src/px/neuclrtab.c

/* Copyright (c) 1994 Regents of the University of California */

#ifndef lint
static char SCCSid[] = "$SunId$ LBL";
#endif

/*
 * Neural-Net quantization algorithm based on work of Anthony Dekker
 */

#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((char *)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((char *)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;
                bzero((char *)cerr, 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:	2.8
Committed:	Mon Dec 12 12:19:04 1994 UTC (29 years, 4 months ago) by greg
Content type:	text/plain
Branch:	MAIN
Changes since 2.7:	+2 -0 lines
Log Message:	added static function definitions
#	User	Rev	Content
1	greg	2.1	/* Copyright (c) 1994 Regents of the University of California */
2
3			#ifndef lint
4			static char SCCSid[] = "$SunId$ LBL";
5			#endif
6
7			/*
8			* Neural-Net quantization algorithm based on work of Anthony Dekker
9			*/
10
11			#include "standard.h"
12
13			#include "color.h"
14
15			#include "random.h"
16
17			#ifdef COMPAT_MODE
18			#define neu_init new_histo
19			#define neu_pixel cnt_pixel
20			#define neu_colrs cnt_colrs
21			#define neu_clrtab new_clrtab
22			#define neu_map_pixel map_pixel
23			#define neu_map_colrs map_colrs
24			#define neu_dith_colrs dith_colrs
25			#endif
26			/* our color table (global) */
27			extern BYTE clrtab[256][3];
28			static int clrtabsiz;
29
30			#ifndef DEFSMPFAC
31			#ifdef SPEED
32			#define DEFSMPFAC (240/SPEED+3)
33			#else
34			#define DEFSMPFAC 30
35			#endif
36			#endif
37
38			int samplefac = DEFSMPFAC; /* sampling factor */
39
40			/* Samples array starts off holding spacing between adjacent
41			* samples, and ends up holding actual BGR sample values.
42			*/
43			static BYTE *thesamples;
44			static int nsamples;
45			static BYTE *cursamp;
46			static long skipcount;
47
48			#define MAXSKIP (1<<24-1)
49
50			#define nskip(sp) ((long)(sp)[0]<<16\|(long)(sp)[1]<<8\|(long)(sp)[2])
51
52			#define setskip(sp,n) ((sp)[0]=(n)>>16,(sp)[1]=((n)>>8)&255,(sp)[2]=(n)&255)
53
54	greg	2.8	static cpyclrtab();
55
56	greg	2.1
57			neu_init(npixels) /* initialize our sample array */
58			long npixels;
59			{
60			register int nsleft;
61			register long sv;
62			double rval, cumprob;
63			long npleft;
64
65			nsamples = npixels/samplefac;
66			if (nsamples < 600)
67			return(-1);
68	greg	2.2	thesamples = (BYTE )malloc(nsamples3);
69	greg	2.1	if (thesamples == NULL)
70			return(-1);
71			cursamp = thesamples;
72			npleft = npixels;
73			nsleft = nsamples;
74			while (nsleft) {
75			rval = frandom(); /* random distance to next sample */
76			sv = 0;
77			cumprob = 0.;
78			while ((cumprob += (1.-cumprob)*nsleft/(npleft-sv)) < rval)
79			sv++;
80	greg	2.2	if (nsleft == nsamples)
81			skipcount = sv;
82			else {
83			setskip(cursamp, sv);
84			cursamp += 3;
85			}
86			npleft -= sv+1;
87	greg	2.1	nsleft--;
88			}
89	greg	2.2	setskip(cursamp, npleft); /* tag on end to skip the rest */
90	greg	2.1	cursamp = thesamples;
91			return(0);
92			}
93
94
95			neu_pixel(col) /* add pixel to our samples */
96			register BYTE col[];
97			{
98			if (!skipcount--) {
99	greg	2.2	skipcount = nskip(cursamp);
100	greg	2.1	cursamp[0] = col[BLU];
101			cursamp[1] = col[GRN];
102			cursamp[2] = col[RED];
103			cursamp += 3;
104			}
105			}
106
107
108			neu_colrs(cs, n) /* add a scanline to our samples */
109			register COLR *cs;
110			register int n;
111			{
112			while (n > skipcount) {
113			cs += skipcount;
114	greg	2.2	n -= skipcount+1;
115			skipcount = nskip(cursamp);
116	greg	2.1	cursamp[0] = cs[0][BLU];
117			cursamp[1] = cs[0][GRN];
118			cursamp[2] = cs[0][RED];
119			cs++;
120			cursamp += 3;
121			}
122			skipcount -= n;
123			}
124
125
126			neu_clrtab(ncolors) /* make new color table using ncolors */
127			int ncolors;
128			{
129			clrtabsiz = ncolors;
130			if (clrtabsiz > 256) clrtabsiz = 256;
131			initnet();
132			learn();
133			unbiasnet();
134			cpyclrtab();
135			inxbuild();
136			/* we're done with our samples */
137			free((char *)thesamples);
138			/* reset dithering function */
139			neu_dith_colrs((BYTE )NULL, (COLR )NULL, 0);
140			/* return new color table size */
141			return(clrtabsiz);
142			}
143
144
145			int
146			neu_map_pixel(col) /* get pixel for color */
147			register BYTE col[];
148			{
149			return(inxsearch(col[BLU],col[GRN],col[RED]));
150			}
151
152
153			neu_map_colrs(bs, cs, n) /* convert a scanline to color index values */
154			register BYTE *bs;
155			register COLR *cs;
156			register int n;
157			{
158			while (n-- > 0) {
159			*bs++ = inxsearch(cs[0][BLU],cs[0][GRN],cs[0][RED]);
160			cs++;
161			}
162			}
163
164
165			neu_dith_colrs(bs, cs, n) /* convert scanline to dithered index values */
166			register BYTE *bs;
167			register COLR *cs;
168			int n;
169			{
170			static short (*cerr)[3] = NULL;
171			static int N = 0;
172			int err[3], errp[3];
173			register int x, i;
174
175			if (n != N) { /* get error propogation array */
176			if (N) {
177			free((char *)cerr);
178			cerr = NULL;
179			}
180			if (n)
181			cerr = (short ()[3])malloc(3n*sizeof(short));
182			if (cerr == NULL) {
183			N = 0;
184			map_colrs(bs, cs, n);
185			return;
186			}
187			N = n;
188			bzero((char )cerr, 3N*sizeof(short));
189			}
190			err[0] = err[1] = err[2] = 0;
191			for (x = 0; x < n; x++) {
192			for (i = 0; i < 3; i++) { /* dither value */
193			errp[i] = err[i];
194			err[i] += cerr[x][i];
195			#ifdef MAXERR
196			if (err[i] > MAXERR) err[i] = MAXERR;
197			else if (err[i] < -MAXERR) err[i] = -MAXERR;
198			#endif
199			err[i] += cs[x][i];
200			if (err[i] < 0) err[i] = 0;
201			else if (err[i] > 255) err[i] = 255;
202			}
203			bs[x] = inxsearch(err[BLU],err[GRN],err[RED]);
204			for (i = 0; i < 3; i++) { /* propagate error */
205			err[i] -= clrtab[bs[x]][i];
206			err[i] /= 3;
207			cerr[x][i] = err[i] + errp[i];
208			}
209			}
210			}
211
212			/* The following was adapted and modified from the original (GW) */
213	greg	2.6
214			/* cheater definitions (GW) */
215			#define thepicture thesamples
216			#define lengthcount (nsamples*3)
217			#define samplefac 1
218
219	greg	2.1	/----------------------------------------------------------------------/
220			/* */
221			/* NeuQuant */
222			/* -------- */
223			/* */
224	greg	2.6	/* Copyright: Anthony Dekker, November 1994 */
225	greg	2.1	/* */
226			/* This program performs colour quantization of graphics images (SUN */
227			/* raster files). It uses a Kohonen Neural Network. It produces */
228			/* better results than existing methods and runs faster, using minimal */
229			/* space (8kB plus the image itself). The algorithm is described in */
230			/* the paper "Kohonen Neural Networks for Optimal Colour Quantization" */
231			/* to appear in the journal "Network: Computation in Neural Systems". */
232			/* It is a significant improvement of an earlier algorithm. */
233			/* */
234			/* This program is distributed free for academic use or for evaluation */
235			/* by commercial organizations. */
236			/* */
237			/* Usage: NeuQuant -n inputfile > outputfile */
238			/* */
239			/* where n is a sampling factor for neural learning. */
240			/* */
241			/* Program performance compared with other methods is as follows: */
242			/* */
243			/* Algorithm \| Av. CPU Time \| Quantization Error */
244			/* ------------------------------------------------------------- */
245			/* NeuQuant -3 \| 314 \| 5.55 */
246			/* NeuQuant -10 \| 119 \| 5.97 */
247			/* NeuQuant -30 \| 65 \| 6.53 */
248			/* Oct-Trees \| 141 \| 8.96 */
249			/* Median Cut (XV -best) \| 420 \| 9.28 */
250			/* Median Cut (XV -slow) \| 72 \| 12.15 */
251			/* */
252			/* Author's address: Dept of ISCS, National University of Singapore */
253			/* Kent Ridge, Singapore 0511 */
254			/* Email: [email protected] */
255			/----------------------------------------------------------------------/
256
257	greg	2.6	#define bool int
258			#define false 0
259			#define true 1
260	greg	2.1
261	greg	2.6	/* network defs */
262	greg	2.7	#define netsize clrtabsiz /* number of colours - can change this */
263	greg	2.6	#define maxnetpos (netsize-1)
264			#define netbiasshift 4 /* bias for colour values */
265			#define ncycles 100 /* no. of learning cycles */
266	greg	2.1
267			/* defs for freq and bias */
268	greg	2.6	#define intbiasshift 16 /* bias for fractions */
269			#define intbias (((int) 1)<<intbiasshift)
270			#define gammashift 10 /* gamma = 1024 */
271			#define gamma (((int) 1)<<gammashift)
272			#define betashift 10
273			#define beta (intbias>>betashift) /* beta = 1/1024 */
274	greg	2.1	#define betagamma (intbias<<(gammashift-betashift))
275
276	greg	2.6	/* defs for decreasing radius factor */
277	greg	2.7	#define initrad (256>>3) /* for 256 cols, radius starts */
278	greg	2.6	#define radiusbiasshift 6 /* at 32.0 biased by 6 bits */
279			#define radiusbias (((int) 1)<<radiusbiasshift)
280			#define initradius (initradradiusbias) / and decreases by a */
281			#define radiusdec 30 /* factor of 1/30 each cycle */
282
283			/* defs for decreasing alpha factor */
284			#define alphabiasshift 10 /* alpha starts at 1.0 */
285			#define initalpha (((int) 1)<<alphabiasshift)
286			int alphadec; /* biased by 10 bits */
287
288			/* radbias and alpharadbias used for radpower calculation */
289	greg	2.1	#define radbiasshift 8
290	greg	2.6	#define radbias (((int) 1)<<radbiasshift)
291	greg	2.1	#define alpharadbshift (alphabiasshift+radbiasshift)
292	greg	2.6	#define alpharadbias (((int) 1)<<alpharadbshift)
293	greg	2.1
294	greg	2.6	/* four primes near 500 - assume no image has a length so large */
295			/* that it is divisible by all four primes */
296			#define prime1 499
297			#define prime2 491
298			#define prime3 487
299			#define prime4 503
300	greg	2.1
301			typedef int pixel[4]; /* BGRc */
302	greg	2.7	pixel network[256];
303	greg	2.1
304	greg	2.6	int netindex[256]; /* for network lookup - really 256 */
305	greg	2.1
306	greg	2.7	int bias [256]; /* bias and freq arrays for learning */
307			int freq [256];
308	greg	2.6	int radpower[initrad]; /* radpower for precomputation */
309	greg	2.1
310
311	greg	2.6	/* initialise network in range (0,0,0) to (255,255,255) */
312	greg	2.1
313	greg	2.6	initnet()
314	greg	2.1	{
315			register int i;
316			register int *p;
317
318	greg	2.7	for (i=0; i<netsize; i++) {
319	greg	2.1	p = network[i];
320	greg	2.7	p[0] = p[1] = p[2] = (i << (netbiasshift+8))/netsize;
321			freq[i] = intbias/netsize; /* 1/netsize */
322	greg	2.1	bias[i] = 0;
323			}
324			}
325
326
327	greg	2.6	/* do after unbias - insertion sort of network and build netindex[0..255] */
328
329	greg	2.1	inxbuild()
330			{
331			register int i,j,smallpos,smallval;
332			register int p,q;
333	greg	2.6	int previouscol,startpos;
334	greg	2.1
335	greg	2.6	previouscol = 0;
336			startpos = 0;
337	greg	2.7	for (i=0; i<netsize; i++) {
338	greg	2.1	p = network[i];
339			smallpos = i;
340			smallval = p[1]; /* index on g */
341	greg	2.7	/* find smallest in i..netsize-1 */
342			for (j=i+1; j<netsize; j++) {
343	greg	2.1	q = network[j];
344			if (q[1] < smallval) { /* index on g */
345			smallpos = j;
346			smallval = q[1]; /* index on g */
347			}
348			}
349			q = network[smallpos];
350	greg	2.6	/* swap p (i) and q (smallpos) entries */
351	greg	2.1	if (i != smallpos) {
352			j = q[0]; q[0] = p[0]; p[0] = j;
353			j = q[1]; q[1] = p[1]; p[1] = j;
354			j = q[2]; q[2] = p[2]; p[2] = j;
355			j = q[3]; q[3] = p[3]; p[3] = j;
356			}
357			/* smallval entry is now in position i */
358	greg	2.6	if (smallval != previouscol) {
359			netindex[previouscol] = (startpos+i)>>1;
360			for (j=previouscol+1; j<smallval; j++) netindex[j] = i;
361			previouscol = smallval;
362			startpos = i;
363	greg	2.1	}
364			}
365	greg	2.6	netindex[previouscol] = (startpos+maxnetpos)>>1;
366			for (j=previouscol+1; j<256; j++) netindex[j] = maxnetpos; /* really 256 */
367	greg	2.1	}
368
369
370	greg	2.6	int inxsearch(b,g,r) /* accepts real BGR values after net is unbiased */
371	greg	2.1	register int b,g,r;
372			{
373	greg	2.6	register int i,j,dist,a,bestd;
374	greg	2.1	register int *p;
375	greg	2.6	int best;
376	greg	2.1
377			bestd = 1000; /* biggest possible dist is 2563 /
378			best = -1;
379			i = netindex[g]; /* index on g */
380	greg	2.6	j = i-1; /* start at netindex[g] and work outwards */
381	greg	2.1
382	greg	2.7	while ((i<netsize) \|\| (j>=0)) {
383			if (i<netsize) {
384	greg	2.1	p = network[i];
385	greg	2.6	dist = p[1] - g; /* inx key */
386	greg	2.7	if (dist >= bestd) i = netsize; /* stop iter */
387	greg	2.1	else {
388			i++;
389	greg	2.6	if (dist<0) dist = -dist;
390			a = p[0] - b; if (a<0) a = -a;
391			dist += a;
392			if (dist<bestd) {
393			a = p[2] - r; if (a<0) a = -a;
394			dist += a;
395			if (dist<bestd) {bestd=dist; best=p[3];}
396	greg	2.1	}
397			}
398			}
399			if (j>=0) {
400			p = network[j];
401	greg	2.6	dist = g - p[1]; /* inx key - reverse dif */
402			if (dist >= bestd) j = -1; /* stop iter */
403	greg	2.1	else {
404			j--;
405	greg	2.6	if (dist<0) dist = -dist;
406			a = p[0] - b; if (a<0) a = -a;
407			dist += a;
408			if (dist<bestd) {
409			a = p[2] - r; if (a<0) a = -a;
410			dist += a;
411			if (dist<bestd) {bestd=dist; best=p[3];}
412	greg	2.1	}
413			}
414			}
415			}
416			return(best);
417			}
418
419
420	greg	2.6	/* finds closest neuron (min dist) and updates freq */
421			/* finds best neuron (min dist-bias) and returns position */
422			/* for frequently chosen neurons, freq[i] is high and bias[i] is negative */
423	greg	2.7	/* bias[i] = gamma((1/netsize)-freq[i]) /
424	greg	2.6
425			int contest(b,g,r) /* accepts biased BGR values */
426	greg	2.1	register int b,g,r;
427			{
428	greg	2.6	register int i,dist,a,biasdist,betafreq;
429			int bestpos,bestbiaspos,bestd,bestbiasd;
430			register int p,f, *n;
431	greg	2.1
432	greg	2.6	bestd = ~(((int) 1)<<31);
433	greg	2.1	bestbiasd = bestd;
434	greg	2.6	bestpos = -1;
435			bestbiaspos = bestpos;
436			p = bias;
437			f = freq;
438
439	greg	2.7	for (i=0; i<netsize; i++) {
440	greg	2.6	n = network[i];
441			dist = n[0] - b; if (dist<0) dist = -dist;
442			a = n[1] - g; if (a<0) a = -a;
443			dist += a;
444			a = n[2] - r; if (a<0) a = -a;
445			dist += a;
446			if (dist<bestd) {bestd=dist; bestpos=i;}
447			biasdist = dist - ((*p)>>(intbiasshift-netbiasshift));
448			if (biasdist<bestbiasd) {bestbiasd=biasdist; bestbiaspos=i;}
449			betafreq = (*f >> betashift);
450			*f++ -= betafreq;
451			*p++ += (betafreq<<gammashift);
452	greg	2.1	}
453	greg	2.6	freq[bestpos] += beta;
454			bias[bestpos] -= betagamma;
455			return(bestbiaspos);
456	greg	2.1	}
457
458
459	greg	2.6	/* move neuron i towards (b,g,r) by factor alpha */
460
461			altersingle(alpha,i,b,g,r) /* accepts biased BGR values */
462			register int alpha,i,b,g,r;
463			{
464			register int *n;
465
466			n = network[i]; /* alter hit neuron */
467			n -= (alpha(*n - b)) / initalpha;
468			n++;
469			n -= (alpha(*n - g)) / initalpha;
470			n++;
471			n -= (alpha(*n - r)) / initalpha;
472			}
473
474
475			/* move neurons adjacent to i towards (b,g,r) by factor */
476			/* alpha(1-((i-j)^2/[r]^2)) precomputed as radpower[\|i-j\|]/
477
478			alterneigh(rad,i,b,g,r) /* accents biased BGR values */
479	greg	2.1	int rad,i;
480			register int b,g,r;
481			{
482			register int j,k,lo,hi,a;
483			register int p, q;
484
485	greg	2.6	lo = i-rad; if (lo<-1) lo= -1;
486	greg	2.7	hi = i+rad; if (hi>netsize) hi=netsize;
487	greg	2.1
488			j = i+1;
489			k = i-1;
490			q = radpower;
491			while ((j<hi) \|\| (k>lo)) {
492			a = (*(++q));
493			if (j<hi) {
494			p = network[j];
495			p -= (a(*p - b)) / alpharadbias;
496			p++;
497			p -= (a(*p - g)) / alpharadbias;
498			p++;
499			p -= (a(*p - r)) / alpharadbias;
500			j++;
501			}
502			if (k>lo) {
503			p = network[k];
504			p -= (a(*p - b)) / alpharadbias;
505			p++;
506			p -= (a(*p - g)) / alpharadbias;
507			p++;
508			p -= (a(*p - r)) / alpharadbias;
509			k--;
510			}
511			}
512			}
513
514
515			learn()
516			{
517			register int i,j,b,g,r;
518	greg	2.6	int radius,rad,alpha,step,delta,samplepixels;
519	greg	2.1	register unsigned char *p;
520			unsigned char *lim;
521
522	greg	2.6	alphadec = 30 + ((samplefac-1)/3);
523			p = thepicture;
524	greg	2.1	lim = thepicture + lengthcount;
525	greg	2.6	samplepixels = lengthcount/(3*samplefac);
526			delta = samplepixels/ncycles;
527	greg	2.1	alpha = initalpha;
528			radius = initradius;
529	greg	2.6
530	greg	2.1	rad = radius >> radiusbiasshift;
531			if (rad <= 1) rad = 0;
532			for (i=0; i<rad; i++)
533			radpower[i] = alpha(((radrad - ii)radbias)/(rad*rad));
534	greg	2.6
535			if ((lengthcount%prime1) != 0) step = 3*prime1;
536	greg	2.1	else {
537	greg	2.6	if ((lengthcount%prime2) !=0) step = 3*prime2;
538	greg	2.1	else {
539	greg	2.6	if ((lengthcount%prime3) !=0) step = 3*prime3;
540			else step = 3*prime4;
541	greg	2.1	}
542			}
543	greg	2.6
544	greg	2.1	i = 0;
545	greg	2.6	while (i < samplepixels) {
546	greg	2.1	b = p[0] << netbiasshift;
547			g = p[1] << netbiasshift;
548			r = p[2] << netbiasshift;
549			j = contest(b,g,r);
550
551			altersingle(alpha,j,b,g,r);
552	greg	2.6	if (rad) alterneigh(rad,j,b,g,r); /* alter neighbours */
553	greg	2.1
554			p += step;
555			if (p >= lim) p -= lengthcount;
556
557			i++;
558			if (i%delta == 0) {
559			alpha -= alpha / alphadec;
560			radius -= radius / radiusdec;
561			rad = radius >> radiusbiasshift;
562			if (rad <= 1) rad = 0;
563			for (j=0; j<rad; j++)
564			radpower[j] = alpha(((radrad - jj)radbias)/(rad*rad));
565			}
566			}
567			}
568
569	greg	2.6	/* unbias network to give 0..255 entries */
570			/* which can then be used for colour map */
571			/* and record position i to prepare for sort */
572
573	greg	2.1	unbiasnet()
574			{
575			int i,j;
576
577	greg	2.7	for (i=0; i<netsize; i++) {
578	greg	2.1	for (j=0; j<3; j++)
579			network[i][j] >>= netbiasshift;
580			network[i][3] = i; /* record colour no */
581			}
582			}
583
584	greg	2.6
585			/* Don't do this until the network has been unbiased (GW) */
586	greg	2.1
587			static
588			cpyclrtab()
589			{
590			register int i,j,k;
591
592	greg	2.7	for (j=0; j<netsize; j++) {
593	greg	2.1	k = network[j][3];
594			for (i = 0; i < 3; i++)
595			clrtab[k][i] = network[j][2-i];
596			}
597			}