path: root/xicc/cam02.c

/* 
 * cam02
 *
 * Color Appearance Model, based on
 * CIECAM02, "The CIECAM02 Color Appearance Model"
 * by Nathan Moroney, Mark D. Fairchild, Robert W.G. Hunt, Changjun Li,
 * M. Ronnier Luo and Todd Newman, IS&T/SID Tenth Color Imaging
 * Conference, with the addition of the Viewing Flare
 * model described on page 487 of "Digital Color Management",
 * by Edward Giorgianni and Thomas Madden, and the
 * Helmholtz-Kohlrausch effect, using the equation from
 * the Bradford-Hunt 96C model as detailed in Mark Fairchild's
 * book "Color Appearance Models". 
 * The Slight modification to the Hunt-Pointer-Estevez matrix
 * recommended by Kuo, Zeis and Lai in IS&T/SID 14th Color Imaging
 * Conference "Robust CIECAM02 Implementation and Numerical
 * Experiment within an ICC Workflow", page 215, together with
 * their matrix formulation of inversion has been adopted.
 *
 * In addition the algorithm has unique extensions to allow
 * it to operate reasonably over an unbounded domain.
 *
 * Author:  Graeme W. Gill
 * Date:    17/1/2004
 * Version: 1.00
 *
 * This file is based on cam97s3.c by Graeme Gill.
 *
 * Copyright 2004 - 2011 Graeme W. Gill
 * Please refer to COPYRIGHT file for details.
 * This material is licenced under the GNU AFFERO GENERAL PUBLIC LICENSE Version 3 :-
 * see the License.txt file for licencing details.
 */


/* Note that XYZ values are normalised to 1.0 consistent */
/* with the ICC convention (not 100.0 as assumed by the CIECAM spec.) */
/* Note that all whites are assumed to be normalised (ie. Y = 1.0) */

/*
	TTBD: Should convert to using Timo Kunkel and Erik Reinhard's simplified
	and improved version of CIECAM02
	[ "A Neurophysiology-Inspired Steady-State Color Appearance Model",
	  ie. "CIECAM02-KR" ? ]

	The rgbp compression has it's problems in terms of perceptual
	uniformity. A color with one component near zero might shift
	all the components to -ve values on inverse conversion - ie.
	a 1 DE shift in Jab becomes a masive DE in XYZ/Lab/perceptual,
	with (say) a dark red becomong black because the blue
	value is small. One way around this is to re-introduce
	a flag to turn off perfect symetry by disabling
	expansion on the reverse conversion.

 */

/*
	Various additions and changes have been made to allow the CAM
	conversions to and from an unbounded range of XYZ and Jab values,
	in a (somewhat) geometrically consistent maner. This is because
	such values arise in the process of gamut mapping, and in scanning through
	the grid of PCS values needed to fill in the A2B table of an
	ICC profile. Such values have no correlation to a real color
	value, but none the less need to be handled without causing an
	exception, in a geometrically consistent and reversible
	fashion.

	The changes are:

	The Hunt-Pointer-Estevez matrix has been increased in precission by
	1 digit [Kuo, Zeise & Lai, IS&T/SID 14th Color Imaging Conference pp. 215-219]

	The sharpened cone response matrix third row has been changed from
	[0.0030, 0.0136, 0.9834] to [0.0000, 0.0000, 1.0000]
	[Susstrunk & Brill, IS&T/SID 14th Color Imaging Conference Late Breaking News sublement]

	To prevent wild Jab values as the rgb' values approach zero, a region close to each
	primary ground plane is compressed. This expands the region of reasonable
	behaviour to encompass XYZ values that are found in practice (ie. ProPhotoRGB).

	To prevent excessive curvature of hue in high saturation blue regions
	due to the non-linearity exagerating the difference between
	r & b values, a heuristic is introduced that blends the r & b
	values, reducing this effect. This degrades the agreement
	with the reference CIECAM implementation by about 1DE in this region,
	but greatly improves the behaviour in mapping and clipping from
	highly saturated blues (ie. ProPhotoRGB).

	The post-adaptation non-linearity response has had a straight
	line negative linear extension added.
	The positive region has a tangential linear extension added, so
	that the range of values is not limited.
	An adaptive threshold for the low level linear extension,
	to avoid coordinates blowing up when one of the cone responses
	is large, while the others become negative.
		
	Re-arrange ss equation into separated effects,
	k1, k2, k3, and then limit these to avoid divide by zero
	and sign change errors in fwd and bwd converson, as well
	as avoiding the blue saturation doubling back for low
	luminance levels.

	To avoid chroma and hue angle information being lost when the
	J value used to scale the chroma is 0, and to ensure
	that J = 0, a,b != 0 values have a valid mapping into
	XYZ space, the J value used to multiply Chroma, is limited
	to be equivalent to not less than A == 0.1.

	The Helmholtz-Kohlrausch effect is crafted to have resonable
	effects over the range of J from 0 to 100, and have more
	moderate effects outside this range. 

*/

#include <copyright.h>
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#include "icc.h"
#include "xcam.h"
#include "cam02.h"
#include "numlib.h"

#define ENABLE_COMPR		/* [Def] Enable XYZ compression  */
#undef ENABLE_DECOMPR		/* [Undef] Enable XYZ de-compression  */
#define ENABLE_BLUE_ANGLE_FIX	/* [Def] Limit maximum blue angle */
#define ENABLE_DDL			/* [Def] Enable k1,k2,k3 overall ss limit values (seems to be the best scheme) */
#undef ENABLE_SS			/* [Undef] Disable overall ss limit values (not the scheme used) */

#undef ENTRACE				/* [Undef] Enable internal value runtime tracing if s->trace != 0 */
#undef DOTRACE				/* [Undef] Trace anyway (ie. set s->trace = 1) */
#undef DIAG1				/* [Undef] Print internal value diagnostics for conditions setup */
#undef DIAG2				/* [Undef] Print internal value diagnostics for each conversion */
#undef TRACKMINMAX			/* [Undef] Track min/max DD & SS limits (run with locus cam02test) */
#undef DISABLE_MATRIX		/* Debug - disable matrix & non-lin, wire XYZ to rgba */
#undef DISABLE_SCR			/* Debug - disable Sharpened Cone Response matrix */
#undef DISABLE_HPE			/* Debug - disable just Hunt-Pointer_Estevez matrix */
#undef DISABLE_NONLIN		/* Debug - wire rgbp to rgba */
#undef DISABLE_TTD			/* Debug - disable ttd vector 'tilt' */
#undef DISABLE_HHKR			/* Debug - disable Helmholtz-Kohlrausch */

							/* We reduce the HK effect from the Hunt equation, in the */
							/* light of real wold experiments. */
#define HHKR_MUL 0.25		/* [0.25] - Helmholtz-Kohlrausch strength multiplier */

#ifdef ENABLE_COMPR
# define BC_WHMINY 0.2		/* [0.2] Compression direction minimum Y value */
# define BC_RANGE_R 0.01	/* [0.02] Set comp. range as prop. of distance to neutral - red */
# define BC_RANGE_G 0.01	/* [0.02] Set comp. range as prop. of distance to neutral - green*/
# define BC_RANGE_B 0.01	/* [0.02] Set comp. range as prop. of distance to neutral - blue */
# define BC_MAXRANGE 0.13	/* [0.13] Maximum compression range */
# define BC_LIMIT 0.7		/* [0.7] Correction limit (abs. rgbp distance shift) */
#endif /* ENABLE_COMPR */

#ifdef ENABLE_BLUE_ANGLE_FIX
# define BLUE_BL_MAX 0.9	/* [0.9] Sets ultimate blue angle, higher = less angle */
# define BLUE_BL_POW 3.5	/* [3.5] Sets rate at which blue angle is limited */
#endif

#define NLDLIMIT 0.00001	/* [0.00001] Non-linearity minimum lower crossover to straight line */
#define NLDICEPT -0.18		/* [-0.18] Input intercept of straight line with 0.1 output */

#define NLULIMIT 1e5		/* Non-linearity upper crossover to straight line */

#ifdef ENABLE_SS			/* [Undef] */
# define SSLLIMIT 0.22		/* Overall ab scale lower limit */
# define SSULIMIT 580.0      /* Overall ab scale upper limit */
#endif /* ENABLE_SS */

#define SYMETRICJ			/* [Undef] Use symetric power about zero, else straigt line -ve */

#define DDLLIMIT 0.55		/* [0.55] ab component -k3:k2 ratio limit (must be < 1.0) */
//#define DDULIMIT 0.9993		/* [0.9993] ab component k3:k1 ratio limit (must be < 1.0) */
#define DDULIMIT 0.34		/* [0.34] ab component k3:k1 ratio limit (must be < 1.0) */
#define SSMINcJ 0.005		/* [0.005] ab scale cJ minimum value */
#define JLIMIT 0.005		/* [0.005] J encoding cutover point straight line (0 - 1.0 range) */
#define HKLIMIT 0.7			/* [0.7] Maximum Helmholtz-Kohlrausch lift out of 1.0 */

#ifdef TRACKMINMAX
double minss = 1e60;
double maxss = -1e60;
double minlrat = 0.0;
double maxurat = 0.0;
#define noslots 103
double slotsd[noslots];
double slotsu[noslots];
double minj = 1e38, maxj = -1e38;
#endif /* TRACKMINMAX */

#if defined(ENTRACE) || defined(DOTRACE)
#define TRACE(xxxx) if (s->trace) printf xxxx ;
#else
#define TRACE(xxxx) 
#endif

static void cam_free(cam02 *s);
static int set_view(struct _cam02 *s, ViewingCondition Ev, double Wxyz[3],
	                double La, double Yb, double Lv, double Yf, double Yg, double Gxyz[3],
					int hk, double hkscale, double mtaf, double Wxyz2[3]);
static int XYZ_to_cam(struct _cam02 *s, double *Jab, double *xyz);
static int cam_to_XYZ(struct _cam02 *s, double *xyz, double *Jab);

static double spow(double val, double pp) {
	if (val < 0.0)
		return -pow(-val, pp);
	else
		return pow(val, pp);
}

/* Create a cam02 conversion object, with default viewing conditions */
cam02 *new_cam02(void) {
	cam02 *s;
//	double D50[3] = { 0.9642, 1.0000, 0.8249 };

	if ((s = (cam02 *)calloc(1, sizeof(cam02))) == NULL) {
		fprintf(stderr,"cam02: malloc failed allocating object\n");
		exit(-1);
	}
	
	/* Initialise methods */
	s->del      = cam_free;
	s->set_view = set_view;
	s->XYZ_to_cam = XYZ_to_cam;
	s->cam_to_XYZ = cam_to_XYZ;

	/* Initialise default parameters */
	s->hkscale = 1.0;

	/* Set default range handling limits */
	s->nldlimit = NLDLIMIT;
	s->nldicept = NLDICEPT;
	s->nlulimit = NLULIMIT;
	s->ddllimit = DDLLIMIT;
	s->ddulimit = DDULIMIT;
	s->ssmincj = SSMINcJ;
	s->jlimit = JLIMIT;
	s->hklimit = 1.0 / HKLIMIT;

	/* Set a default viewing condition ?? */
	/* set_view(s, vc_average, D50, 33, 0.2, 0.0, 0.0, D50, 0); */

#ifdef DOTRACE
	s->trace = 1;
#endif

#ifdef TRACKMINMAX
	{
		int i;
		for (i = 0; i < noslots; i++) {
			slotsd[i] = 0.0;
			slotsu[i] = 0.0;
		}
	}
#endif /* TRACKMINMAX */

	return s;
}

static void cam_free(cam02 *s) {

#ifdef TRACKMINMAX
	{
		int i;
		for (i = 0; i < noslots; i++) {
			printf("Slot %d = %f, %f\n",i,slotsd[i], slotsu[i]);
		}
		printf("minj = %f, maxj = %f\n",minj,maxj);
	
		printf("minss = %f\n",minss);
		printf("maxss = %f\n",maxss);
		printf("minlrat = %f\n",minlrat);
		printf("maxurat = %f\n",maxurat);
	}
#endif /* TRACKMINMAX */

	if (s != NULL)
		free(s);
}

/* Return value is always 0 */
static int set_view(
cam02 *s,
ViewingCondition Ev,	/* Enumerated Viewing Condition */
double Wxyz[3],	/* Reference/Adapted White XYZ (Y range 0.0 .. 1.0) */
double La,		/* Adapting/Surround Luminance cd/m^2 */
double Yb,		/* Relative Luminance of Background to reference white (range 0.0 .. 1.0) */
double Lv,		/* Luminance of white in the Viewing/Scene/Image field (cd/m^2) */
				/* Ignored if Ev is set to other than vc_none */
double Yf,		/* Flare as a fraction of the reference white (Y range 0.0 .. 1.0) */
double Yg,		/* Flare as a fraction of the adapting/surround (Y range 0.0 .. 1.0) */
double Gxyz[3],	/* The Glare white coordinates (typically the Ambient color) */
				/* If <= 0 will Wxyz will be used. */
int hk,			/* Flag, NZ to use Helmholtz-Kohlrausch effect */
double hkscale,	/* HK effect scaling factor */
double mtaf,	/* Mid tone partial adapation factor from Wxyz to Wxyz2, <= 0.0 if none */
double Wxyz2[3] /* Mid tone Adapted White XYZ (Y range 0.0 .. 1.0) */
) {
	double tt, t1, t2;
	double tm[3][3];
	int i;

	if (Ev == vc_none) {
		/* Compute the internal parameters from the */
		/* ratio of La to Lv by interpolation */
		int i;
		double r, bf;
		/* Dark, dim, average, above average */
		double t_C[4]  = { 0.525, 0.59, 0.69, 1.0 }; 
		double t_Nc[4] = { 0.800, 0.95, 1.00, 1.0 }; 
		double t_F[4]  = { 0.800, 0.90, 1.00, 1.0 }; 

		if (La < 1e-10) 		/* Hmm. */
			La = 1e-10;
		r = La/Lv;
		if (r < 0.0)
			r = 0.0;
		else if (r > 1.0)
			r = 1.0;
	
		if (r < 0.1) {			/* Dark to Dim */
			i = 0;
			bf = r/0.1;
		} else if (r < 0.2) {	/* Dim to Average */
			i = 1;
			bf = (r - 0.1)/0.1;
		} else {				/* Average to above average */
			i = 2;
			bf = (r - 0.2)/0.8;
		}
		s->C  = t_C[i] * (1.0 - bf)  + t_C[i+1] * bf;
		s->Nc = t_Nc[i] * (1.0 - bf) + t_Nc[i+1] * bf;
		s->F  = t_F[i] * (1.0 - bf)  + t_F[i+1] * bf;
	} else {
		/* Compute the internal parameters by category */
		/* Fake up Lv according to category */
		switch(Ev) {
			case vc_dark:
				s->C = 0.525;
				s->Nc = 0.8;
				s->F = 0.8;
				Lv = La/0.033; 
				break;
			case vc_dim:
				s->C = 0.59;
				s->Nc = 0.95;
				s->F = 0.9;
				Lv = La/0.1; 
				break;
			case vc_average:
			default:
				s->C = 0.69;
				s->Nc = 1.0;
				s->F = 1.0;
				Lv = La/0.2; 
				break;
			case vc_cut_sheet:
				s->C = 0.41;
				s->Nc = 0.8;
				s->F = 0.8;
				Lv = La/0.02; 	// ???
				break;
		}
	}

	/* Transfer parameters to the object */
	s->Ev = Ev;
	s->Wxyz[0] = Wxyz[0];
	s->Wxyz[1] = Wxyz[1];
	s->Wxyz[2] = Wxyz[2];
	s->La = La;
	s->Yb = Yb > 0.005 ? Yb : 0.005;	/* Set minimum to avoid divide by 0.0 */
	s->Lv = Lv;
	s->Yf = Yf;
	s->Yg = Yg;
	if (Gxyz[0] > 0.0 && Gxyz[1] > 0.0 && Gxyz[2] > 0.0) {
		tt = Wxyz[1]/Gxyz[1];		/* Scale to white ref white */
		s->Gxyz[0] = tt * Gxyz[0];
		s->Gxyz[1] = tt * Gxyz[1];
		s->Gxyz[2] = tt * Gxyz[2];
	} else {
		s->Gxyz[0] = Wxyz[0];
		s->Gxyz[1] = Wxyz[1];
		s->Gxyz[2] = Wxyz[2];
	}
	s->hk = hk;
	s->hkscale = hkscale;
	if (mtaf > 0.0 && Wxyz2 != NULL) {		/* Mid tone Adaption hack enabled */
		if (mtaf > 1.0)
			mtaf = 1.0;
		s->mtaf = mtaf;
		s->Wxyz2[0] = Wxyz2[0]/Wxyz2[1] * Wxyz[1];		/* Normalize scale */
		s->Wxyz2[2] = Wxyz2[2]/Wxyz2[1] * Wxyz[1];
		s->Wxyz2[1] =                     Wxyz[1];
		s->mtap = (4.0 * mtaf);		/* 0.5 -> pow 4 */
		s->mtap *= s->mtaf;
		s->pmta_en = 1;
	} else {
		s->mtaf = mtaf;
		s->Wxyz2[0] = Wxyz[0];
		s->Wxyz2[1] = Wxyz[1];
		s->Wxyz2[2] = Wxyz[2];
		s->mtap = 1.0;
		s->pmta_en = 0;
	}

	/* The rgba vectors */
	s->Va[0] = 1.0;
	s->Va[1] = -12.0/11.0;
	s->Va[2] = 1.0/11.0;

	s->Vb[0] = 1.0/9.0;
	s->Vb[1] = 1.0/9.0;
	s->Vb[2] = -2.0/9.0;

	s->VttA[0] = 2.0;
	s->VttA[1] = 1.0;
	s->VttA[2] = 1.0/20.0;

	/* (Not used) */
	s->Vttd[0] = 1.0;
	s->Vttd[1] = 1.0;
	s->Vttd[2] = 21.0/20.0;

	/* Vttd in terms of the VttA, Va and Vb vectors */
	s->dcomp[0] = 1.0;
	s->dcomp[1] = -11.0/23.0;
	s->dcomp[2] = -108.0/23.0;

	/* Compute values that only change with viewing parameters */

	/* Figure out the Flare contribution to the flareless XYZ input */
	s->Fsxyz[0] = s->Yf * s->Wxyz[0];
	s->Fsxyz[1] = s->Yf * s->Wxyz[1];
	s->Fsxyz[2] = s->Yf * s->Wxyz[2];

	/* Add in the Glare contribution from the adapting/surround */
	tt = s->Yg * s->La/s->Lv;
	s->Fsxyz[0] += tt * s->Gxyz[0];
	s->Fsxyz[1] += tt * s->Gxyz[1];
	s->Fsxyz[2] += tt * s->Gxyz[2];

	/* Rescale so that the sum of the flare and the input doesn't exceed white */
	s->Fsc = s->Wxyz[1]/(s->Fsxyz[1] + s->Wxyz[1]);
	s->Fsxyz[0] *= s->Fsc;
	s->Fsxyz[1] *= s->Fsc;
	s->Fsxyz[2] *= s->Fsc;
	s->Fisc = 1.0/s->Fsc;

#ifndef DISABLE_SCR
	/* Sharpened cone response white values */
	s->rgbW[0] =  0.7328 * s->Wxyz[0] + 0.4296 * s->Wxyz[1] - 0.1624 * s->Wxyz[2];
	s->rgbW[1] = -0.7036 * s->Wxyz[0] + 1.6975 * s->Wxyz[1] + 0.0061 * s->Wxyz[2];
	s->rgbW[2] =  0.0000 * s->Wxyz[0] + 0.0000 * s->Wxyz[1] + 1.0000 * s->Wxyz[2];
	s->rgbW2[0] =  0.7328 * s->Wxyz2[0] + 0.4296 * s->Wxyz2[1] - 0.1624 * s->Wxyz2[2];
	s->rgbW2[1] = -0.7036 * s->Wxyz2[0] + 1.6975 * s->Wxyz2[1] + 0.0061 * s->Wxyz2[2];
	s->rgbW2[2] =  0.0000 * s->Wxyz2[0] + 0.0000 * s->Wxyz2[1] + 1.0000 * s->Wxyz2[2];
#else
	s->rgbW[0] = s->Wxyz[0];
	s->rgbW[1] = s->Wxyz[1];
	s->rgbW[2] = s->Wxyz[2];
	s->rgbW2[0] = s->Wxyz2[0];
	s->rgbW2[1] = s->Wxyz2[1];
	s->rgbW2[2] = s->Wxyz2[2];
#endif

	/* Degree of chromatic adaptation */
	s->D = s->F * (1.0 - exp((-s->La - 42.0)/92.0)/3.6);

	/* Precompute Chromatic transform values */
	s->Drgb[0] = s->D * (s->Wxyz[1]/s->rgbW[0]) + 1.0 - s->D;
	s->Drgb[1] = s->D * (s->Wxyz[1]/s->rgbW[1]) + 1.0 - s->D;
	s->Drgb[2] = s->D * (s->Wxyz[1]/s->rgbW[2]) + 1.0 - s->D;
	s->Drgb2[0] = s->D * (s->Wxyz2[1]/s->rgbW2[0]) + 1.0 - s->D;
	s->Drgb2[1] = s->D * (s->Wxyz2[1]/s->rgbW2[1]) + 1.0 - s->D;
	s->Drgb2[2] = s->D * (s->Wxyz2[1]/s->rgbW2[2]) + 1.0 - s->D;

	/* Chromaticaly transformed sharpened white value */
	s->rgbcW[0] = s->Drgb[0] * s->rgbW[0];
	s->rgbcW[1] = s->Drgb[1] * s->rgbW[1];
	s->rgbcW[2] = s->Drgb[2] * s->rgbW[2];
	s->rgbcW2[0] = s->Drgb2[0] * s->rgbW2[0];
	s->rgbcW2[1] = s->Drgb2[1] * s->rgbW2[1];
	s->rgbcW2[2] = s->Drgb2[2] * s->rgbW2[2];
	
#ifndef DISABLE_HPE
	/* Transform from spectrally sharpened, to Hunt-Pointer_Estevez cone space */
	s->rgbpW[0] =  0.7409744840453773 * s->rgbcW[0]
	            +  0.2180245944753982 * s->rgbcW[1]
	            +  0.0410009214792244 * s->rgbcW[2];
	s->rgbpW[1] =  0.2853532916858801 * s->rgbcW[0]
	            +  0.6242015741188157 * s->rgbcW[1]
	            +  0.0904451341953042 * s->rgbcW[2];
	s->rgbpW[2] = -0.0096276087384294 * s->rgbcW[0]
	            -  0.0056980312161134 * s->rgbcW[1]
	            +  1.0153256399545427 * s->rgbcW[2];
	s->rgbpW2[0] =  0.7409744840453773 * s->rgbcW2[0]
	             +  0.2180245944753982 * s->rgbcW2[1]
	             +  0.0410009214792244 * s->rgbcW2[2];
	s->rgbpW2[1] =  0.2853532916858801 * s->rgbcW2[0]
	             +  0.6242015741188157 * s->rgbcW2[1]
	             +  0.0904451341953042 * s->rgbcW2[2];
	s->rgbpW2[2] = -0.0096276087384294 * s->rgbcW2[0]
	             -  0.0056980312161134 * s->rgbcW2[1]
	             +  1.0153256399545427 * s->rgbcW2[2];
#else
	s->rgbpW[0] = s->rgbcW[0];
	s->rgbpW[1] = s->rgbcW[1];
	s->rgbpW[2] = s->rgbcW[2];
	s->rgbpW2[0] = s->rgbcW2[0];
	s->rgbpW2[1] = s->rgbcW2[1];
	s->rgbpW2[2] = s->rgbcW2[2];
#endif /* DISABLE_HPE */

#ifndef DISABLE_SCR
	/* Create combined cone and chromatic transform matrix: */
	/* Spectrally sharpened cone responses matrix */
	s->cc[0][0] =  0.7328; s->cc[0][1] = 0.4296; s->cc[0][2] = -0.1624;
	s->cc[1][0] = -0.7036; s->cc[1][1] = 1.6975; s->cc[1][2] =  0.0061;
	s->cc[2][0] =  0.0000; s->cc[2][1] = 0.0000; s->cc[2][2] =  1.0000;
#else
	s->cc[0][0] = 1.0; s->cc[0][1] = 0.0; s->cc[0][2] = 0.0;
	s->cc[1][0] = 0.0; s->cc[1][1] = 1.0; s->cc[1][2] = 0.0;
	s->cc[2][0] = 0.0; s->cc[2][1] = 0.0; s->cc[2][2] = 1.0;
#endif
	icmCpy3x3(s->cc2, s->cc);
	
	/* Chromaticaly transformed sample value */
	icmSetUnity3x3(tm);
	tm[0][0] = s->Drgb[0];
	tm[1][1] = s->Drgb[1];
	tm[2][2] = s->Drgb[2];
	icmMul3x3(s->cc, tm);
	
	tm[0][0] = s->Drgb2[0];
	tm[1][1] = s->Drgb2[1];
	tm[2][2] = s->Drgb2[2];
	icmMul3x3(s->cc2, tm);
	
#ifndef DISABLE_HPE
	/* Transform from spectrally sharpened, to Hunt-Pointer_Estevez cone space */
	tm[0][0] =  0.7409744840453773;
	tm[0][1] =  0.2180245944753982;
	tm[0][2] =  0.0410009214792244;
	tm[1][0] =  0.2853532916858801;
	tm[1][1] =  0.6242015741188157;
	tm[1][2] =  0.0904451341953042;
	tm[2][0] = -0.0096276087384294;
	tm[2][1] = -0.0056980312161134;
	tm[2][2] =  1.0153256399545427;
	icmMul3x3(s->cc, tm);
	icmMul3x3(s->cc2, tm);
#endif /* DISABLE_HPE */

	/* Create inverse combined cone and chromatic transform matrix: */
	icmInverse3x3(s->icc, s->cc);		/* Hmm. we assume it cannot fail */
	icmInverse3x3(s->icc2, s->cc2);

#ifdef ENABLE_COMPR
	/* Compression ranges */
	s->crange[0] = BC_RANGE_R; 
	s->crange[1] = BC_RANGE_G; 
	s->crange[2] = BC_RANGE_B; 
#endif /* ENABLE_COMPR */

	/* Background induction factor */
	s->n = s->Yb/ s->Wxyz[1];
	s->nn = pow(1.64 - pow(0.29, s->n), 0.73);	/* Pre computed value */

	/* Lightness contrast factor ?? */
	{
		double k;

		k = 1.0 / (5.0 * s->La + 1.0);
		s->Fl = 0.2 * pow(k , 4.0) * 5.0 * s->La
		      + 0.1 * pow(1.0 - pow(k , 4.0) , 2.0) * pow(5.0 * s->La , 1.0/3.0);
	}

	/* Background and Chromatic brightness induction factors */
	s->Nbb   = 0.725 * pow(1.0/s->n, 0.2);
	s->Ncb   = s->Nbb;

	/* Base exponential nonlinearity */
	s->z = 1.48 + pow(s->n , 0.5);

	/* Post-adapted cone response of white */
	tt = pow(s->Fl * s->rgbpW[0], 0.42);
	s->rgbaW[0] = 400.0 * tt / (tt + 27.13) + 0.1;
	tt = pow(s->Fl * s->rgbpW[1], 0.42);
	s->rgbaW[1] = 400.0 * tt / (tt + 27.13) + 0.1;
	tt = pow(s->Fl * s->rgbpW[2], 0.42);
	s->rgbaW[2] = 400.0 * tt / (tt + 27.13) + 0.1;

	/* Achromatic response of white */
	s->Aw = (s->VttA[0] * s->rgbaW[0]
	      +  s->VttA[1] * s->rgbaW[1]
	      +  s->VttA[2] * s->rgbaW[2] - 0.305) * s->Nbb;

	/* Non-linearity lower crossover output value */
	tt = pow(s->Fl * s->nldlimit, 0.42);
	s->nldxval = 400.0 * tt / (tt + 27.13) + 0.1;

	/* Non-linearity lower crossover slope from lower crossover */
	/* to intercept with 0.1 output */
	s->nldxslope = (s->nldxval - 0.1)/(s->nldlimit - s->nldicept);

	/* Non-linearity upper crossover value */
	tt = pow(s->Fl * s->nlulimit, 0.42);
	s->nluxval = 400.0 * tt / (tt + 27.13) + 0.1;

	/* Non-linearity upper crossover slope, set to asymtope */
	t1 = s->Fl * s->nlulimit;
	t2 = pow(t1, 0.42) + 27.13;
	s->nluxslope = 0.42 * s->Fl * 400.0 * 27.13 / (pow(t1,0.58) * t2 * t2);


	/* Limited A value at J = JLIMIT */
	s->lA = pow(s->jlimit, 1.0/(s->C * s->z)) * s->Aw;

#ifdef DIAG1
	printf("Scene parameters:\n");
	printf("Viewing condition Ev = %d\n",s->Ev);
	printf("Ref white Wxyz = %f %f %f\n", s->Wxyz[0], s->Wxyz[1], s->Wxyz[2]);
	printf("Relative luminance of background Yb = %f\n", s->Yb);
	printf("Adapting luminance La = %f\n", s->La);
	printf("Flare Yf = %f\n", s->Yf);
	printf("Glare Yg = %f\n", s->Yg);
	printf("Glare color Gxyz = %f %f %f\n", s->Gxyz[0], s->Gxyz[1], s->Gxyz[2]);

	printf("Internal parameters:\n");
	printf("Surround Impact C = %f\n", s->C);
	printf("Chromatic Induction Nc = %f\n", s->Nc);
	printf("Adaptation Degree F = %f\n", s->F);

	printf("Pre-computed values\n");
	printf("Sharpened cone white rgbW = %f %f %f\n", s->rgbW[0], s->rgbW[1], s->rgbW[2]);
	printf("Degree of chromatic adaptation D = %f\n", s->D);
	printf("Chromatic transform values Drgb = %f %f %f\n", s->Drgb[0], s->Drgb[1], s->Drgb[2]);
	printf("Chromatically transformed white rgbcW = %f %f %f\n", s->rgbcW[0], s->rgbcW[1], s->rgbcW[2]);
	printf("Hunter-P-E cone response white rgbpW = %f %f %f\n", s->rgbpW[0], s->rgbpW[1], s->rgbpW[2]);
	printf("Background induction factor n = %f\n", s->n);
	printf("                            nn = %f\n", s->nn);
	printf("Lightness contrast factor Fl = %f\n", s->Fl);
	printf("Background brightness induction factor Nbb = %f\n", s->Nbb);
	printf("Chromatic brightness induction factor Ncb = %f\n", s->Ncb);
	printf("Base exponential nonlinearity z = %f\n", s->z);
	printf("Post adapted cone response white rgbaW = %f %f %f\n", s->rgbaW[0], s->rgbaW[1], s->rgbaW[2]);
	printf("Achromatic response of white Aw = %f\n", s->Aw);
	if (s->pmta_en) {
		printf("Mid tone partial adapation factor = %f\n", s->mtaf);
		printf("Mid tone Adapted White Wxyz2 = %f %f %f\n", s->Wxyz2[0], s->Wxyz2[1], s->Wxyz2[2]);
		printf("Mid tone partial adapation power = %f\n", s->mtap);
	}
	printf("\n");
#endif
	return 0;
}

/* Conversions. Return values are always 0 */
static int XYZ_to_cam(
struct _cam02 *s,
double Jab[3],
double XYZ[3]
) {
	int i;
	double xyz[3], rgbp[3], rgba[3];
	double a, b, ja, jb, J, JJ, C, h, e, A, ss;
	double ttA, rS, cJ, tt;
	double k1, k2, k3;
	int clip = 0;

#ifdef DIAG2		/* Incase in == out */
	double XYZi[3];

	XYZi[0] = XYZ[0];
	XYZi[1] = XYZ[1];
	XYZi[2] = XYZ[2];
#endif

	TRACE(("\nCIECAM02 Forward conversion:\n"))
	TRACE(("XYZ = %f %f %f\n",XYZ[0], XYZ[1], XYZ[2]))

#ifdef DISABLE_MATRIX

	rgba[0] = XYZ[0];
	rgba[1] = XYZ[1];
	rgba[2] = XYZ[2];

#else /* !DISABLE_MATRIX */

	/* Add in flare */
	xyz[0] = s->Fsc * XYZ[0] + s->Fsxyz[0];
	xyz[1] = s->Fsc * XYZ[1] + s->Fsxyz[1];
	xyz[2] = s->Fsc * XYZ[2] + s->Fsxyz[2];

	TRACE(("XYZ inc flare = %f %f %f\n",xyz[0], xyz[1], xyz[2]))

	/* Transfor to spectrally sharpened cone responses, */
	/* apply chromaticaly transform, */
	/* and transform from spectrally sharpened to Hunt-Pointer_Estevez cone space, */
	/* all in one step. */
	icmMulBy3x3(rgbp, s->cc, xyz);

	/* Partial mid-tone adapation hack. */
	/* Blend between two adapation states along Y */
	if (s->pmta_en) {
		double rgbp2[3], t;

//printf("\nIn xyz %f %f %f\n",xyz[0], xyz[1], xyz[2]);
		icmMulBy3x3(rgbp2, s->cc2, xyz);
		t = xyz[1]/s->Wxyz[1];
		if (t < 0.0)
			t = 0.0;
		else if (t > 1.0)
			t = 1.0;
		t = pow(t, s->mtap);
//printf("Blend %f from %f %f %f and %f %f %f\n",t, rgbp2[0], rgbp2[1], rgbp2[2], rgbp[0], rgbp[1], rgbp[2]);
		icmBlend3(rgbp, rgbp2, rgbp, t);
	}

	TRACE(("rgbp = %f %f %f\n", rgbp[0], rgbp[1], rgbp[2]))

#ifdef ENABLE_COMPR
	/* Try and prevent crazy out of cam02 gamut behaviour, by compressing */
	/* the rgbp so as to prevent it becoming less than zero. */
	{
		double tt;			/* Temporary */
		double wrgb[3];		/* White target */

		/* Make white target white point with same Y value */
		tt = xyz[1] > BC_WHMINY ? xyz[1] : BC_WHMINY;	/* Limit to minimum Y */
		icmScale3(wrgb, s->rgbpW, tt/s->Wxyz[1]);	/* White target at same Y */
		TRACE(("wrgb %f %f %f\n", wrgb[0], wrgb[1], wrgb[2]))

		/* Compress r,g,b in turn */
		for (i = 0; i < 3; i++) {
			double cv;		/* Compression value */
			double ctv;		/* Compression target value */
			double cd;		/* Compression displacement needed */
			double cvec[3];	/* Normalized correction vector */
			double isec[3];	/* Intersection with plane */
			double offs;	/* Offset of point from orgin*/
			double range;	/* Threshold to start compression */
			double asym;	/* Compression asymtope */

			/* Compute compression direction as vector towards white */
			/* (We did try correcting in a blend of limit plane normal and white, */
			/*  but compressing towards white seems to be the best.) */
			icmSub3(cvec, wrgb, rgbp);					/* Direction of white target */

			TRACE(("ch %d, rgbp %f %f %f\n", i, rgbp[0], rgbp[1], rgbp[2]))
			TRACE(("cvec %f %f %f\n", cvec[0], cvec[1], cvec[2]))

			if (cvec[i] < 1e-9) {		/* compression direction can't correct this coord */
				TRACE(("Intersection with limit plane failed\n"))
				continue;
			}

			/* Scale compression vector to make it move a unit in normal direction */
			icmScale3(cvec, cvec, 1.0/cvec[i]);		/* Normalized vector to white */
			TRACE(("ncvec %f %f %f\n", cvec[0], cvec[1], cvec[2]))

			/* Compute intersection of correction direction with this limit plane */
			/* (This corresponds with finding displacement of rgbp by cvec */
			/*  such that the current coord value = 0) */
			icmScale3(isec, cvec, -rgbp[i]);		/* (since cvec[i] == 1.0) */
			icmAdd3(isec, isec, rgbp);
			TRACE(("isec %f %f %f\n", isec[0], isec[1], isec[2]))

			/* Compute distance from intersection to origin */
			offs = pow(icmNorm3(isec), 0.85);

			range = s->crange[i] * offs;	/* Scale range by distance to origin */
			if (range > BC_MAXRANGE)		/* so that it tapers down as we approach it */
				range = BC_MAXRANGE;		/* and limit maximum */

			/* Aiming above plane when far from origin, */
			/* but below plane at the origin, so that black isn't affected. */
			asym = range - 0.2 * (range + (0.01 * s->crange[i]));

			ctv = cv = rgbp[i];		/* Distance above/below limit plane */

			TRACE(("ch %d, offs %f, range %f asym %f, cv %f\n",i, offs,range,asym,cv))
			if (cv < (range - 1e-12)) {		/* Need to compress */
				double aa, bb;
				aa = 1.0/(range - cv);
				bb = 1.0/(range - asym);
				ctv = range - 1.0/(aa + bb);

				cd = ctv - cv;				/* Displacement needed */
				if (cd > BC_LIMIT)
					cd = BC_LIMIT;
				TRACE(("ch %d cd = %f, scaled cd %f\n",i,cd,cd))

				/* Apply correction */
				icmScale3(cvec, cvec, cd);			/* Compression vector */
				icmAdd3(rgbp, rgbp, cvec);			/* Compress by displacement */
				TRACE(("rgbp after comp. = %f %f %f\n",rgbp[0], rgbp[1], rgbp[2]))
			}
		}
	}
#endif /* ENABLE_COMPR */

#ifdef ENABLE_BLUE_ANGLE_FIX
	ss = rgbp[0] + rgbp[1] + rgbp[2];
	if (ss < 1e-9) {
		ss = 0.0;
	} else {
		ss = (rgbp[2]/ss - 1.0/3.0) * 3.0/2.0;
		if (ss > 0.0)
			ss = BLUE_BL_MAX * pow(ss, BLUE_BL_POW);
	}
	if (ss < 0.0)
		ss = 0.0;
	else if (ss > 1.0)
		ss = 1.0;
	tt = 0.5 * (rgbp[0] + rgbp[1]);
	rgbp[0] = ss * tt + (1.0 - ss) * rgbp[0];
	rgbp[1] = ss * tt + (1.0 - ss) * rgbp[1];
	TRACE(("rgbp after blue fix ss = %f, rgbp = %f %f %f\n",ss,rgbp[0], rgbp[1], rgbp[2]))
#endif

#ifdef DISABLE_NONLIN
	for (i = 0; i < 3; i++) {
		rgba[i] = 400.0/27.13 * rgbp[i];
	}
#else	/* !DISABLE_NONLIN */
	/* Post-adapted cone response of sample. */
	/* rgba[] has a minimum value of 0.1 for XYZ[] = 0 and no flare. */
	/* We add a negative linear region, plus a linear segment at */
	/* the end of the +ve conversion to allow numerical handling of a */
	/* very wide range of values. */

	for (i = 0; i < 3; i++) {
		if (rgbp[i] < s->nldlimit) {
			rgba[i] = s->nldxval + s->nldxslope * (rgbp[i] - s->nldlimit);
		} else {
			if (rgbp[i] <= s->nlulimit) {
				tt = pow(s->Fl * rgbp[i], 0.42);
				rgba[i] = 400.0 * tt / (tt + 27.13) + 0.1;
			} else {
				rgba[i] = s->nluxval + s->nluxslope * (rgbp[i] - s->nlulimit);
			}
		}
	}
#endif	/* !DISABLE_NONLIN */

//tt = 0.5 * (rgba[0] + rgba[1]);
//rgba[0] = (rgba[0] - ss * tt)/(1.0 - ss);
//rgba[1] = (rgba[1] - ss * tt)/(1.0 - ss);

#endif /* !DISABLE_MATRIX */

	/* Note that the minimum values of rgba[] for XYZ = 0 is 0.1, */
	/* hence magic 0.305 below comes from the following weighting of rgba[], */
	/* to base A at 0.0 */
	/* Deal with the core rgb to A, S & h conversion: */

	TRACE(("rgba = %f %f %f\n", rgba[0], rgba[1], rgba[2]))

	/* Preliminary Acromatic response */
	ttA = s->VttA[0] * rgba[0] + s->VttA[1] * rgba[1] + s->VttA[2] * rgba[2];

	/* Achromatic response */
	A = (ttA - 0.305) * s->Nbb;

	/* Preliminary red-green & yellow-blue opponent dimensions */
	/* a, b & ttd form an (almost) orthogonal coordinate set. */
	/* ttA is in a different direction */
	a = s->Va[0] * rgba[0] + s->Va[1] * rgba[1] + s->Va[2] * rgba[2];
	b = s->Vb[0] * rgba[0] + s->Vb[1] * rgba[1] + s->Vb[2] * rgba[2];

	/* restricted Saturation to stop divide by zero */
	/* (The exact value isn't important because the numerator dominates as a,b aproach 0 */
	rS = sqrt(a * a + b * b);
	if (rS < DBL_EPSILON)
		rS = DBL_EPSILON;
	TRACE(("ttA = %f, a = %f, b = %f, rS = %f, A = %f\n", ttA,a,b,rS,A))

	/* Lightness J, Derived directly from Acromatic response. */
	/* Cuttover to a straight line segment when J < 0.005, */
#ifndef SYMETRICJ		/* Cut to a straight line */
	if (A >= s->lA) {
		J = pow(A/s->Aw, s->C * s->z);		/* J/100  - keep Sign */
	} else {
		J = s->jlimit/s->lA * A;			/* Straight line */
		TRACE(("limited Acromatic to straight line\n"))
	}
#else			/* Symetric */
	if (A >= 0.0) {
		J = pow(A/s->Aw, s->C * s->z);		/* J/100  - keep Sign */
	} else {
		J = -pow(-A/s->Aw, s->C * s->z);		/* J/100  - keep Sign */
		TRACE(("symetric Acromatic\n"))
	}
#endif

	/* Constrained (+ve, non-zero) J */
	if (A > 0.0) {
		cJ = pow(A/s->Aw, s->C * s->z);
		if (cJ < s->ssmincj)
			cJ = s->ssmincj;
	} else {
		cJ = s->ssmincj;
	}

	TRACE(("J = %f, cJ = %f\n",J,cJ))

	/* Final hue angle */
	h = (180.0/DBL_PI) * atan2(b,a);
	h = (h < 0.0) ? h + 360.0 : h;

	/* Eccentricity factor */
	e = (12500.0/13.0 * s->Nc * s->Ncb * (cos(h * DBL_PI/180.0 + 2.0) + 3.8));

	/* ab scale components */
	k1 = pow(s->nn, 1.0/0.9) * e * pow(cJ, 1.0/1.8)/pow(rS, 1.0/9.0);
	k2 = pow(cJ, 1.0/(s->C * s->z)) * s->Aw/s->Nbb + 0.305;
	k3 = s->dcomp[1] * a + s->dcomp[2] * b;

	TRACE(("Raw k1 = %f, k2 = %f, k3 = %f, raw ss = %f\n",k1, k2, k3, pow(k1/(k2 + k3), 0.9)))

#ifdef TRACKMINMAX
	{
		int sno;
		double lrat, urat;

		ss = pow(k1/(k2 + k3), 0.9);

		if (ss < minss)
			minss = ss;
		if (ss > maxss)
			maxss = ss;

		lrat = -k3/k2;
		urat =  k3 * pow(ss, 10.0/9.0) / k1; 
		if (lrat > minlrat)
			minlrat = lrat;
		if (urat > maxurat)
			maxurat = urat;

		/* Record distribution of ss min/max vs. J for */
		/* regions outside a,b == 0 */

		sno = (int)(J * 100.0 + 0.5);
		if (sno < 0) {
			sno = 101;
			if (J < minj)
				minj = J;
		} else if (sno > 100) {
			sno = 102;
			if (J > maxj)
				maxj = J;
		}
		if (slotsd[sno] < lrat)
			slotsd[sno] = lrat;
		if (slotsu[sno] < urat)
			slotsu[sno] = urat;
	}
#endif /* TRACKMINMAX */

#ifdef ENABLE_DDL	

	/* Limit ratio of k3 to k2 to stop zero or -ve ss */
	if (k3 < -k2 * s->ddllimit) {
		k3 = -k2 * s->ddllimit;
		TRACE(("k3 limited to %f due to k3:k2 ratio, ss = %f\n",k3,pow(k1/(k2 + k3), 0.9)))
		clip = 1;
	}

	/* See if there is going to be a problem in bwd, and limit k3 if there is */
	if (k3 > (k2 * s->ddulimit/(1.0 - s->ddulimit))) {
		k3 = (k2 * s->ddulimit/(1.0 - s->ddulimit));
		TRACE(("k3 limited to %f to allow for bk3:bk1 bwd limit\n",k3))
		clip = 1;
	}

#endif /* !ENABLE_DDL */	

#ifdef DISABLE_TTD 

	ss = pow((k1/k2), 0.9);

#else	/* !TRACKMINMAX */

	/* Compute the ab scale factor */
	ss = pow(k1/(k2 + k3), 0.9);

#endif	/* !ENABLE_DDL */

#ifdef ENABLE_SS
	if (ss < SSLLIMIT)
		ss = SSLLIMIT;
	else if (ss > SSULIMIT)
		ss = SSULIMIT;
#endif /* ENABLE_SS */

#ifdef NEVER
// -------------------------------------------
	/* Show ss components */
	if (s->retss) {
		Jab[0] = 1.0;
		if (clip)
			Jab[0] = 0.0;
		Jab[1] = (log10(ss) - +1.5)/(2.0 - +1.5);
		Jab[2] = (log10(ss) - +1.0)/(2.5 - +1.0);

		for (i = 0; i < 3; i++) {
			if (Jab[i] < 0.0)
				Jab[i] = 0.0;
			else if (Jab[i] > 1.0)
				Jab[i] = 1.0;
		}
		return 0;
	}
// -------------------------------------------
#endif /* NEVER */

	ja = a * ss;
	jb = b * ss;

	/* Chroma - always +ve, used just for HHKR */
	C = sqrt(ja * ja + jb * jb);
	
	TRACE(("ss = %f, A = %f, J = %f, C = %f, h = %f\n",ss,A,J,C,h))

	JJ = J;
#ifndef DISABLE_HHKR
 	/* Helmholtz-Kohlrausch effect */
	if (s->hk && J < 1.0) {
//		double kk = C/300.0 * sin(DBL_PI * fabs(0.5 * (h - 90.0))/180.0);
		double kk = s->hkscale * HHKR_MUL * C/300.0 * sin(DBL_PI * fabs(0.5 * (h - 90.0))/180.0);
		if (kk > 1e-6) 	/* Limit kk to a reasonable range */
			kk = 1.0/(s->hklimit + 1.0/kk);
		JJ = J + (1.0 - (J > 0.0 ? J : 0.0)) * kk;
		TRACE(("JJ = %f from J = %f, kk = %f\n",JJ,J,kk))
	}
#endif /* DISABLE_HHKR */

	JJ *= 100.0;	/* Scale J */

	/* Compute Jab value */
	Jab[0] = JJ;
	Jab[1] = ja;
	Jab[2] = jb;

#ifdef NEVER	/* Brightness/Colorfulness */
	{
		double M, Q;

		Q = (4.0/s->C) * sqrt(JJ/100.0) * (s->Aw + 4.0) * pow(s->Fl, 0.25);
		M = C * pow(s->Fl, 0.25);

		printf("Lightness = %f, Chroma = %f\n",J,C);
		printf("Brightness = %f, Colorfulness = %f\n",M,Q);
	}
#endif /* NEVER */

	TRACE(("Jab %f %f %f\n",Jab[0], Jab[1], Jab[2]))
	TRACE(("\n"))

#ifdef DIAG2
	printf("Processing XYZ->Jab:\n");
	printf("XYZ = %f %f %f\n", XYZi[0], XYZi[1], XYZi[2]);
	printf("Including flare XYZ = %f %f %f\n", xyz[0], xyz[1], xyz[2]);
	printf("Huntpace rgbpP-E cone space rgbp = %f %f %f\n", rgbp[0], rgbp[1], rgbp[2]);
	printf("Post adapted cone response rgba = %f %f %f\n", rgba[0], rgba[1], rgba[2]);
	printf("Prelim red green a = %f, b = %f\n", a, b);
	printf("Hue angle h = %f\n", h);
	printf("Eccentricity factor e = %f\n", e);
	printf("Achromatic response A = %f\n", A);
	printf("Lightness J = %f, H.K. Lightness = %f\n", J * 100, JJ);
	printf("Prelim Saturation ss = %f\n", ss);
	printf("Chroma C = %f\n", C);
	printf("Jab = %f %f %f\n", Jab[0], Jab[1], Jab[2]);
	printf("\n");
#endif
	return 0;
}

static int cam_to_XYZ(
struct _cam02 *s,
double XYZ[3],
double Jab[3]
) {
	int i;
	double xyz[3], rgbp[3], rgba[3];
	double a, b, ja, jb, J, JJ, C, rC, h, e, A, ss;
	double tt, cJ, ttA;
	double k1, k2, k3;		/* (k1 & k3 are different to the fwd k1 & k3) */

#ifdef DIAG2		/* Incase in == out */
	double Jabi[3];

	Jabi[0] = Jab[0];
	Jabi[1] = Jab[1];
	Jabi[2] = Jab[2];

#endif

	TRACE(("\nCIECAM02 Reverse conversion:\n"))
	TRACE(("Jab %f %f %f\n",Jab[0], Jab[1], Jab[2]))

	JJ = Jab[0] * 0.01;	/* J/100 */
	ja = Jab[1];
	jb = Jab[2];

	/* Convert Jab to core A, S & h values: */

	/* Compute hue angle */
	h = (180.0/DBL_PI) * atan2(jb, ja);
	h = (h < 0.0) ? h + 360.0 : h;
	
	/* Compute chroma value */
	C = sqrt(ja * ja + jb * jb);	/* Must be Always +ve, Can be NZ even if J == 0 */

	/* Preliminary Restricted chroma, always +ve and NZ */
	/* (The exact value isn't important because the numerator dominates as a,b aproach 0) */
	rC = C;
	if (rC < DBL_EPSILON)
		rC = DBL_EPSILON;

	J = JJ;
#ifndef DISABLE_HHKR
 	/* Undo Helmholtz-Kohlrausch effect */
	if (s->hk && J < 1.0) {
//		double kk = C/300.0 * sin(DBL_PI * fabs(0.5 * (h - 90.0))/180.0);
		double kk = s->hkscale * HHKR_MUL * C/300.0 * sin(DBL_PI * fabs(0.5 * (h - 90.0))/180.0);
		if (kk > 1e-6) 	/* Limit kk to a reasonable range */
			kk = 1.0/(s->hklimit + 1.0/kk);
		J = (JJ - kk)/(1.0 - kk);
		if (J < 0.0)
			J = JJ - kk;
		TRACE(("J = %f from JJ = %f, kk = %f\n",J,JJ,kk))
	}
#endif /* DISABLE_HHKR */

	/* Achromatic response */
#ifndef SYMETRICJ		/* Cut to a straight line */
	if (J >= s->jlimit) {
		A = pow(J, 1.0/(s->C * s->z)) * s->Aw;
	} else {	/* In the straight line segment */
		A = s->lA/s->jlimit * J;
		TRACE(("Undo Acromatic straight line\n"))
	}
#else			/* Symetric */
	if (J >= 0.0) {
		A = pow(J, 1.0/(s->C * s->z)) * s->Aw;
	} else {	/* In the straight line segment */
		A = -pow(-J, 1.0/(s->C * s->z)) * s->Aw;
		TRACE(("Undo symetric Acromatic\n"))
	}
#endif

	/* Preliminary Acromatic response +ve */ 
	ttA = (A/s->Nbb)+0.305;

	if (A > 0.0) {
		cJ = pow(A/s->Aw, s->C * s->z);
		if (cJ < s->ssmincj)
			cJ = s->ssmincj;
	} else {
		cJ = s->ssmincj;
	}
	TRACE(("C = %f, A = %f from J = %f, cJ = %f\n",C, A,J,cJ))

	/* Eccentricity factor */
	e = (12500.0/13.0 * s->Nc * s->Ncb * (cos(h * DBL_PI/180.0 + 2.0) + 3.8));

	/* ab scale components */
	k1 = pow(s->nn, 1.0/0.9) * e * pow(cJ, 1.0/1.8)/pow(rC, 1.0/9.0);
	k2 = pow(cJ, 1.0/(s->C * s->z)) * s->Aw/s->Nbb + 0.305;
	k3 = s->dcomp[1] * ja + s->dcomp[2] * jb;

	TRACE(("Raw k1 = %f, k2 = %f, k3 = %f, raw ss = %f\n",k1, k2, k3, (k1 - k3)/k2))

#ifdef ENABLE_DDL 

	/* Limit ratio of k3 to k1 to stop zero or -ve ss */
	if (k3 > (k1 * s->ddulimit)) {
		k3 = k1 * s->ddulimit;
		TRACE(("k3 limited to %f due to k3:k1 ratio, ss = %f\n",k3,(k1 - k3)/k2))
	}

	/* See if there is going to be a problem in fwd */
	if (k3 < -k1 * s->ddllimit/(1.0 - s->ddllimit)) {
		/* Adjust ss to allow for fwd limitd computation */
		k3 = -k1 * s->ddllimit/(1.0 - s->ddllimit);
		TRACE(("k3 set to %f to allow for fk3:fk2 fwd limit\n",k3))
	}

#endif /* ENABLE_DDL */	

#ifdef DISABLE_TTD 

	ss = k1/k2;

#else /* !DISABLE_TTD */

	/* Compute the ab scale factor */
	ss = (k1 - k3)/k2;

#endif /* !DISABLE_TTD */

#ifdef ENABLE_SS
	if (ss < SSLLIMIT)
		ss = SSLLIMIT;
	else if (ss > SSULIMIT)
		ss = SSULIMIT;
#endif /* ENABLE_SS */

	/* Unscale a and b */
	a = ja / ss;
	b = jb / ss;

	TRACE(("ss = %f, ttA = %f, a = %f, b = %f\n",ss,ttA,a,b))

	/* Solve for post-adapted cone response of sample */
	/* using inverse matrix on ttA, a, b */
	rgba[0] = (20.0/61.0) * ttA
	        + ((41.0 * 11.0)/(61.0 * 23.0)) * a
	        + ((288.0 * 1.0)/(61.0 * 23.0)) * b;
	rgba[1] = (20.0/61.0) * ttA
	        - ((81.0 * 11.0)/(61.0 * 23.0)) * a
	        - ((261.0 * 1.0)/(61.0 * 23.0)) * b;
	rgba[2] = (20.0/61.0) * ttA
	        - ((20.0 * 11.0)/(61.0 * 23.0)) * a
	        - ((20.0 * 315.0)/(61.0 * 23.0)) * b;

	TRACE(("rgba %f %f %f\n",rgba[0], rgba[1], rgba[2]))
	
#ifdef DISABLE_MATRIX

	XYZ[0] = rgba[0];
	XYZ[1] = rgba[1];
	XYZ[2] = rgba[2];

#else /* !DISABLE_MATRIX */

#ifdef DISABLE_NONLIN
	rgbp[0] = 27.13/400.0 * rgba[0];
	rgbp[1] = 27.13/400.0 * rgba[1];
	rgbp[2] = 27.13/400.0 * rgba[2];
#else	/* !DISABLE_NONLIN */

	/* Hunt-Pointer_Estevez cone space */
	/* (with linear segment at the +ve end) */
	for (i = 0; i < 3; i++) {
		if (rgba[i] < s->nldxval) {
			rgbp[i] = s->nldlimit + (rgba[i] - s->nldxval)/s->nldxslope;
		} else if (rgba[i] <= s->nluxval) {
			tt = rgba[i] - 0.1;
			rgbp[i] = pow((27.13 * tt)/(400.0 - tt), 1.0/0.42)/s->Fl;
//			rgbp[i] = pow((27.13 * tt)/(400.0 - tt), 1.0/1.0)/s->Fl;
		} else {
			rgbp[i] = s->nlulimit + (rgba[i] - s->nluxval)/s->nluxslope;
		}
	}
#endif /* !DISABLE_NONLIN */

	TRACE(("rgbp %f %f %f\n",rgbp[0], rgbp[1], rgbp[2]))

#ifdef ENABLE_BLUE_ANGLE_FIX
	ss = rgbp[0] + rgbp[1] + rgbp[2];
	if (ss < 1e-9)
		ss = 0.0;
	else {
		ss = (rgbp[2]/ss - 1.0/3.0) * 3.0/2.0;
		if (ss > 0.0)
			ss = BLUE_BL_MAX * pow(ss, BLUE_BL_POW);
	}
	if (ss < 0.0)
		ss = 0.0;
	else if (ss > 1.0)
		ss = 1.0;
	tt = 0.5 * (rgbp[0] + rgbp[1]);
	rgbp[0] = (rgbp[0] - ss * tt)/(1.0 - ss);
	rgbp[1] = (rgbp[1] - ss * tt)/(1.0 - ss);

	TRACE(("rgbp after blue fix %f = %f %f %f\n",ss,rgbp[0], rgbp[1], rgbp[2]))
#endif


#ifdef ENABLE_DECOMPR
	/* Undo soft limiting */
	{
		double tt;			/* Temporary */
		double wrgb[3];		/* White target */

		/* Make white target white point with same Y value */
		tt = rgbp[0] * s->icc[1][0] + rgbp[1] * s->icc[1][1] + rgbp[2] * s->icc[1][2];
		tt = tt > BC_WHMINY ? tt : BC_WHMINY;	/* Limit to minimum Y */
		icmScale3(wrgb, s->rgbpW, tt/s->Wxyz[1]);	/* White target at same Y */
		TRACE(("wrgb %f %f %f\n", wrgb[0], wrgb[1], wrgb[2]))

		/* Un-limit b,g,r in turn */
		for (i = 2; i >= 0; i--) {
			double cv;		/* Compression value */
			double ctv;		/* Compression target value */
			double cd;		/* Compression displacement needed */
			double cvec[3];	/* Normalized correction vector */
			double isec[3];	/* Intersection with plane */
			double offs;	/* Offset of point from orgin*/
			double range;	/* Threshold to start compression */
			double asym;	/* Compression asymtope */

			/* Compute compression direction as vector towards white */
			/* (We did try correcting in a blend of limit plane normal and white, */
			/*  but compressing towards white seems to be the best.) */
			icmSub3(cvec, wrgb, rgbp);					/* Direction of white target */

			TRACE(("ch %d, rgbp %f %f %f\n", i, rgbp[0], rgbp[1], rgbp[2]))
			TRACE(("cvec %f %f %f\n", cvec[0], cvec[1], cvec[2]))

			if (cvec[i] < 1e-9) {		/* compression direction can't correct this coord */
				TRACE(("Intersection with limit plane failed\n"))
				continue;
			}

			/* Scale compression vector to make it move a unit in normal direction */
			icmScale3(cvec, cvec, 1.0/cvec[i]);		/* Normalized vector to white */
			TRACE(("cvec %f %f %f\n", cvec[0], cvec[1], cvec[2]))

			/* Compute intersection of correction direction with this limit plane */
			/* (This corresponds with finding displacement of rgbp by cvec */
			/*  such that the current coord value = 0) */
			icmScale3(isec, cvec, -rgbp[i]);		/* (since cvec[i] == 1.0) */
			icmAdd3(isec, isec, rgbp);
			TRACE(("isec %f %f %f\n", isec[0], isec[1], isec[2]))

			/* Compute distance from intersection to origin */
			offs = pow(icmNorm3(isec), 0.85);

			range = s->crange[i] * offs;	/* Scale range by distance to origin */
			if (range > BC_MAXRANGE)		/* so that it tapers down as we approach it */
				range = BC_MAXRANGE;		/* and limit maximum */

			/* Aiming above plane when far from origin, */
			/* but below plane at the origin, so that black isn't affected. */
			asym = range - 0.2 * (range + (0.01 * s->crange[i]));

			ctv = cv = rgbp[i];		/* Distance above/below limit plane */

			TRACE(("ch %d, offs %f, range %f asym %f, cv %f\n",i, offs,range,asym,cv))

			if (ctv < (range - 1e-12)) {		/* Need to expand */

				if (ctv <= asym) {
					cd = BC_LIMIT;
					TRACE(("ctv %f < asym %f\n",ctv,asym))
				} else {
					double aa, bb;
					aa = 1.0/(range - ctv);
					bb = 1.0/(range - asym);
					if (aa > (bb + 1e-12))
						cv = range - 1.0/(aa - bb);
					cd = ctv - cv;				/* Displacement needed */
				}
				if (cd > BC_LIMIT)
					cd = BC_LIMIT;
				TRACE(("ch %d cd = %f, scaled cd %f\n",i,cd,cd))

				if (cd > 1e-9) {
					icmScale3(cvec, cvec, -cd);			/* Compression vector */
					icmAdd3(rgbp, rgbp, cvec);			/* Compress by displacement */
					TRACE(("rgbp after decomp. = %f %f %f\n",rgbp[0], rgbp[1], rgbp[2]))
				}
			}
		}
	}
#endif /* ENABLE_COMPR */

	/* Chromaticaly transformed sample value */
	/* Spectrally sharpened cone responses */
	/* XYZ values */
	if (!s->pmta_en) {
		icmMulBy3x3(xyz, s->icc, rgbp);

	/* Partial mid-tone adapation hack. */
	} else {
		double xyz1[3], xyz2[3], t;
		int i;

		/* Aprox bwd transform to compute starting value */
		icmMulBy3x3(xyz1, s->icc,  rgbp);
		icmMulBy3x3(xyz2, s->icc2, rgbp);
		t = 0.5 * (xyz1[1] + xyz2[1])/s->Wxyz[1];
		if (t < 0.0)
			t = 0.0;
		else if (t > 1.0)
			t = 1.0;
		t = pow(t, s->mtap);
		icmBlend3(xyz, xyz2, xyz1, t);

		/* Simple Newton itteration to more accurately invert */
		for (i = 0; i < 5; i++) {
			double rgbp0[3], rgbp1[3], rgbp2[3];
			double rgbd[3], xyzd0[3], xyzd1[3], xyzd2[3];

			/* Fwd transform */
			icmMulBy3x3(rgbp1, s->cc, xyz);
			icmMulBy3x3(rgbp2, s->cc2, xyz);
			t = xyz[1]/s->Wxyz[1];
			if (t < 0.0)
				t = 0.0;
			else if (t > 1.0)
				t = 1.0;
			t = pow(t, s->mtap);
			icmBlend3(rgbp0, rgbp2, rgbp1, t);

			icmSub3(rgbd, rgbp, rgbp0);			/* Error to input value */

			/* Bwd transform of error */
			icmMulBy3x3(xyzd1, s->icc,  rgbd);
			icmMulBy3x3(xyzd2, s->icc2, rgbd);
			icmBlend3(xyzd0, xyzd2, xyzd1, t);

			/* Adjust solution for error */
			icmAdd3(xyz, xyz, xyzd0);
		}
	}

	TRACE(("XYZ = %f %f %f\n",xyz[0], xyz[1], xyz[2]))

	/* Subtract flare */
	XYZ[0] = s->Fisc * (xyz[0] - s->Fsxyz[0]);
	XYZ[1] = s->Fisc * (xyz[1] - s->Fsxyz[1]);
	XYZ[2] = s->Fisc * (xyz[2] - s->Fsxyz[2]);

#endif /* !DISABLE_MATRIX */

	TRACE(("XYZ after flare = %f %f %f\n",XYZ[0], XYZ[1], XYZ[2]))
	TRACE(("\n"))

#ifdef DIAG2
	printf("Processing:\n");
	printf("Jab = %f %f %f\n", Jabi[0], Jabi[1], Jabi[2]);
	printf("Chroma C = %f\n", C);
	printf("Preliminary Saturation ss = %f\n", ss);
	printf("Lightness J = %f, H.K. Lightness = %f\n", J * 100, JJ * 100);
	printf("Achromatic response A = %f\n", A);
	printf("Eccentricity factor e = %f\n", e);
	printf("Hue angle h = %f\n", h);
	printf("Post adapted cone response rgba = %f %f %f\n", rgba[0], rgba[1], rgba[2]);
	printf("Hunundeft-P-E cone space rgbp = %f %f %f\n", rgbp[0], rgbp[1], rgbp[2]);
	printf("Including flare XYZ = %f %f %f\n", xyz[0], xyz[1], xyz[2]);
	printf("XYZ = %f %f %f\n", XYZ[0], XYZ[1], XYZ[2]);
	printf("\n");
#endif
	return 0;
}