23/02/2009 PFSTMO 1.4
	* mantiuk08: added color correction method from (http://zgk.wi.ps.pl/color_correction/)
	* mantiuk08: fixed dependency on libgsl (thanks to Niels)
	* durand02: added color correction method from (http://zgk.wi.ps.pl/color_correction/)
	
04/11/2008
	* all: variables with underscore replaced with "n<variable name>"
	as they caused compilation problems under cygwin
	
21/09/2008
	* mantiuk08: Refactored for exception safety

21/09/2008

	* build: Removed bogus flag -march=nocona; added CPU detection and
	--disable-portable option for enabling non-portable optimizations.
	Tunes for local CPU by default, can be disabled using
	--without-gcc-arch

21/09/2008

	* pfstmo.h: Templatized Array2D; added copy constructor,
	assignment operator, and function `allocate' for re-initializing
	array after construction

09/09/2008 PFSTMO 1.3.2

	* durand02: removed downsampling option - introduced large errors
	into the results
	
08/09/2008

	* durand02: division by zero bug fixed (fastbilateral)
	
	* durand02: range boundary bug fixed (fastbilateral)

	* durand02: added progress reporting (fastbilateral)

	* durand02: performance imrovements in fastbilateral
	
	* mantiuk08: found the cause for the uninitialized value in the cqpminimizer.
	Initialization moved to more a suitable code location
	
05/09/2008

	* durand02: disabled fast_bilateral after finding some bugs. A
	slow bilateral filter will be used until bugs are fixed.

04/09/2008

	* all: removed dependency on pfs::Array2D[Impl], replaced by new
	class pfstmo::Array2D.  Apart from getting rid of the dependency,
	this also speeds thing up considerably because pfstmo::Array2D is
	non-virtual and thus can be inlined.
	
03/09/2008

	* durand02: added missing delete in fastBilateralFilter. All TMOs
	are now valgrind memcheck-clean.

03/09/2008

	* mantiuk08: fixed uninitialized value in gsl_cqpminimizer_alloc.

02/09/2008

	* fattal02: replaced delete with correct delete[] operators in PDE
	solver.

27/08/2008

	* fixed: added missing FFTW3F_CFLAGS for durand02 operator 
	
	* pattanaik00: Fixed a bug that would cause every frame to be
	treated as the first frame, never invoking
	VisualAdaptationModel::calculateAdaptation.
	
27/08/2008

	* configure.ac: Added missing likely/unlikely macros

26/08/2008 PFSTMO 1.3.1

	* mantiuk06: Ed Brambley's bug-fix and convergence patch:

	 - There was a bug in the upsampling code that could cause an
	   out-of- bound read past the end of allocated memory.  This was
	   occasionally causing NaNs to creep in, at least in a few runs I
	   did.  This has been fixed.

         - The patch adds some code to monitor how we're converging to a
	   solution, and to reset the iteration if necessary.  In the few
	   examples I've tried, this has made the convergence much more
	   robust.

         - The patch changes the start point of the iteration.  Before, we
	   started from a uniform gray ("x" was initialized to 0.0f
	   everywhere). Now, we start from the original image.  I found
	   this also helps the iteration be more robust, although
	   sometimes its quicker and sometimes its slower this way.

         - The clipping prevention code has been modified to be scale
	   independent.  A badly-scaled input file should no longer cause
	   a loss of dynamic range.

         Overall, the code runs sometimes a little more slowly and
	 sometimes about the same speed as before, although it is, I
	 believe, now far more robust and much more likely to converge.

	more at: http://groups.google.com/group/pfstools/browse_thread/thread/de2378af98ec6185/0dee5304fc14e99d?hl=en#0dee5304fc14e99d

	* mantiuk06: Ed Brambley's OpenMP patch:

	I've attached a second patch (which depends on the first one).
	This patch is a first attempt at parallelizing the code to run on
	OpenMP. I've tried to avoid assuming how many processors there
	will be, so it should scale to 4 or 8 processors quite happily.
	I'll let you know when I get my Quad core set up :-)

        There are probably more efficient ways to parallelize the code;
	for example, having only one or two "#pragma omp parallel"
	sections and lots of "#pragma omp for" or "#pragma omp sections"
	sections, but I've not tried to do this.  It happily uses 100% of
	both cores on the dual-core box I tried it on, and gives knocks
	about 20 seconds of a typical 1 minute tone mapping.
	
	* mantiuk06: fixed references to exp10f (missing in some compilers)

	* mantiuk06: added branch prediction optimizations (thanks to Lebed Dmytry)

	* mantiuk08: better handling of zero-values in HDR images - previously it could fail on some images

	* mantiuk08: removed some debug code

	* mantiuk08: added branch prediction optimizations
	
	* added: autoconf checks for OpenMP and some gcc optimization options
	
16/06/2008 PFSTMO 1.3

	* added: mantiuk08 - display adaptive tone mapping

	* mantiuk06: updated documentation, added --quiet option,
	        can now process more than one image in a loop
	
19/02/2008 PFSTMO 1.2

	* mantiuk06: fixed some serious bugs in the solver and subsampling
	        procedure the artifacts found in some odd-sized images
		should be gone

	* mantiuk06: faster up-/down-sampling functions make the
	        algorithm 25% faster. Thanks to Ed Brambley.

	* mantiuk06: performance improvements from Ed Brambley: lower
	        memory consumption, faster solver, better convergence,
	        additional scale parameter for the contrast equalization
	        method.
		
11/07/2007 PFSTMO 1.1

	* pfstmo-1.1: first release on sourceforge
		
	* small fixes with parameters (Miguel Granados)
		
	* reinhard05: renamed from reinhard04 to correct publication year.
		substantial corrections to the code
		  
	* fattal02: major bug fix,
		added noise level parameter (Przemyslaw Bazarnik),
	  manpage updates - explanation of parameters
		  
	* mantiuk06: new gradient domain operator
		implementation by Radoslaw Mantiuk
		
	* ashikhmin02: implementation temporarily dropped
		imposible to reproduce results from paper

05/12/2005

	* added temporal coherence to reinhard02
	
29/11/2005

	* fixed ansi compatibilities (fattal02)

14/12/2004

	* new algorithms
		ashikhmin02, fattal02

	* durand02
		corrected interpretation of compression factor
		percentile to estimate min and max in final mapping -> more stable

	* pattanaik02
		bug fix: local adaptation
		bug fix: r,g,b intensities normalization

	* reinhard02
		fix: memory leaks in building the pyramid

22/09/2004 PFStmo 1.0
	
	* first unofficial release

	* implemented algorithms
		drago03, durand02, pattanaik00, reinhard02, reinhard04
