Help for RNGERRS

PURPOSE: Determine relative accuracy of outputs from program rng.

EXECUTION:

  rng inp=(x,y) out=tran disparity=(dx,dy)
  rngerrs inp=(x,y) disparity=(dx,dy) pseudox=z diff=diff q=q nq=nq

The stereo pair (x,y) is input to rng, which computes the horizontal
and vertical disparities (dx,dy) of the right image y relative to the left
image x.  rngerrs uses these disparities to distort the right image into a
pseudo-image.  Differences between the left and pseudo-left image are then
measured by correlating over local areas (q) and differencing of the two
images (diff).

OPERATION:

The pseudo-left image z is computed as follows:

	z(i,j) = y(i+dx(i,j),j+dy(i,j))

where (i,j) are the (sample,line) image coordinates.  Since the coordinates
(i+dx(i,j),j+dy(i,j) is generally not at an integral pixel position, bilinear
interpolation is used to compute the value of y at that location.

The difference image diff is simply:

	diff(i,j) = x(i,j) = z(i,j)

The correlation quality image q is computed by matching a nlw x nsw area
surrounding each pixel location x(i,j) and z(i,j).  See NLW and NSW parameters.

	q(i,j) = correlation coefficient for area surrounding (i,j)

There are gaps in the disparity map (dx,dy).  All pixel locations where the
disparity was not measured is flagged as dx(i,j)=-999 and dy(i,j)=-999.
Whenever such a pixel is encountered during area matching, that pixel is
ignored during the match.  A count of all the valid pixels in the correlation
area is output to image nq:

	0 < nq(i,j) < nlw*nsw

If program rng has succeeded in computing the disparities, the pseudo-left
image should match the left image, except for differences in intensity
between the images.  The correlation quality is a measure of how well rng
has succeeded at each pixel location.  The mean value of q is computed and
printed for a specified area of the image (see AREA parameter).

HISTORY:

Written by: Gary Yagi, Feb 6, 2007
Cognizant programmer: Gary Yagi

PARAMETERS: