| 1 |
– |
/* Copyright (c) 1994 Regents of the University of California */ |
| 2 |
– |
|
| 1 |
|
#ifndef lint |
| 2 |
< |
static char SCCSid[] = "$SunId$ LBL"; |
| 2 |
> |
static const char RCSid[] = "$Id$"; |
| 3 |
|
#endif |
| 6 |
– |
|
| 4 |
|
/* |
| 5 |
|
* Neural-Net quantization algorithm based on work of Anthony Dekker |
| 6 |
|
*/ |
| 48 |
|
|
| 49 |
|
#define setskip(sp,n) ((sp)[0]=(n)>>16,(sp)[1]=((n)>>8)&255,(sp)[2]=(n)&255) |
| 50 |
|
|
| 51 |
+ |
static cpyclrtab(); |
| 52 |
|
|
| 53 |
+ |
|
| 54 |
|
neu_init(npixels) /* initialize our sample array */ |
| 55 |
|
long npixels; |
| 56 |
|
{ |
| 131 |
|
cpyclrtab(); |
| 132 |
|
inxbuild(); |
| 133 |
|
/* we're done with our samples */ |
| 134 |
< |
free((char *)thesamples); |
| 134 |
> |
free((void *)thesamples); |
| 135 |
|
/* reset dithering function */ |
| 136 |
|
neu_dith_colrs((BYTE *)NULL, (COLR *)NULL, 0); |
| 137 |
|
/* return new color table size */ |
| 171 |
|
|
| 172 |
|
if (n != N) { /* get error propogation array */ |
| 173 |
|
if (N) { |
| 174 |
< |
free((char *)cerr); |
| 174 |
> |
free((void *)cerr); |
| 175 |
|
cerr = NULL; |
| 176 |
|
} |
| 177 |
|
if (n) |
| 256 |
|
#define true 1 |
| 257 |
|
|
| 258 |
|
/* network defs */ |
| 259 |
< |
#define netsize 256 /* number of colours - can change this */ |
| 259 |
> |
#define netsize clrtabsiz /* number of colours - can change this */ |
| 260 |
|
#define maxnetpos (netsize-1) |
| 261 |
|
#define netbiasshift 4 /* bias for colour values */ |
| 262 |
|
#define ncycles 100 /* no. of learning cycles */ |
| 271 |
|
#define betagamma (intbias<<(gammashift-betashift)) |
| 272 |
|
|
| 273 |
|
/* defs for decreasing radius factor */ |
| 274 |
< |
#define initrad (netsize>>3) /* for 256 cols, radius starts */ |
| 274 |
> |
#define initrad (256>>3) /* for 256 cols, radius starts */ |
| 275 |
|
#define radiusbiasshift 6 /* at 32.0 biased by 6 bits */ |
| 276 |
|
#define radiusbias (((int) 1)<<radiusbiasshift) |
| 277 |
|
#define initradius (initrad*radiusbias) /* and decreases by a */ |
| 296 |
|
#define prime4 503 |
| 297 |
|
|
| 298 |
|
typedef int pixel[4]; /* BGRc */ |
| 299 |
< |
pixel network[netsize]; |
| 299 |
> |
pixel network[256]; |
| 300 |
|
|
| 301 |
|
int netindex[256]; /* for network lookup - really 256 */ |
| 302 |
|
|
| 303 |
< |
int bias [netsize]; /* bias and freq arrays for learning */ |
| 304 |
< |
int freq [netsize]; |
| 303 |
> |
int bias [256]; /* bias and freq arrays for learning */ |
| 304 |
> |
int freq [256]; |
| 305 |
|
int radpower[initrad]; /* radpower for precomputation */ |
| 306 |
|
|
| 307 |
|
|
| 312 |
|
register int i; |
| 313 |
|
register int *p; |
| 314 |
|
|
| 315 |
< |
for (i=0; i<clrtabsiz; i++) { |
| 315 |
> |
for (i=0; i<netsize; i++) { |
| 316 |
|
p = network[i]; |
| 317 |
< |
p[0] = p[1] = p[2] = (i << (netbiasshift+8))/clrtabsiz; |
| 318 |
< |
freq[i] = intbias/clrtabsiz; /* 1/clrtabsiz */ |
| 317 |
> |
p[0] = p[1] = p[2] = (i << (netbiasshift+8))/netsize; |
| 318 |
> |
freq[i] = intbias/netsize; /* 1/netsize */ |
| 319 |
|
bias[i] = 0; |
| 320 |
|
} |
| 321 |
|
} |
| 331 |
|
|
| 332 |
|
previouscol = 0; |
| 333 |
|
startpos = 0; |
| 334 |
< |
for (i=0; i<clrtabsiz; i++) { |
| 334 |
> |
for (i=0; i<netsize; i++) { |
| 335 |
|
p = network[i]; |
| 336 |
|
smallpos = i; |
| 337 |
|
smallval = p[1]; /* index on g */ |
| 338 |
< |
/* find smallest in i..clrtabsiz-1 */ |
| 339 |
< |
for (j=i+1; j<clrtabsiz; j++) { |
| 338 |
> |
/* find smallest in i..netsize-1 */ |
| 339 |
> |
for (j=i+1; j<netsize; j++) { |
| 340 |
|
q = network[j]; |
| 341 |
|
if (q[1] < smallval) { /* index on g */ |
| 342 |
|
smallpos = j; |
| 376 |
|
i = netindex[g]; /* index on g */ |
| 377 |
|
j = i-1; /* start at netindex[g] and work outwards */ |
| 378 |
|
|
| 379 |
< |
while ((i<clrtabsiz) || (j>=0)) { |
| 380 |
< |
if (i<clrtabsiz) { |
| 379 |
> |
while ((i<netsize) || (j>=0)) { |
| 380 |
> |
if (i<netsize) { |
| 381 |
|
p = network[i]; |
| 382 |
|
dist = p[1] - g; /* inx key */ |
| 383 |
< |
if (dist >= bestd) i = clrtabsiz; /* stop iter */ |
| 383 |
> |
if (dist >= bestd) i = netsize; /* stop iter */ |
| 384 |
|
else { |
| 385 |
|
i++; |
| 386 |
|
if (dist<0) dist = -dist; |
| 417 |
|
/* finds closest neuron (min dist) and updates freq */ |
| 418 |
|
/* finds best neuron (min dist-bias) and returns position */ |
| 419 |
|
/* for frequently chosen neurons, freq[i] is high and bias[i] is negative */ |
| 420 |
< |
/* bias[i] = gamma*((1/clrtabsiz)-freq[i]) */ |
| 420 |
> |
/* bias[i] = gamma*((1/netsize)-freq[i]) */ |
| 421 |
|
|
| 422 |
|
int contest(b,g,r) /* accepts biased BGR values */ |
| 423 |
|
register int b,g,r; |
| 433 |
|
p = bias; |
| 434 |
|
f = freq; |
| 435 |
|
|
| 436 |
< |
for (i=0; i<clrtabsiz; i++) { |
| 436 |
> |
for (i=0; i<netsize; i++) { |
| 437 |
|
n = network[i]; |
| 438 |
|
dist = n[0] - b; if (dist<0) dist = -dist; |
| 439 |
|
a = n[1] - g; if (a<0) a = -a; |
| 480 |
|
register int *p, *q; |
| 481 |
|
|
| 482 |
|
lo = i-rad; if (lo<-1) lo= -1; |
| 483 |
< |
hi = i+rad; if (hi>clrtabsiz) hi=clrtabsiz; |
| 483 |
> |
hi = i+rad; if (hi>netsize) hi=netsize; |
| 484 |
|
|
| 485 |
|
j = i+1; |
| 486 |
|
k = i-1; |
| 571 |
|
{ |
| 572 |
|
int i,j; |
| 573 |
|
|
| 574 |
< |
for (i=0; i<clrtabsiz; i++) { |
| 574 |
> |
for (i=0; i<netsize; i++) { |
| 575 |
|
for (j=0; j<3; j++) |
| 576 |
|
network[i][j] >>= netbiasshift; |
| 577 |
|
network[i][3] = i; /* record colour no */ |
| 586 |
|
{ |
| 587 |
|
register int i,j,k; |
| 588 |
|
|
| 589 |
< |
for (j=0; j<clrtabsiz; j++) { |
| 589 |
> |
for (j=0; j<netsize; j++) { |
| 590 |
|
k = network[j][3]; |
| 591 |
|
for (i = 0; i < 3; i++) |
| 592 |
|
clrtab[k][i] = network[j][2-i]; |
