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