[Octave-cvsupdate] SF.net SVN: octave:[10925] trunk/octave-forge/main/signal

[<car...@us...>]

Revision: 10925
          http://octave.svn.sourceforge.net/octave/?rev=10925&view=rev
Author:   carandraug
Date:     2012-08-29 02:45:12 +0000 (Wed, 29 Aug 2012)
Log Message:
-----------
xcorr2: remove old coeff option code, replace by new method

Modified Paths:
--------------
    trunk/octave-forge/main/signal/NEWS
    trunk/octave-forge/main/signal/inst/xcorr2.m

Modified: trunk/octave-forge/main/signal/NEWS
===================================================================
--- trunk/octave-forge/main/signal/NEWS	2012-08-27 13:28:45 UTC (rev 10924)
+++ trunk/octave-forge/main/signal/NEWS	2012-08-29 02:45:12 UTC (rev 10925)
@@ -18,8 +18,8 @@
  ** Bugfix in `xcorr2' introduced in 1.1.2 that would not accept "none" as
     scale option.
 
- ** `xcorr2' accepts a new scale option, "norm", to return the normalized
-    cross correlation.
+ ** `xcorr2' scaling option "coeff" was changed to return the normalized
+    cross-correlation.
 
  ** The following functions are new:
     movingrms  schtrig  clustersegment

Modified: trunk/octave-forge/main/signal/inst/xcorr2.m
===================================================================
--- trunk/octave-forge/main/signal/inst/xcorr2.m	2012-08-27 13:28:45 UTC (rev 10924)
+++ trunk/octave-forge/main/signal/inst/xcorr2.m	2012-08-29 02:45:12 UTC (rev 10925)
@@ -41,10 +41,6 @@
 ## element.
 ##
 ## @item "coeff"
-## Normalizes the sequence dividing by the max of the cross-correlation, so that
-## the largest cross-correlation element is 1.
-##
-## @item "norm"
 ## Returns the normalized cross-correlation.
 ## @end table
 ##
@@ -75,7 +71,7 @@
 
   ## bias routines by Dave Cogdell (cog...@as...)
   ## optimized by Paul Kienzle (pki...@us...)
-  ## norm routine by Carnë Draug (car...@gm...)
+  ## coeff routine by Carnë Draug (car...@gm...)
   switch lower (biasflag)
     case {"none"}
       ## do nothing, it's all done
@@ -94,15 +90,10 @@
       c    = c./bias;
 
     case {"coeff"}
-      c = c/max(c(:))';
-
-    case {"norm"}
-      ## FIXME maybe these conversions for double can be removed when
-      ## https://savannah.gnu.org/bugs/?37199 gets fixed?
       a = double (a);
       b = double (b);
       a = conv2 (a.^2, ones (size (b)));
-      b = dot (b(:), b(:));
+      b = sumsq (b(:));
       c(:,:) = c(:,:) ./ sqrt (a(:,:) * b);
 
     otherwise
@@ -110,7 +101,7 @@
   endswitch
 endfunction
 
-%!test
+%!test  # basic usage
 %! a = magic (5);
 %! b = [6 13 22; 10 18 23; 8 15 23];
 %! c = [391  807  519  391  473  289 120
@@ -121,3 +112,22 @@
 %!      473 1006 1643 1457  946  347 108
 %!      242  539  850  477  374  129  54];
 %! assert (xcorr2 (a, b), c);
+
+%!shared a, b, c, row_shift, col_shift
+%! row_shift = 18;
+%! col_shift = 20;
+%! a = randi (255, 30, 30);
+%! b = a(row_shift-10:row_shift, col_shift-7:col_shift);
+%! c = xcorr2 (a, b, "coeff");
+%!assert (nthargout ([1 2], @find, c == max (c(:))), {row_shift, col_shift});   # should return exact coordinates
+%! m = rand (size (b)) > 0.5;
+%! b(m)  = b(m)  * 0.95;
+%! b(!m) = b(!m) * 1.05;
+%! c = xcorr2 (a, b, "coeff");
+%!assert (nthargout ([1 2], @find, c == max (c(:))), {row_shift, col_shift});   # even with some small noise, should return exact coordinates
+
+%!test  # coeff of autocorrelation must be same as negavtive of correlation by additive inverse
+%! a = 10 * randn (100, 100);
+%! auto    = xcorr2 (a, "coeff");
+%! add_in  = xcorr2 (a, -a, "coeff");
+%! assert ([min(auto(:)), max(auto(:))], -[max(add_in(:)), min(add_in(:))]);

This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.