version.  It called a version
             of FFTT that had 4 arguments.  The current FFTT has 3 arguments.
             The phase correlation values are supposed to be normalized to
             1.0 (for an exact match) but the version I am delivering
             does not seem to give values greater than .5.  This program
             probably needs more debugging before it is used, which may
             be never.  The exact meaning of the offset column in the
             output file remains fuzzy.


PARAMETERS:


INP

Input images to correlate

OUT

Interface file for corr info

ITIE

Samp coord of tiepoints(input)

OTIE

Samp coord of tiepoints(output)

LINES

Line numbers to correlate

MINS

First sample for matching

MAXS

Last sample for matching

SPACING

Step in samples

AREA

Search area in samples

POWER

Size of sampling (power of 2)

NOPRINT

Suppress printout

See Examples:


Cognizant Programmer: