As an example of what can be achieved without "image processing".

This simple script compares two images and produces a single xx.xxx% similarity figure:

#! perl -slw
use strict;
use GD;

GD::Image->trueColor( 1 );

my $im1 = GD::Image->new( $ARGV[ 0 ] );
my $im2 = GD::Image->new( $ARGV[ 1 ] );

my $raw1 = $im1->gd;
my $raw2 = $im2->gd;

my $xored = $raw1 ^ $raw2;
my( $all, $diff ) = (0)x2;

$all += 255, $diff += ord substr $xored, $_, 1 for 0 .. length( $xored
+ ) - 1;

print $all, ' ', $diff;

printf "The simlarity is %.3f%%\n", ( $all - $diff ) / $all * 100;
[download]

And here are the results of applying that to my previous 'problem' examples.

• The 'spot-the-differences' ballons A & B:

  C:\test\xx>..\907337 ballons1.jpg ballons2.jpg
  672410265 62443267
  The simlarity is 90.714%
  [download]

• The 'identical except slight color fade' ballons; A & A':

  C:\test\xx>..\907337 ballons1.jpg ballons1a.jpg
  672410265 92011553
  The simlarity is 86.316%
  [download]

• The two BrowserUk late period variations of "Red rectangle on white background"; X & Y:

  C:\test\xx>..\907337 1.png 2.png
  244802805 241803
  The simlarity is 99.901%
  [download]

• And perhaps a more poignant example. MicroArray1 & MicroArray2:

  C:\test\xx>..\907337 microarray1.jpg microarray2.jpg
  340764405 41409302
  The simlarity is 87.848%
  [download]

  In this example, the second microarray is just a mirror image of the first for test purposes as I couldn't find two similarly size examples. The reflection means that the registration is probably not as good as you'd expect from a true comparison, but it serves its purpose to highlight the possibilities and downsides of the technique.

The technique could be much improved with good examples and a clearer statement of requirements.