src/px/neuclrtab.c

#ifndef lint
static const char       RCSid[] = "$Id: neuclrtab.c,v 2.11 2004/03/28 20:33:14 schorsch 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"
#include "clrtab.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
#define DEFSMPFAC       3
#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 void initnet(void);
static void inxbuild(void);
static int inxsearch(int b, int g, int r);
static int contest(int b, int g, int r);
static void altersingle(int alpha, int i, int b, int g, int r);
static void alterneigh(int rad, int i, int b, int g, int r);
static void learn(void);
static void unbiasnet(void);
static void cpyclrtab(void);


extern int
neu_init(               /* 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);
}


extern void
neu_pixel(                      /* 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;
        }
}


extern void
neu_colrs(              /* 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;
}


extern int
neu_clrtab(             /* 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);
}


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


extern void
neu_map_colrs(  /* 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++;
        }
}


extern void
neu_dith_colrs( /* 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) */

static void
initnet(void)   
{
        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] */

static void
inxbuild(void)
{
        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 */
}


static int
inxsearch(  /* accepts real BGR values after net is unbiased */
        register int b,
        register int g,
        register int 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]) */

static int
contest(        /* accepts biased BGR values */
        register int b,
        register int g,
        register int 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 */

static void
altersingle(    /* accepts biased BGR values */
        register int alpha,
        register int i,
        register int b,
        register int g,
        register int 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|]*/

static void
alterneigh(     /* accents biased BGR values */
        int rad,
        int i,
        register int b,
        register int g,
        register int 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--;
                }
        }
}


static void
learn(void)
{
        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 */

static void
unbiasnet(void)
{
        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 void
cpyclrtab(void)
{
        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:	2.12
Committed:	Mon Sep 19 02:23:58 2005 UTC (18 years, 7 months ago) by greg
Content type:	text/plain
Branch:	MAIN
CVS Tags:	rad3R8
Changes since 2.11:	+2 -6 lines
Log Message:	Eliminated SPEED macro from makeall and source tree
#	User	Rev	Content
1	greg	2.1	#ifndef lint
2	greg	2.12	static const char RCSid[] = "$Id: neuclrtab.c,v 2.11 2004/03/28 20:33:14 schorsch 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	schorsch	2.11	#include "clrtab.h"
16	greg	2.1
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	greg	2.12	#define DEFSMPFAC 3
32	greg	2.1	#endif
33
34			int samplefac = DEFSMPFAC; /* sampling factor */
35
36			/* Samples array starts off holding spacing between adjacent
37			* samples, and ends up holding actual BGR sample values.
38			*/
39			static BYTE *thesamples;
40			static int nsamples;
41			static BYTE *cursamp;
42			static long skipcount;
43
44			#define MAXSKIP (1<<24-1)
45
46			#define nskip(sp) ((long)(sp)[0]<<16\|(long)(sp)[1]<<8\|(long)(sp)[2])
47
48			#define setskip(sp,n) ((sp)[0]=(n)>>16,(sp)[1]=((n)>>8)&255,(sp)[2]=(n)&255)
49
50	schorsch	2.11	static void initnet(void);
51			static void inxbuild(void);
52			static int inxsearch(int b, int g, int r);
53			static int contest(int b, int g, int r);
54			static void altersingle(int alpha, int i, int b, int g, int r);
55			static void alterneigh(int rad, int i, int b, int g, int r);
56			static void learn(void);
57			static void unbiasnet(void);
58			static void cpyclrtab(void);
59	greg	2.8
60	greg	2.1
61	schorsch	2.11	extern int
62			neu_init( /* initialize our sample array */
63			long npixels
64			)
65	greg	2.1	{
66			register int nsleft;
67			register long sv;
68			double rval, cumprob;
69			long npleft;
70
71			nsamples = npixels/samplefac;
72			if (nsamples < 600)
73			return(-1);
74	greg	2.2	thesamples = (BYTE )malloc(nsamples3);
75	greg	2.1	if (thesamples == NULL)
76			return(-1);
77			cursamp = thesamples;
78			npleft = npixels;
79			nsleft = nsamples;
80			while (nsleft) {
81			rval = frandom(); /* random distance to next sample */
82			sv = 0;
83			cumprob = 0.;
84			while ((cumprob += (1.-cumprob)*nsleft/(npleft-sv)) < rval)
85			sv++;
86	greg	2.2	if (nsleft == nsamples)
87			skipcount = sv;
88			else {
89			setskip(cursamp, sv);
90			cursamp += 3;
91			}
92			npleft -= sv+1;
93	greg	2.1	nsleft--;
94			}
95	greg	2.2	setskip(cursamp, npleft); /* tag on end to skip the rest */
96	greg	2.1	cursamp = thesamples;
97			return(0);
98			}
99
100
101	schorsch	2.11	extern void
102			neu_pixel( /* add pixel to our samples */
103			register BYTE col[]
104			)
105	greg	2.1	{
106			if (!skipcount--) {
107	greg	2.2	skipcount = nskip(cursamp);
108	greg	2.1	cursamp[0] = col[BLU];
109			cursamp[1] = col[GRN];
110			cursamp[2] = col[RED];
111			cursamp += 3;
112			}
113			}
114
115
116	schorsch	2.11	extern void
117			neu_colrs( /* add a scanline to our samples */
118			register COLR *cs,
119			register int n
120			)
121	greg	2.1	{
122			while (n > skipcount) {
123			cs += skipcount;
124	greg	2.2	n -= skipcount+1;
125			skipcount = nskip(cursamp);
126	greg	2.1	cursamp[0] = cs[0][BLU];
127			cursamp[1] = cs[0][GRN];
128			cursamp[2] = cs[0][RED];
129			cs++;
130			cursamp += 3;
131			}
132			skipcount -= n;
133			}
134
135
136	schorsch	2.11	extern int
137			neu_clrtab( /* make new color table using ncolors */
138			int ncolors
139			)
140	greg	2.1	{
141			clrtabsiz = ncolors;
142			if (clrtabsiz > 256) clrtabsiz = 256;
143			initnet();
144			learn();
145			unbiasnet();
146			cpyclrtab();
147			inxbuild();
148			/* we're done with our samples */
149	greg	2.9	free((void *)thesamples);
150	greg	2.1	/* reset dithering function */
151			neu_dith_colrs((BYTE )NULL, (COLR )NULL, 0);
152			/* return new color table size */
153			return(clrtabsiz);
154			}
155
156
157	schorsch	2.11	extern int
158			neu_map_pixel( /* get pixel for color */
159			register BYTE col[]
160			)
161	greg	2.1	{
162			return(inxsearch(col[BLU],col[GRN],col[RED]));
163			}
164
165
166	schorsch	2.11	extern void
167			neu_map_colrs( /* convert a scanline to color index values */
168			register BYTE *bs,
169			register COLR *cs,
170			register int n
171			)
172	greg	2.1	{
173			while (n-- > 0) {
174			*bs++ = inxsearch(cs[0][BLU],cs[0][GRN],cs[0][RED]);
175			cs++;
176			}
177			}
178
179
180	schorsch	2.11	extern void
181			neu_dith_colrs( /* convert scanline to dithered index values */
182			register BYTE *bs,
183			register COLR *cs,
184			int n
185			)
186	greg	2.1	{
187			static short (*cerr)[3] = NULL;
188			static int N = 0;
189			int err[3], errp[3];
190			register int x, i;
191
192			if (n != N) { /* get error propogation array */
193			if (N) {
194	greg	2.9	free((void *)cerr);
195	greg	2.1	cerr = NULL;
196			}
197			if (n)
198			cerr = (short ()[3])malloc(3n*sizeof(short));
199			if (cerr == NULL) {
200			N = 0;
201			map_colrs(bs, cs, n);
202			return;
203			}
204			N = n;
205	schorsch	2.10	memset((char )cerr, '\0', 3N*sizeof(short));
206	greg	2.1	}
207			err[0] = err[1] = err[2] = 0;
208			for (x = 0; x < n; x++) {
209			for (i = 0; i < 3; i++) { /* dither value */
210			errp[i] = err[i];
211			err[i] += cerr[x][i];
212			#ifdef MAXERR
213			if (err[i] > MAXERR) err[i] = MAXERR;
214			else if (err[i] < -MAXERR) err[i] = -MAXERR;
215			#endif
216			err[i] += cs[x][i];
217			if (err[i] < 0) err[i] = 0;
218			else if (err[i] > 255) err[i] = 255;
219			}
220			bs[x] = inxsearch(err[BLU],err[GRN],err[RED]);
221			for (i = 0; i < 3; i++) { /* propagate error */
222			err[i] -= clrtab[bs[x]][i];
223			err[i] /= 3;
224			cerr[x][i] = err[i] + errp[i];
225			}
226			}
227			}
228
229			/* The following was adapted and modified from the original (GW) */
230	greg	2.6
231			/* cheater definitions (GW) */
232			#define thepicture thesamples
233			#define lengthcount (nsamples*3)
234			#define samplefac 1
235
236	greg	2.1	/----------------------------------------------------------------------/
237			/* */
238			/* NeuQuant */
239			/* -------- */
240			/* */
241	greg	2.6	/* Copyright: Anthony Dekker, November 1994 */
242	greg	2.1	/* */
243			/* This program performs colour quantization of graphics images (SUN */
244			/* raster files). It uses a Kohonen Neural Network. It produces */
245			/* better results than existing methods and runs faster, using minimal */
246			/* space (8kB plus the image itself). The algorithm is described in */
247			/* the paper "Kohonen Neural Networks for Optimal Colour Quantization" */
248			/* to appear in the journal "Network: Computation in Neural Systems". */
249			/* It is a significant improvement of an earlier algorithm. */
250			/* */
251			/* This program is distributed free for academic use or for evaluation */
252			/* by commercial organizations. */
253			/* */
254			/* Usage: NeuQuant -n inputfile > outputfile */
255			/* */
256			/* where n is a sampling factor for neural learning. */
257			/* */
258			/* Program performance compared with other methods is as follows: */
259			/* */
260			/* Algorithm \| Av. CPU Time \| Quantization Error */
261			/* ------------------------------------------------------------- */
262			/* NeuQuant -3 \| 314 \| 5.55 */
263			/* NeuQuant -10 \| 119 \| 5.97 */
264			/* NeuQuant -30 \| 65 \| 6.53 */
265			/* Oct-Trees \| 141 \| 8.96 */
266			/* Median Cut (XV -best) \| 420 \| 9.28 */
267			/* Median Cut (XV -slow) \| 72 \| 12.15 */
268			/* */
269			/* Author's address: Dept of ISCS, National University of Singapore */
270			/* Kent Ridge, Singapore 0511 */
271			/* Email: [email protected] */
272			/----------------------------------------------------------------------/
273
274	greg	2.6	#define bool int
275			#define false 0
276			#define true 1
277	greg	2.1
278	greg	2.6	/* network defs */
279	greg	2.7	#define netsize clrtabsiz /* number of colours - can change this */
280	greg	2.6	#define maxnetpos (netsize-1)
281			#define netbiasshift 4 /* bias for colour values */
282			#define ncycles 100 /* no. of learning cycles */
283	greg	2.1
284			/* defs for freq and bias */
285	greg	2.6	#define intbiasshift 16 /* bias for fractions */
286			#define intbias (((int) 1)<<intbiasshift)
287			#define gammashift 10 /* gamma = 1024 */
288			#define gamma (((int) 1)<<gammashift)
289			#define betashift 10
290			#define beta (intbias>>betashift) /* beta = 1/1024 */
291	greg	2.1	#define betagamma (intbias<<(gammashift-betashift))
292
293	greg	2.6	/* defs for decreasing radius factor */
294	greg	2.7	#define initrad (256>>3) /* for 256 cols, radius starts */
295	greg	2.6	#define radiusbiasshift 6 /* at 32.0 biased by 6 bits */
296			#define radiusbias (((int) 1)<<radiusbiasshift)
297			#define initradius (initradradiusbias) / and decreases by a */
298			#define radiusdec 30 /* factor of 1/30 each cycle */
299
300			/* defs for decreasing alpha factor */
301			#define alphabiasshift 10 /* alpha starts at 1.0 */
302			#define initalpha (((int) 1)<<alphabiasshift)
303			int alphadec; /* biased by 10 bits */
304
305			/* radbias and alpharadbias used for radpower calculation */
306	greg	2.1	#define radbiasshift 8
307	greg	2.6	#define radbias (((int) 1)<<radbiasshift)
308	greg	2.1	#define alpharadbshift (alphabiasshift+radbiasshift)
309	greg	2.6	#define alpharadbias (((int) 1)<<alpharadbshift)
310	greg	2.1
311	greg	2.6	/* four primes near 500 - assume no image has a length so large */
312			/* that it is divisible by all four primes */
313			#define prime1 499
314			#define prime2 491
315			#define prime3 487
316			#define prime4 503
317	greg	2.1
318			typedef int pixel[4]; /* BGRc */
319	greg	2.7	pixel network[256];
320	greg	2.1
321	greg	2.6	int netindex[256]; /* for network lookup - really 256 */
322	greg	2.1
323	greg	2.7	int bias [256]; /* bias and freq arrays for learning */
324			int freq [256];
325	greg	2.6	int radpower[initrad]; /* radpower for precomputation */
326	greg	2.1
327
328	greg	2.6	/* initialise network in range (0,0,0) to (255,255,255) */
329	greg	2.1
330	schorsch	2.11	static void
331			initnet(void)
332	greg	2.1	{
333			register int i;
334			register int *p;
335
336	greg	2.7	for (i=0; i<netsize; i++) {
337	greg	2.1	p = network[i];
338	greg	2.7	p[0] = p[1] = p[2] = (i << (netbiasshift+8))/netsize;
339			freq[i] = intbias/netsize; /* 1/netsize */
340	greg	2.1	bias[i] = 0;
341			}
342			}
343
344
345	greg	2.6	/* do after unbias - insertion sort of network and build netindex[0..255] */
346
347	schorsch	2.11	static void
348			inxbuild(void)
349	greg	2.1	{
350			register int i,j,smallpos,smallval;
351			register int p,q;
352	greg	2.6	int previouscol,startpos;
353	greg	2.1
354	greg	2.6	previouscol = 0;
355			startpos = 0;
356	greg	2.7	for (i=0; i<netsize; i++) {
357	greg	2.1	p = network[i];
358			smallpos = i;
359			smallval = p[1]; /* index on g */
360	greg	2.7	/* find smallest in i..netsize-1 */
361			for (j=i+1; j<netsize; j++) {
362	greg	2.1	q = network[j];
363			if (q[1] < smallval) { /* index on g */
364			smallpos = j;
365			smallval = q[1]; /* index on g */
366			}
367			}
368			q = network[smallpos];
369	greg	2.6	/* swap p (i) and q (smallpos) entries */
370	greg	2.1	if (i != smallpos) {
371			j = q[0]; q[0] = p[0]; p[0] = j;
372			j = q[1]; q[1] = p[1]; p[1] = j;
373			j = q[2]; q[2] = p[2]; p[2] = j;
374			j = q[3]; q[3] = p[3]; p[3] = j;
375			}
376			/* smallval entry is now in position i */
377	greg	2.6	if (smallval != previouscol) {
378			netindex[previouscol] = (startpos+i)>>1;
379			for (j=previouscol+1; j<smallval; j++) netindex[j] = i;
380			previouscol = smallval;
381			startpos = i;
382	greg	2.1	}
383			}
384	greg	2.6	netindex[previouscol] = (startpos+maxnetpos)>>1;
385			for (j=previouscol+1; j<256; j++) netindex[j] = maxnetpos; /* really 256 */
386	greg	2.1	}
387
388
389	schorsch	2.11	static int
390			inxsearch( /* accepts real BGR values after net is unbiased */
391			register int b,
392			register int g,
393			register int r
394			)
395	greg	2.1	{
396	greg	2.6	register int i,j,dist,a,bestd;
397	greg	2.1	register int *p;
398	greg	2.6	int best;
399	greg	2.1
400			bestd = 1000; /* biggest possible dist is 2563 /
401			best = -1;
402			i = netindex[g]; /* index on g */
403	greg	2.6	j = i-1; /* start at netindex[g] and work outwards */
404	greg	2.1
405	greg	2.7	while ((i<netsize) \|\| (j>=0)) {
406			if (i<netsize) {
407	greg	2.1	p = network[i];
408	greg	2.6	dist = p[1] - g; /* inx key */
409	greg	2.7	if (dist >= bestd) i = netsize; /* stop iter */
410	greg	2.1	else {
411			i++;
412	greg	2.6	if (dist<0) dist = -dist;
413			a = p[0] - b; if (a<0) a = -a;
414			dist += a;
415			if (dist<bestd) {
416			a = p[2] - r; if (a<0) a = -a;
417			dist += a;
418			if (dist<bestd) {bestd=dist; best=p[3];}
419	greg	2.1	}
420			}
421			}
422			if (j>=0) {
423			p = network[j];
424	greg	2.6	dist = g - p[1]; /* inx key - reverse dif */
425			if (dist >= bestd) j = -1; /* stop iter */
426	greg	2.1	else {
427			j--;
428	greg	2.6	if (dist<0) dist = -dist;
429			a = p[0] - b; if (a<0) a = -a;
430			dist += a;
431			if (dist<bestd) {
432			a = p[2] - r; if (a<0) a = -a;
433			dist += a;
434			if (dist<bestd) {bestd=dist; best=p[3];}
435	greg	2.1	}
436			}
437			}
438			}
439			return(best);
440			}
441
442
443	greg	2.6	/* finds closest neuron (min dist) and updates freq */
444			/* finds best neuron (min dist-bias) and returns position */
445			/* for frequently chosen neurons, freq[i] is high and bias[i] is negative */
446	greg	2.7	/* bias[i] = gamma((1/netsize)-freq[i]) /
447	greg	2.6
448	schorsch	2.11	static int
449			contest( /* accepts biased BGR values */
450			register int b,
451			register int g,
452			register int r
453			)
454	greg	2.1	{
455	greg	2.6	register int i,dist,a,biasdist,betafreq;
456			int bestpos,bestbiaspos,bestd,bestbiasd;
457			register int p,f, *n;
458	greg	2.1
459	greg	2.6	bestd = ~(((int) 1)<<31);
460	greg	2.1	bestbiasd = bestd;
461	greg	2.6	bestpos = -1;
462			bestbiaspos = bestpos;
463			p = bias;
464			f = freq;
465
466	greg	2.7	for (i=0; i<netsize; i++) {
467	greg	2.6	n = network[i];
468			dist = n[0] - b; if (dist<0) dist = -dist;
469			a = n[1] - g; if (a<0) a = -a;
470			dist += a;
471			a = n[2] - r; if (a<0) a = -a;
472			dist += a;
473			if (dist<bestd) {bestd=dist; bestpos=i;}
474			biasdist = dist - ((*p)>>(intbiasshift-netbiasshift));
475			if (biasdist<bestbiasd) {bestbiasd=biasdist; bestbiaspos=i;}
476			betafreq = (*f >> betashift);
477			*f++ -= betafreq;
478			*p++ += (betafreq<<gammashift);
479	greg	2.1	}
480	greg	2.6	freq[bestpos] += beta;
481			bias[bestpos] -= betagamma;
482			return(bestbiaspos);
483	greg	2.1	}
484
485
486	greg	2.6	/* move neuron i towards (b,g,r) by factor alpha */
487
488	schorsch	2.11	static void
489			altersingle( /* accepts biased BGR values */
490			register int alpha,
491			register int i,
492			register int b,
493			register int g,
494			register int r
495			)
496	greg	2.6	{
497			register int *n;
498
499			n = network[i]; /* alter hit neuron */
500			n -= (alpha(*n - b)) / initalpha;
501			n++;
502			n -= (alpha(*n - g)) / initalpha;
503			n++;
504			n -= (alpha(*n - r)) / initalpha;
505			}
506
507
508			/* move neurons adjacent to i towards (b,g,r) by factor */
509			/* alpha(1-((i-j)^2/[r]^2)) precomputed as radpower[\|i-j\|]/
510
511	schorsch	2.11	static void
512			alterneigh( /* accents biased BGR values */
513			int rad,
514			int i,
515			register int b,
516			register int g,
517			register int r
518			)
519	greg	2.1	{
520			register int j,k,lo,hi,a;
521			register int p, q;
522
523	greg	2.6	lo = i-rad; if (lo<-1) lo= -1;
524	greg	2.7	hi = i+rad; if (hi>netsize) hi=netsize;
525	greg	2.1
526			j = i+1;
527			k = i-1;
528			q = radpower;
529			while ((j<hi) \|\| (k>lo)) {
530			a = (*(++q));
531			if (j<hi) {
532			p = network[j];
533			p -= (a(*p - b)) / alpharadbias;
534			p++;
535			p -= (a(*p - g)) / alpharadbias;
536			p++;
537			p -= (a(*p - r)) / alpharadbias;
538			j++;
539			}
540			if (k>lo) {
541			p = network[k];
542			p -= (a(*p - b)) / alpharadbias;
543			p++;
544			p -= (a(*p - g)) / alpharadbias;
545			p++;
546			p -= (a(*p - r)) / alpharadbias;
547			k--;
548			}
549			}
550			}
551
552
553	schorsch	2.11	static void
554			learn(void)
555	greg	2.1	{
556			register int i,j,b,g,r;
557	greg	2.6	int radius,rad,alpha,step,delta,samplepixels;
558	greg	2.1	register unsigned char *p;
559			unsigned char *lim;
560
561	greg	2.6	alphadec = 30 + ((samplefac-1)/3);
562			p = thepicture;
563	greg	2.1	lim = thepicture + lengthcount;
564	greg	2.6	samplepixels = lengthcount/(3*samplefac);
565			delta = samplepixels/ncycles;
566	greg	2.1	alpha = initalpha;
567			radius = initradius;
568	greg	2.6
569	greg	2.1	rad = radius >> radiusbiasshift;
570			if (rad <= 1) rad = 0;
571			for (i=0; i<rad; i++)
572			radpower[i] = alpha(((radrad - ii)radbias)/(rad*rad));
573	greg	2.6
574			if ((lengthcount%prime1) != 0) step = 3*prime1;
575	greg	2.1	else {
576	greg	2.6	if ((lengthcount%prime2) !=0) step = 3*prime2;
577	greg	2.1	else {
578	greg	2.6	if ((lengthcount%prime3) !=0) step = 3*prime3;
579			else step = 3*prime4;
580	greg	2.1	}
581			}
582	greg	2.6
583	greg	2.1	i = 0;
584	greg	2.6	while (i < samplepixels) {
585	greg	2.1	b = p[0] << netbiasshift;
586			g = p[1] << netbiasshift;
587			r = p[2] << netbiasshift;
588			j = contest(b,g,r);
589
590			altersingle(alpha,j,b,g,r);
591	greg	2.6	if (rad) alterneigh(rad,j,b,g,r); /* alter neighbours */
592	greg	2.1
593			p += step;
594			if (p >= lim) p -= lengthcount;
595
596			i++;
597			if (i%delta == 0) {
598			alpha -= alpha / alphadec;
599			radius -= radius / radiusdec;
600			rad = radius >> radiusbiasshift;
601			if (rad <= 1) rad = 0;
602			for (j=0; j<rad; j++)
603			radpower[j] = alpha(((radrad - jj)radbias)/(rad*rad));
604			}
605			}
606			}
607
608	greg	2.6	/* unbias network to give 0..255 entries */
609			/* which can then be used for colour map */
610			/* and record position i to prepare for sort */
611
612	schorsch	2.11	static void
613			unbiasnet(void)
614	greg	2.1	{
615			int i,j;
616
617	greg	2.7	for (i=0; i<netsize; i++) {
618	greg	2.1	for (j=0; j<3; j++)
619			network[i][j] >>= netbiasshift;
620			network[i][3] = i; /* record colour no */
621			}
622			}
623
624	greg	2.6
625			/* Don't do this until the network has been unbiased (GW) */
626	greg	2.1
627	schorsch	2.11	static void
628			cpyclrtab(void)
629	greg	2.1	{
630			register int i,j,k;
631
632	greg	2.7	for (j=0; j<netsize; j++) {
633	greg	2.1	k = network[j][3];
634			for (i = 0; i < 3; i++)
635			clrtab[k][i] = network[j][2-i];
636			}
637			}