src/px/neuclrtab.c

#ifndef lint
static const char       RCSid[] = "$Id: neuclrtab.c,v 2.9 2003/02/22 02:07:27 greg Exp $";
#endif
/*
 * Neural-Net quantization algorithm based on work of Anthony Dekker
 */

#include "copyright.h"

#include <string.h>

#include "standard.h"
#include "color.h"
#include "random.h"

#ifdef COMPAT_MODE
#define neu_init        new_histo
#define neu_pixel       cnt_pixel
#define neu_colrs       cnt_colrs
#define neu_clrtab      new_clrtab
#define neu_map_pixel   map_pixel
#define neu_map_colrs   map_colrs
#define neu_dith_colrs  dith_colrs
#endif
                                /* our color table (global) */
extern BYTE     clrtab[256][3];
static int      clrtabsiz;

#ifndef DEFSMPFAC
#ifdef SPEED
#define DEFSMPFAC       (240/SPEED+3)
#else
#define DEFSMPFAC       30
#endif
#endif

int     samplefac = DEFSMPFAC;  /* sampling factor */

                /* Samples array starts off holding spacing between adjacent
                 * samples, and ends up holding actual BGR sample values.
                 */
static BYTE     *thesamples;
static int      nsamples;
static BYTE     *cursamp;
static long     skipcount;

#define MAXSKIP         (1<<24-1)

#define nskip(sp)       ((long)(sp)[0]<<16|(long)(sp)[1]<<8|(long)(sp)[2])

#define setskip(sp,n)   ((sp)[0]=(n)>>16,(sp)[1]=((n)>>8)&255,(sp)[2]=(n)&255)

static  cpyclrtab();


neu_init(npixels)               /* initialize our sample array */
long    npixels;
{
        register int    nsleft;
        register long   sv;
        double  rval, cumprob;
        long    npleft;

        nsamples = npixels/samplefac;
        if (nsamples < 600)
                return(-1);
        thesamples = (BYTE *)malloc(nsamples*3);
        if (thesamples == NULL)
                return(-1);
        cursamp = thesamples;
        npleft = npixels;
        nsleft = nsamples;
        while (nsleft) {
                rval = frandom();       /* random distance to next sample */
                sv = 0;
                cumprob = 0.;
                while ((cumprob += (1.-cumprob)*nsleft/(npleft-sv)) < rval)
                        sv++;
                if (nsleft == nsamples)
                        skipcount = sv;
                else {
                        setskip(cursamp, sv);
                        cursamp += 3;
                }
                npleft -= sv+1;
                nsleft--;
        }
        setskip(cursamp, npleft);       /* tag on end to skip the rest */
        cursamp = thesamples;
        return(0);
}


neu_pixel(col)                  /* add pixel to our samples */
register BYTE   col[];
{
        if (!skipcount--) {
                skipcount = nskip(cursamp);
                cursamp[0] = col[BLU];
                cursamp[1] = col[GRN];
                cursamp[2] = col[RED];
                cursamp += 3;
        }
}


neu_colrs(cs, n)                /* add a scanline to our samples */
register COLR   *cs;
register int    n;
{
        while (n > skipcount) {
                cs += skipcount;
                n -= skipcount+1;
                skipcount = nskip(cursamp);
                cursamp[0] = cs[0][BLU];
                cursamp[1] = cs[0][GRN];
                cursamp[2] = cs[0][RED];
                cs++;
                cursamp += 3;
        }
        skipcount -= n;
}


neu_clrtab(ncolors)             /* make new color table using ncolors */
int     ncolors;
{
        clrtabsiz = ncolors;
        if (clrtabsiz > 256) clrtabsiz = 256;
        initnet();
        learn();
        unbiasnet();
        cpyclrtab();
        inxbuild();
                                /* we're done with our samples */
        free((void *)thesamples);
                                /* reset dithering function */
        neu_dith_colrs((BYTE *)NULL, (COLR *)NULL, 0);
                                /* return new color table size */
        return(clrtabsiz);
}


int
neu_map_pixel(col)              /* get pixel for color */
register BYTE   col[];
{
        return(inxsearch(col[BLU],col[GRN],col[RED]));
}


neu_map_colrs(bs, cs, n)        /* convert a scanline to color index values */
register BYTE   *bs;
register COLR   *cs;
register int    n;
{
        while (n-- > 0) {
                *bs++ = inxsearch(cs[0][BLU],cs[0][GRN],cs[0][RED]);
                cs++;
        }
}


neu_dith_colrs(bs, cs, n)       /* convert scanline to dithered index values */
register BYTE   *bs;
register COLR   *cs;
int     n;
{
        static short    (*cerr)[3] = NULL;
        static int      N = 0;
        int     err[3], errp[3];
        register int    x, i;

        if (n != N) {           /* get error propogation array */
                if (N) {
                        free((void *)cerr);
                        cerr = NULL;
                }
                if (n)
                        cerr = (short (*)[3])malloc(3*n*sizeof(short));
                if (cerr == NULL) {
                        N = 0;
                        map_colrs(bs, cs, n);
                        return;
                }
                N = n;
                memset((char *)cerr, '\0', 3*N*sizeof(short));
        }
        err[0] = err[1] = err[2] = 0;
        for (x = 0; x < n; x++) {
                for (i = 0; i < 3; i++) {       /* dither value */
                        errp[i] = err[i];
                        err[i] += cerr[x][i];
#ifdef MAXERR
                        if (err[i] > MAXERR) err[i] = MAXERR;
                        else if (err[i] < -MAXERR) err[i] = -MAXERR;
#endif
                        err[i] += cs[x][i];
                        if (err[i] < 0) err[i] = 0;
                        else if (err[i] > 255) err[i] = 255;
                }
                bs[x] = inxsearch(err[BLU],err[GRN],err[RED]);
                for (i = 0; i < 3; i++) {       /* propagate error */
                        err[i] -= clrtab[bs[x]][i];
                        err[i] /= 3;
                        cerr[x][i] = err[i] + errp[i];
                }
        }
}

/* The following was adapted and modified from the original (GW)        */

/* cheater definitions (GW) */
#define thepicture      thesamples
#define lengthcount     (nsamples*3)
#define samplefac       1

/*----------------------------------------------------------------------*/
/*                                                                      */
/*                              NeuQuant                                */
/*                              --------                                */
/*                                                                      */
/*              Copyright: Anthony Dekker, November 1994                */
/*                                                                      */
/* This program performs colour quantization of graphics images (SUN    */
/* raster files).  It uses a Kohonen Neural Network.  It produces       */
/* better results than existing methods and runs faster, using minimal  */
/* space (8kB plus the image itself).  The algorithm is described in    */
/* the paper "Kohonen Neural Networks for Optimal Colour Quantization"  */
/* to appear in the journal "Network: Computation in Neural Systems".   */
/* It is a significant improvement of an earlier algorithm.             */
/*                                                                      */
/* This program is distributed free for academic use or for evaluation  */
/* by commercial organizations.                                         */
/*                                                                      */
/*      Usage:  NeuQuant -n inputfile > outputfile                      */
/*                                                                      */
/* where n is a sampling factor for neural learning.                    */
/*                                                                      */
/* Program performance compared with other methods is as follows:       */
/*                                                                      */
/*      Algorithm               |  Av. CPU Time |  Quantization Error   */
/*      -------------------------------------------------------------   */
/*      NeuQuant -3             |  314          |  5.55                 */
/*      NeuQuant -10            |  119          |  5.97                 */
/*      NeuQuant -30            |  65           |  6.53                 */
/*      Oct-Trees               |  141          |  8.96                 */
/*      Median Cut (XV -best)   |  420          |  9.28                 */
/*      Median Cut (XV -slow)   |  72           |  12.15                */
/*                                                                      */
/* Author's address:    Dept of ISCS, National University of Singapore  */
/*                      Kent Ridge, Singapore 0511                      */
/* Email:       [email protected]                                     */
/*----------------------------------------------------------------------*/

#define bool            int
#define false           0
#define true            1

/* network defs */
#define netsize         clrtabsiz               /* number of colours - can change this */
#define maxnetpos       (netsize-1)
#define netbiasshift    4                       /* bias for colour values */
#define ncycles         100                     /* no. of learning cycles */

/* defs for freq and bias */
#define intbiasshift    16                      /* bias for fractions */
#define intbias         (((int) 1)<<intbiasshift)
#define gammashift      10                      /* gamma = 1024 */
#define gamma           (((int) 1)<<gammashift)
#define betashift       10
#define beta            (intbias>>betashift)    /* beta = 1/1024 */
#define betagamma       (intbias<<(gammashift-betashift))

/* defs for decreasing radius factor */
#define initrad         (256>>3)                /* for 256 cols, radius starts */
#define radiusbiasshift 6                       /* at 32.0 biased by 6 bits */
#define radiusbias      (((int) 1)<<radiusbiasshift)
#define initradius      (initrad*radiusbias)    /* and decreases by a */
#define radiusdec       30                      /* factor of 1/30 each cycle */ 

/* defs for decreasing alpha factor */
#define alphabiasshift  10                      /* alpha starts at 1.0 */
#define initalpha       (((int) 1)<<alphabiasshift)
int alphadec;                                   /* biased by 10 bits */

/* radbias and alpharadbias used for radpower calculation */
#define radbiasshift    8
#define radbias         (((int) 1)<<radbiasshift)
#define alpharadbshift  (alphabiasshift+radbiasshift)
#define alpharadbias    (((int) 1)<<alpharadbshift)

/* four primes near 500 - assume no image has a length so large */
/* that it is divisible by all four primes */
#define prime1          499
#define prime2          491
#define prime3          487
#define prime4          503

typedef int pixel[4];  /* BGRc */
pixel network[256];

int netindex[256];      /* for network lookup - really 256 */

int bias [256];         /* bias and freq arrays for learning */
int freq [256];
int radpower[initrad];  /* radpower for precomputation */


/* initialise network in range (0,0,0) to (255,255,255) */

initnet()       
{
        register int i;
        register int *p;
        
        for (i=0; i<netsize; i++) {
                p = network[i];
                p[0] = p[1] = p[2] = (i << (netbiasshift+8))/netsize;
                freq[i] = intbias/netsize;  /* 1/netsize */
                bias[i] = 0;
        }
}


/* do after unbias - insertion sort of network and build netindex[0..255] */

inxbuild()
{
        register int i,j,smallpos,smallval;
        register int *p,*q;
        int previouscol,startpos;

        previouscol = 0;
        startpos = 0;
        for (i=0; i<netsize; i++) {
                p = network[i];
                smallpos = i;
                smallval = p[1];        /* index on g */
                /* find smallest in i..netsize-1 */
                for (j=i+1; j<netsize; j++) {
                        q = network[j];
                        if (q[1] < smallval) {  /* index on g */
                                smallpos = j;
                                smallval = q[1]; /* index on g */
                        }
                }
                q = network[smallpos];
                /* swap p (i) and q (smallpos) entries */
                if (i != smallpos) {
                        j = q[0];   q[0] = p[0];   p[0] = j;
                        j = q[1];   q[1] = p[1];   p[1] = j;
                        j = q[2];   q[2] = p[2];   p[2] = j;
                        j = q[3];   q[3] = p[3];   p[3] = j;
                }
                /* smallval entry is now in position i */
                if (smallval != previouscol) {
                        netindex[previouscol] = (startpos+i)>>1;
                        for (j=previouscol+1; j<smallval; j++) netindex[j] = i;
                        previouscol = smallval;
                        startpos = i;
                }
        }
        netindex[previouscol] = (startpos+maxnetpos)>>1;
        for (j=previouscol+1; j<256; j++) netindex[j] = maxnetpos; /* really 256 */
}


int inxsearch(b,g,r)  /* accepts real BGR values after net is unbiased */
register int b,g,r;
{
        register int i,j,dist,a,bestd;
        register int *p;
        int best;

        bestd = 1000;   /* biggest possible dist is 256*3 */
        best = -1;
        i = netindex[g]; /* index on g */
        j = i-1;         /* start at netindex[g] and work outwards */

        while ((i<netsize) || (j>=0)) {
                if (i<netsize) {
                        p = network[i];
                        dist = p[1] - g;        /* inx key */
                        if (dist >= bestd) i = netsize; /* stop iter */
                        else {
                                i++;
                                if (dist<0) dist = -dist;
                                a = p[0] - b;   if (a<0) a = -a;
                                dist += a;
                                if (dist<bestd) {
                                        a = p[2] - r;   if (a<0) a = -a;
                                        dist += a;
                                        if (dist<bestd) {bestd=dist; best=p[3];}
                                }
                        }
                }
                if (j>=0) {
                        p = network[j];
                        dist = g - p[1]; /* inx key - reverse dif */
                        if (dist >= bestd) j = -1; /* stop iter */
                        else {
                                j--;
                                if (dist<0) dist = -dist;
                                a = p[0] - b;   if (a<0) a = -a;
                                dist += a;
                                if (dist<bestd) {
                                        a = p[2] - r;   if (a<0) a = -a;
                                        dist += a;
                                        if (dist<bestd) {bestd=dist; best=p[3];}
                                }
                        }
                }
        }
        return(best);
}


/* finds closest neuron (min dist) and updates freq */
/* finds best neuron (min dist-bias) and returns position */
/* for frequently chosen neurons, freq[i] is high and bias[i] is negative */
/* bias[i] = gamma*((1/netsize)-freq[i]) */

int contest(b,g,r)      /* accepts biased BGR values */
register int b,g,r;
{
        register int i,dist,a,biasdist,betafreq;
        int bestpos,bestbiaspos,bestd,bestbiasd;
        register int *p,*f, *n;

        bestd = ~(((int) 1)<<31);
        bestbiasd = bestd;
        bestpos = -1;
        bestbiaspos = bestpos;
        p = bias;
        f = freq;

        for (i=0; i<netsize; i++) {
                n = network[i];
                dist = n[0] - b;   if (dist<0) dist = -dist;
                a = n[1] - g;   if (a<0) a = -a;
                dist += a;
                a = n[2] - r;   if (a<0) a = -a;
                dist += a;
                if (dist<bestd) {bestd=dist; bestpos=i;}
                biasdist = dist - ((*p)>>(intbiasshift-netbiasshift));
                if (biasdist<bestbiasd) {bestbiasd=biasdist; bestbiaspos=i;}
                betafreq = (*f >> betashift);
                *f++ -= betafreq;
                *p++ += (betafreq<<gammashift);
        }
        freq[bestpos] += beta;
        bias[bestpos] -= betagamma;
        return(bestbiaspos);
}


/* move neuron i towards (b,g,r) by factor alpha */

altersingle(alpha,i,b,g,r)      /* accepts biased BGR values */
register int alpha,i,b,g,r;
{
        register int *n;

        n = network[i];         /* alter hit neuron */
        *n -= (alpha*(*n - b)) / initalpha;
        n++;
        *n -= (alpha*(*n - g)) / initalpha;
        n++;
        *n -= (alpha*(*n - r)) / initalpha;
}


/* move neurons adjacent to i towards (b,g,r) by factor */
/* alpha*(1-((i-j)^2/[r]^2)) precomputed as radpower[|i-j|]*/

alterneigh(rad,i,b,g,r) /* accents biased BGR values */
int rad,i;
register int b,g,r;
{
        register int j,k,lo,hi,a;
        register int *p, *q;

        lo = i-rad;   if (lo<-1) lo= -1;
        hi = i+rad;   if (hi>netsize) hi=netsize;

        j = i+1;
        k = i-1;
        q = radpower;
        while ((j<hi) || (k>lo)) {
                a = (*(++q));
                if (j<hi) {
                        p = network[j];
                        *p -= (a*(*p - b)) / alpharadbias;
                        p++;
                        *p -= (a*(*p - g)) / alpharadbias;
                        p++;
                        *p -= (a*(*p - r)) / alpharadbias;
                        j++;
                }
                if (k>lo) {
                        p = network[k];
                        *p -= (a*(*p - b)) / alpharadbias;
                        p++;
                        *p -= (a*(*p - g)) / alpharadbias;
                        p++;
                        *p -= (a*(*p - r)) / alpharadbias;
                        k--;
                }
        }
}


learn()
{
        register int i,j,b,g,r;
        int radius,rad,alpha,step,delta,samplepixels;
        register unsigned char *p;
        unsigned char *lim;

        alphadec = 30 + ((samplefac-1)/3);
        p = thepicture;
        lim = thepicture + lengthcount;
        samplepixels = lengthcount/(3*samplefac);
        delta = samplepixels/ncycles;
        alpha = initalpha;
        radius = initradius;
        
        rad = radius >> radiusbiasshift;
        if (rad <= 1) rad = 0;
        for (i=0; i<rad; i++) 
                radpower[i] = alpha*(((rad*rad - i*i)*radbias)/(rad*rad));
        
        if ((lengthcount%prime1) != 0) step = 3*prime1;
        else {
                if ((lengthcount%prime2) !=0) step = 3*prime2;
                else {
                        if ((lengthcount%prime3) !=0) step = 3*prime3;
                        else step = 3*prime4;
                }
        }
        
        i = 0;
        while (i < samplepixels) {
                b = p[0] << netbiasshift;
                g = p[1] << netbiasshift;
                r = p[2] << netbiasshift;
                j = contest(b,g,r);

                altersingle(alpha,j,b,g,r);
                if (rad) alterneigh(rad,j,b,g,r);   /* alter neighbours */

                p += step;
                if (p >= lim) p -= lengthcount;
        
                i++;
                if (i%delta == 0) {     
                        alpha -= alpha / alphadec;
                        radius -= radius / radiusdec;
                        rad = radius >> radiusbiasshift;
                        if (rad <= 1) rad = 0;
                        for (j=0; j<rad; j++) 
                                radpower[j] = alpha*(((rad*rad - j*j)*radbias)/(rad*rad));
                }
        }
}
        
/* unbias network to give 0..255 entries */
/* which can then be used for colour map */
/* and record position i to prepare for sort */

unbiasnet()
{
        int i,j;

        for (i=0; i<netsize; i++) {
                for (j=0; j<3; j++)
                        network[i][j] >>= netbiasshift;
                network[i][3] = i; /* record colour no */
        }
}


/* Don't do this until the network has been unbiased (GW) */
                
static
cpyclrtab()
{
        register int i,j,k;
        
        for (j=0; j<netsize; j++) {
                k = network[j][3];
                for (i = 0; i < 3; i++)
                        clrtab[k][i] = network[j][2-i];
        }
}
Revision:	2.10
Committed:	Mon Jun 30 14:59:12 2003 UTC (20 years, 10 months ago) by schorsch
Content type:	text/plain
Branch:	MAIN
Changes since 2.9:	+6 -4 lines
Log Message:	Replaced most outdated BSD function calls with their posix equivalents, and cleaned up a few other platform dependencies.
#	User	Rev	Content
1	greg	2.1	#ifndef lint
2	schorsch	2.10	static const char RCSid[] = "$Id: neuclrtab.c,v 2.9 2003/02/22 02:07:27 greg Exp $";
3	greg	2.1	#endif
4			/*
5			* Neural-Net quantization algorithm based on work of Anthony Dekker
6			*/
7
8	schorsch	2.10	#include "copyright.h"
9
10			#include <string.h>
11
12	greg	2.1	#include "standard.h"
13			#include "color.h"
14			#include "random.h"
15
16			#ifdef COMPAT_MODE
17			#define neu_init new_histo
18			#define neu_pixel cnt_pixel
19			#define neu_colrs cnt_colrs
20			#define neu_clrtab new_clrtab
21			#define neu_map_pixel map_pixel
22			#define neu_map_colrs map_colrs
23			#define neu_dith_colrs dith_colrs
24			#endif
25			/* our color table (global) */
26			extern BYTE clrtab[256][3];
27			static int clrtabsiz;
28
29			#ifndef DEFSMPFAC
30			#ifdef SPEED
31			#define DEFSMPFAC (240/SPEED+3)
32			#else
33			#define DEFSMPFAC 30
34			#endif
35			#endif
36
37			int samplefac = DEFSMPFAC; /* sampling factor */
38
39			/* Samples array starts off holding spacing between adjacent
40			* samples, and ends up holding actual BGR sample values.
41			*/
42			static BYTE *thesamples;
43			static int nsamples;
44			static BYTE *cursamp;
45			static long skipcount;
46
47			#define MAXSKIP (1<<24-1)
48
49			#define nskip(sp) ((long)(sp)[0]<<16\|(long)(sp)[1]<<8\|(long)(sp)[2])
50
51			#define setskip(sp,n) ((sp)[0]=(n)>>16,(sp)[1]=((n)>>8)&255,(sp)[2]=(n)&255)
52
53	greg	2.8	static cpyclrtab();
54
55	greg	2.1
56			neu_init(npixels) /* initialize our sample array */
57			long npixels;
58			{
59			register int nsleft;
60			register long sv;
61			double rval, cumprob;
62			long npleft;
63
64			nsamples = npixels/samplefac;
65			if (nsamples < 600)
66			return(-1);
67	greg	2.2	thesamples = (BYTE )malloc(nsamples3);
68	greg	2.1	if (thesamples == NULL)
69			return(-1);
70			cursamp = thesamples;
71			npleft = npixels;
72			nsleft = nsamples;
73			while (nsleft) {
74			rval = frandom(); /* random distance to next sample */
75			sv = 0;
76			cumprob = 0.;
77			while ((cumprob += (1.-cumprob)*nsleft/(npleft-sv)) < rval)
78			sv++;
79	greg	2.2	if (nsleft == nsamples)
80			skipcount = sv;
81			else {
82			setskip(cursamp, sv);
83			cursamp += 3;
84			}
85			npleft -= sv+1;
86	greg	2.1	nsleft--;
87			}
88	greg	2.2	setskip(cursamp, npleft); /* tag on end to skip the rest */
89	greg	2.1	cursamp = thesamples;
90			return(0);
91			}
92
93
94			neu_pixel(col) /* add pixel to our samples */
95			register BYTE col[];
96			{
97			if (!skipcount--) {
98	greg	2.2	skipcount = nskip(cursamp);
99	greg	2.1	cursamp[0] = col[BLU];
100			cursamp[1] = col[GRN];
101			cursamp[2] = col[RED];
102			cursamp += 3;
103			}
104			}
105
106
107			neu_colrs(cs, n) /* add a scanline to our samples */
108			register COLR *cs;
109			register int n;
110			{
111			while (n > skipcount) {
112			cs += skipcount;
113	greg	2.2	n -= skipcount+1;
114			skipcount = nskip(cursamp);
115	greg	2.1	cursamp[0] = cs[0][BLU];
116			cursamp[1] = cs[0][GRN];
117			cursamp[2] = cs[0][RED];
118			cs++;
119			cursamp += 3;
120			}
121			skipcount -= n;
122			}
123
124
125			neu_clrtab(ncolors) /* make new color table using ncolors */
126			int ncolors;
127			{
128			clrtabsiz = ncolors;
129			if (clrtabsiz > 256) clrtabsiz = 256;
130			initnet();
131			learn();
132			unbiasnet();
133			cpyclrtab();
134			inxbuild();
135			/* we're done with our samples */
136	greg	2.9	free((void *)thesamples);
137	greg	2.1	/* reset dithering function */
138			neu_dith_colrs((BYTE )NULL, (COLR )NULL, 0);
139			/* return new color table size */
140			return(clrtabsiz);
141			}
142
143
144			int
145			neu_map_pixel(col) /* get pixel for color */
146			register BYTE col[];
147			{
148			return(inxsearch(col[BLU],col[GRN],col[RED]));
149			}
150
151
152			neu_map_colrs(bs, cs, n) /* convert a scanline to color index values */
153			register BYTE *bs;
154			register COLR *cs;
155			register int n;
156			{
157			while (n-- > 0) {
158			*bs++ = inxsearch(cs[0][BLU],cs[0][GRN],cs[0][RED]);
159			cs++;
160			}
161			}
162
163
164			neu_dith_colrs(bs, cs, n) /* convert scanline to dithered index values */
165			register BYTE *bs;
166			register COLR *cs;
167			int n;
168			{
169			static short (*cerr)[3] = NULL;
170			static int N = 0;
171			int err[3], errp[3];
172			register int x, i;
173
174			if (n != N) { /* get error propogation array */
175			if (N) {
176	greg	2.9	free((void *)cerr);
177	greg	2.1	cerr = NULL;
178			}
179			if (n)
180			cerr = (short ()[3])malloc(3n*sizeof(short));
181			if (cerr == NULL) {
182			N = 0;
183			map_colrs(bs, cs, n);
184			return;
185			}
186			N = n;
187	schorsch	2.10	memset((char )cerr, '\0', 3N*sizeof(short));
188	greg	2.1	}
189			err[0] = err[1] = err[2] = 0;
190			for (x = 0; x < n; x++) {
191			for (i = 0; i < 3; i++) { /* dither value */
192			errp[i] = err[i];
193			err[i] += cerr[x][i];
194			#ifdef MAXERR
195			if (err[i] > MAXERR) err[i] = MAXERR;
196			else if (err[i] < -MAXERR) err[i] = -MAXERR;
197			#endif
198			err[i] += cs[x][i];
199			if (err[i] < 0) err[i] = 0;
200			else if (err[i] > 255) err[i] = 255;
201			}
202			bs[x] = inxsearch(err[BLU],err[GRN],err[RED]);
203			for (i = 0; i < 3; i++) { /* propagate error */
204			err[i] -= clrtab[bs[x]][i];
205			err[i] /= 3;
206			cerr[x][i] = err[i] + errp[i];
207			}
208			}
209			}
210
211			/* The following was adapted and modified from the original (GW) */
212	greg	2.6
213			/* cheater definitions (GW) */
214			#define thepicture thesamples
215			#define lengthcount (nsamples*3)
216			#define samplefac 1
217
218	greg	2.1	/----------------------------------------------------------------------/
219			/* */
220			/* NeuQuant */
221			/* -------- */
222			/* */
223	greg	2.6	/* Copyright: Anthony Dekker, November 1994 */
224	greg	2.1	/* */
225			/* This program performs colour quantization of graphics images (SUN */
226			/* raster files). It uses a Kohonen Neural Network. It produces */
227			/* better results than existing methods and runs faster, using minimal */
228			/* space (8kB plus the image itself). The algorithm is described in */
229			/* the paper "Kohonen Neural Networks for Optimal Colour Quantization" */
230			/* to appear in the journal "Network: Computation in Neural Systems". */
231			/* It is a significant improvement of an earlier algorithm. */
232			/* */
233			/* This program is distributed free for academic use or for evaluation */
234			/* by commercial organizations. */
235			/* */
236			/* Usage: NeuQuant -n inputfile > outputfile */
237			/* */
238			/* where n is a sampling factor for neural learning. */
239			/* */
240			/* Program performance compared with other methods is as follows: */
241			/* */
242			/* Algorithm \| Av. CPU Time \| Quantization Error */
243			/* ------------------------------------------------------------- */
244			/* NeuQuant -3 \| 314 \| 5.55 */
245			/* NeuQuant -10 \| 119 \| 5.97 */
246			/* NeuQuant -30 \| 65 \| 6.53 */
247			/* Oct-Trees \| 141 \| 8.96 */
248			/* Median Cut (XV -best) \| 420 \| 9.28 */
249			/* Median Cut (XV -slow) \| 72 \| 12.15 */
250			/* */
251			/* Author's address: Dept of ISCS, National University of Singapore */
252			/* Kent Ridge, Singapore 0511 */
253			/* Email: [email protected] */
254			/----------------------------------------------------------------------/
255
256	greg	2.6	#define bool int
257			#define false 0
258			#define true 1
259	greg	2.1
260	greg	2.6	/* network defs */
261	greg	2.7	#define netsize clrtabsiz /* number of colours - can change this */
262	greg	2.6	#define maxnetpos (netsize-1)
263			#define netbiasshift 4 /* bias for colour values */
264			#define ncycles 100 /* no. of learning cycles */
265	greg	2.1
266			/* defs for freq and bias */
267	greg	2.6	#define intbiasshift 16 /* bias for fractions */
268			#define intbias (((int) 1)<<intbiasshift)
269			#define gammashift 10 /* gamma = 1024 */
270			#define gamma (((int) 1)<<gammashift)
271			#define betashift 10
272			#define beta (intbias>>betashift) /* beta = 1/1024 */
273	greg	2.1	#define betagamma (intbias<<(gammashift-betashift))
274
275	greg	2.6	/* defs for decreasing radius factor */
276	greg	2.7	#define initrad (256>>3) /* for 256 cols, radius starts */
277	greg	2.6	#define radiusbiasshift 6 /* at 32.0 biased by 6 bits */
278			#define radiusbias (((int) 1)<<radiusbiasshift)
279			#define initradius (initradradiusbias) / and decreases by a */
280			#define radiusdec 30 /* factor of 1/30 each cycle */
281
282			/* defs for decreasing alpha factor */
283			#define alphabiasshift 10 /* alpha starts at 1.0 */
284			#define initalpha (((int) 1)<<alphabiasshift)
285			int alphadec; /* biased by 10 bits */
286
287			/* radbias and alpharadbias used for radpower calculation */
288	greg	2.1	#define radbiasshift 8
289	greg	2.6	#define radbias (((int) 1)<<radbiasshift)
290	greg	2.1	#define alpharadbshift (alphabiasshift+radbiasshift)
291	greg	2.6	#define alpharadbias (((int) 1)<<alpharadbshift)
292	greg	2.1
293	greg	2.6	/* four primes near 500 - assume no image has a length so large */
294			/* that it is divisible by all four primes */
295			#define prime1 499
296			#define prime2 491
297			#define prime3 487
298			#define prime4 503
299	greg	2.1
300			typedef int pixel[4]; /* BGRc */
301	greg	2.7	pixel network[256];
302	greg	2.1
303	greg	2.6	int netindex[256]; /* for network lookup - really 256 */
304	greg	2.1
305	greg	2.7	int bias [256]; /* bias and freq arrays for learning */
306			int freq [256];
307	greg	2.6	int radpower[initrad]; /* radpower for precomputation */
308	greg	2.1
309
310	greg	2.6	/* initialise network in range (0,0,0) to (255,255,255) */
311	greg	2.1
312	greg	2.6	initnet()
313	greg	2.1	{
314			register int i;
315			register int *p;
316
317	greg	2.7	for (i=0; i<netsize; i++) {
318	greg	2.1	p = network[i];
319	greg	2.7	p[0] = p[1] = p[2] = (i << (netbiasshift+8))/netsize;
320			freq[i] = intbias/netsize; /* 1/netsize */
321	greg	2.1	bias[i] = 0;
322			}
323			}
324
325
326	greg	2.6	/* do after unbias - insertion sort of network and build netindex[0..255] */
327
328	greg	2.1	inxbuild()
329			{
330			register int i,j,smallpos,smallval;
331			register int p,q;
332	greg	2.6	int previouscol,startpos;
333	greg	2.1
334	greg	2.6	previouscol = 0;
335			startpos = 0;
336	greg	2.7	for (i=0; i<netsize; i++) {
337	greg	2.1	p = network[i];
338			smallpos = i;
339			smallval = p[1]; /* index on g */
340	greg	2.7	/* find smallest in i..netsize-1 */
341			for (j=i+1; j<netsize; j++) {
342	greg	2.1	q = network[j];
343			if (q[1] < smallval) { /* index on g */
344			smallpos = j;
345			smallval = q[1]; /* index on g */
346			}
347			}
348			q = network[smallpos];
349	greg	2.6	/* swap p (i) and q (smallpos) entries */
350	greg	2.1	if (i != smallpos) {
351			j = q[0]; q[0] = p[0]; p[0] = j;
352			j = q[1]; q[1] = p[1]; p[1] = j;
353			j = q[2]; q[2] = p[2]; p[2] = j;
354			j = q[3]; q[3] = p[3]; p[3] = j;
355			}
356			/* smallval entry is now in position i */
357	greg	2.6	if (smallval != previouscol) {
358			netindex[previouscol] = (startpos+i)>>1;
359			for (j=previouscol+1; j<smallval; j++) netindex[j] = i;
360			previouscol = smallval;
361			startpos = i;
362	greg	2.1	}
363			}
364	greg	2.6	netindex[previouscol] = (startpos+maxnetpos)>>1;
365			for (j=previouscol+1; j<256; j++) netindex[j] = maxnetpos; /* really 256 */
366	greg	2.1	}
367
368
369	greg	2.6	int inxsearch(b,g,r) /* accepts real BGR values after net is unbiased */
370	greg	2.1	register int b,g,r;
371			{
372	greg	2.6	register int i,j,dist,a,bestd;
373	greg	2.1	register int *p;
374	greg	2.6	int best;
375	greg	2.1
376			bestd = 1000; /* biggest possible dist is 2563 /
377			best = -1;
378			i = netindex[g]; /* index on g */
379	greg	2.6	j = i-1; /* start at netindex[g] and work outwards */
380	greg	2.1
381	greg	2.7	while ((i<netsize) \|\| (j>=0)) {
382			if (i<netsize) {
383	greg	2.1	p = network[i];
384	greg	2.6	dist = p[1] - g; /* inx key */
385	greg	2.7	if (dist >= bestd) i = netsize; /* stop iter */
386	greg	2.1	else {
387			i++;
388	greg	2.6	if (dist<0) dist = -dist;
389			a = p[0] - b; if (a<0) a = -a;
390			dist += a;
391			if (dist<bestd) {
392			a = p[2] - r; if (a<0) a = -a;
393			dist += a;
394			if (dist<bestd) {bestd=dist; best=p[3];}
395	greg	2.1	}
396			}
397			}
398			if (j>=0) {
399			p = network[j];
400	greg	2.6	dist = g - p[1]; /* inx key - reverse dif */
401			if (dist >= bestd) j = -1; /* stop iter */
402	greg	2.1	else {
403			j--;
404	greg	2.6	if (dist<0) dist = -dist;
405			a = p[0] - b; if (a<0) a = -a;
406			dist += a;
407			if (dist<bestd) {
408			a = p[2] - r; if (a<0) a = -a;
409			dist += a;
410			if (dist<bestd) {bestd=dist; best=p[3];}
411	greg	2.1	}
412			}
413			}
414			}
415			return(best);
416			}
417
418
419	greg	2.6	/* finds closest neuron (min dist) and updates freq */
420			/* finds best neuron (min dist-bias) and returns position */
421			/* for frequently chosen neurons, freq[i] is high and bias[i] is negative */
422	greg	2.7	/* bias[i] = gamma((1/netsize)-freq[i]) /
423	greg	2.6
424			int contest(b,g,r) /* accepts biased BGR values */
425	greg	2.1	register int b,g,r;
426			{
427	greg	2.6	register int i,dist,a,biasdist,betafreq;
428			int bestpos,bestbiaspos,bestd,bestbiasd;
429			register int p,f, *n;
430	greg	2.1
431	greg	2.6	bestd = ~(((int) 1)<<31);
432	greg	2.1	bestbiasd = bestd;
433	greg	2.6	bestpos = -1;
434			bestbiaspos = bestpos;
435			p = bias;
436			f = freq;
437
438	greg	2.7	for (i=0; i<netsize; i++) {
439	greg	2.6	n = network[i];
440			dist = n[0] - b; if (dist<0) dist = -dist;
441			a = n[1] - g; if (a<0) a = -a;
442			dist += a;
443			a = n[2] - r; if (a<0) a = -a;
444			dist += a;
445			if (dist<bestd) {bestd=dist; bestpos=i;}
446			biasdist = dist - ((*p)>>(intbiasshift-netbiasshift));
447			if (biasdist<bestbiasd) {bestbiasd=biasdist; bestbiaspos=i;}
448			betafreq = (*f >> betashift);
449			*f++ -= betafreq;
450			*p++ += (betafreq<<gammashift);
451	greg	2.1	}
452	greg	2.6	freq[bestpos] += beta;
453			bias[bestpos] -= betagamma;
454			return(bestbiaspos);
455	greg	2.1	}
456
457
458	greg	2.6	/* move neuron i towards (b,g,r) by factor alpha */
459
460			altersingle(alpha,i,b,g,r) /* accepts biased BGR values */
461			register int alpha,i,b,g,r;
462			{
463			register int *n;
464
465			n = network[i]; /* alter hit neuron */
466			n -= (alpha(*n - b)) / initalpha;
467			n++;
468			n -= (alpha(*n - g)) / initalpha;
469			n++;
470			n -= (alpha(*n - r)) / initalpha;
471			}
472
473
474			/* move neurons adjacent to i towards (b,g,r) by factor */
475			/* alpha(1-((i-j)^2/[r]^2)) precomputed as radpower[\|i-j\|]/
476
477			alterneigh(rad,i,b,g,r) /* accents biased BGR values */
478	greg	2.1	int rad,i;
479			register int b,g,r;
480			{
481			register int j,k,lo,hi,a;
482			register int p, q;
483
484	greg	2.6	lo = i-rad; if (lo<-1) lo= -1;
485	greg	2.7	hi = i+rad; if (hi>netsize) hi=netsize;
486	greg	2.1
487			j = i+1;
488			k = i-1;
489			q = radpower;
490			while ((j<hi) \|\| (k>lo)) {
491			a = (*(++q));
492			if (j<hi) {
493			p = network[j];
494			p -= (a(*p - b)) / alpharadbias;
495			p++;
496			p -= (a(*p - g)) / alpharadbias;
497			p++;
498			p -= (a(*p - r)) / alpharadbias;
499			j++;
500			}
501			if (k>lo) {
502			p = network[k];
503			p -= (a(*p - b)) / alpharadbias;
504			p++;
505			p -= (a(*p - g)) / alpharadbias;
506			p++;
507			p -= (a(*p - r)) / alpharadbias;
508			k--;
509			}
510			}
511			}
512
513
514			learn()
515			{
516			register int i,j,b,g,r;
517	greg	2.6	int radius,rad,alpha,step,delta,samplepixels;
518	greg	2.1	register unsigned char *p;
519			unsigned char *lim;
520
521	greg	2.6	alphadec = 30 + ((samplefac-1)/3);
522			p = thepicture;
523	greg	2.1	lim = thepicture + lengthcount;
524	greg	2.6	samplepixels = lengthcount/(3*samplefac);
525			delta = samplepixels/ncycles;
526	greg	2.1	alpha = initalpha;
527			radius = initradius;
528	greg	2.6
529	greg	2.1	rad = radius >> radiusbiasshift;
530			if (rad <= 1) rad = 0;
531			for (i=0; i<rad; i++)
532			radpower[i] = alpha(((radrad - ii)radbias)/(rad*rad));
533	greg	2.6
534			if ((lengthcount%prime1) != 0) step = 3*prime1;
535	greg	2.1	else {
536	greg	2.6	if ((lengthcount%prime2) !=0) step = 3*prime2;
537	greg	2.1	else {
538	greg	2.6	if ((lengthcount%prime3) !=0) step = 3*prime3;
539			else step = 3*prime4;
540	greg	2.1	}
541			}
542	greg	2.6
543	greg	2.1	i = 0;
544	greg	2.6	while (i < samplepixels) {
545	greg	2.1	b = p[0] << netbiasshift;
546			g = p[1] << netbiasshift;
547			r = p[2] << netbiasshift;
548			j = contest(b,g,r);
549
550			altersingle(alpha,j,b,g,r);
551	greg	2.6	if (rad) alterneigh(rad,j,b,g,r); /* alter neighbours */
552	greg	2.1
553			p += step;
554			if (p >= lim) p -= lengthcount;
555
556			i++;
557			if (i%delta == 0) {
558			alpha -= alpha / alphadec;
559			radius -= radius / radiusdec;
560			rad = radius >> radiusbiasshift;
561			if (rad <= 1) rad = 0;
562			for (j=0; j<rad; j++)
563			radpower[j] = alpha(((radrad - jj)radbias)/(rad*rad));
564			}
565			}
566			}
567
568	greg	2.6	/* unbias network to give 0..255 entries */
569			/* which can then be used for colour map */
570			/* and record position i to prepare for sort */
571
572	greg	2.1	unbiasnet()
573			{
574			int i,j;
575
576	greg	2.7	for (i=0; i<netsize; i++) {
577	greg	2.1	for (j=0; j<3; j++)
578			network[i][j] >>= netbiasshift;
579			network[i][3] = i; /* record colour no */
580			}
581			}
582
583	greg	2.6
584			/* Don't do this until the network has been unbiased (GW) */
585	greg	2.1
586			static
587			cpyclrtab()
588			{
589			register int i,j,k;
590
591	greg	2.7	for (j=0; j<netsize; j++) {
592	greg	2.1	k = network[j][3];
593			for (i = 0; i < 3; i++)
594			clrtab[k][i] = network[j][2-i];
595			}
596			}