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 |
|
*/ |
7 |
|
|
8 |
< |
#include "standard.h" |
8 |
> |
#include "copyright.h" |
9 |
|
|
10 |
< |
#include "color.h" |
10 |
> |
#include <string.h> |
11 |
|
|
12 |
+ |
#include "standard.h" |
13 |
+ |
#include "color.h" |
14 |
|
#include "random.h" |
15 |
+ |
#include "clrtab.h" |
16 |
|
|
17 |
|
#ifdef COMPAT_MODE |
18 |
|
#define neu_init new_histo |
51 |
|
|
52 |
|
#define setskip(sp,n) ((sp)[0]=(n)>>16,(sp)[1]=((n)>>8)&255,(sp)[2]=(n)&255) |
53 |
|
|
54 |
+ |
static void initnet(void); |
55 |
+ |
static void inxbuild(void); |
56 |
+ |
static int inxsearch(int b, int g, int r); |
57 |
+ |
static int contest(int b, int g, int r); |
58 |
+ |
static void altersingle(int alpha, int i, int b, int g, int r); |
59 |
+ |
static void alterneigh(int rad, int i, int b, int g, int r); |
60 |
+ |
static void learn(void); |
61 |
+ |
static void unbiasnet(void); |
62 |
+ |
static void cpyclrtab(void); |
63 |
|
|
64 |
< |
neu_init(npixels) /* initialize our sample array */ |
65 |
< |
long npixels; |
64 |
> |
|
65 |
> |
extern int |
66 |
> |
neu_init( /* initialize our sample array */ |
67 |
> |
long npixels |
68 |
> |
) |
69 |
|
{ |
70 |
|
register int nsleft; |
71 |
|
register long sv; |
102 |
|
} |
103 |
|
|
104 |
|
|
105 |
< |
neu_pixel(col) /* add pixel to our samples */ |
106 |
< |
register BYTE col[]; |
105 |
> |
extern void |
106 |
> |
neu_pixel( /* add pixel to our samples */ |
107 |
> |
register BYTE col[] |
108 |
> |
) |
109 |
|
{ |
110 |
|
if (!skipcount--) { |
111 |
|
skipcount = nskip(cursamp); |
117 |
|
} |
118 |
|
|
119 |
|
|
120 |
< |
neu_colrs(cs, n) /* add a scanline to our samples */ |
121 |
< |
register COLR *cs; |
122 |
< |
register int n; |
120 |
> |
extern void |
121 |
> |
neu_colrs( /* add a scanline to our samples */ |
122 |
> |
register COLR *cs, |
123 |
> |
register int n |
124 |
> |
) |
125 |
|
{ |
126 |
|
while (n > skipcount) { |
127 |
|
cs += skipcount; |
137 |
|
} |
138 |
|
|
139 |
|
|
140 |
< |
neu_clrtab(ncolors) /* make new color table using ncolors */ |
141 |
< |
int ncolors; |
140 |
> |
extern int |
141 |
> |
neu_clrtab( /* make new color table using ncolors */ |
142 |
> |
int ncolors |
143 |
> |
) |
144 |
|
{ |
145 |
|
clrtabsiz = ncolors; |
146 |
|
if (clrtabsiz > 256) clrtabsiz = 256; |
150 |
|
cpyclrtab(); |
151 |
|
inxbuild(); |
152 |
|
/* we're done with our samples */ |
153 |
< |
free((char *)thesamples); |
153 |
> |
free((void *)thesamples); |
154 |
|
/* reset dithering function */ |
155 |
|
neu_dith_colrs((BYTE *)NULL, (COLR *)NULL, 0); |
156 |
|
/* return new color table size */ |
158 |
|
} |
159 |
|
|
160 |
|
|
161 |
< |
int |
162 |
< |
neu_map_pixel(col) /* get pixel for color */ |
163 |
< |
register BYTE col[]; |
161 |
> |
extern int |
162 |
> |
neu_map_pixel( /* get pixel for color */ |
163 |
> |
register BYTE col[] |
164 |
> |
) |
165 |
|
{ |
166 |
|
return(inxsearch(col[BLU],col[GRN],col[RED])); |
167 |
|
} |
168 |
|
|
169 |
|
|
170 |
< |
neu_map_colrs(bs, cs, n) /* convert a scanline to color index values */ |
171 |
< |
register BYTE *bs; |
172 |
< |
register COLR *cs; |
173 |
< |
register int n; |
170 |
> |
extern void |
171 |
> |
neu_map_colrs( /* convert a scanline to color index values */ |
172 |
> |
register BYTE *bs, |
173 |
> |
register COLR *cs, |
174 |
> |
register int n |
175 |
> |
) |
176 |
|
{ |
177 |
|
while (n-- > 0) { |
178 |
|
*bs++ = inxsearch(cs[0][BLU],cs[0][GRN],cs[0][RED]); |
181 |
|
} |
182 |
|
|
183 |
|
|
184 |
< |
neu_dith_colrs(bs, cs, n) /* convert scanline to dithered index values */ |
185 |
< |
register BYTE *bs; |
186 |
< |
register COLR *cs; |
187 |
< |
int n; |
184 |
> |
extern void |
185 |
> |
neu_dith_colrs( /* convert scanline to dithered index values */ |
186 |
> |
register BYTE *bs, |
187 |
> |
register COLR *cs, |
188 |
> |
int n |
189 |
> |
) |
190 |
|
{ |
191 |
|
static short (*cerr)[3] = NULL; |
192 |
|
static int N = 0; |
195 |
|
|
196 |
|
if (n != N) { /* get error propogation array */ |
197 |
|
if (N) { |
198 |
< |
free((char *)cerr); |
198 |
> |
free((void *)cerr); |
199 |
|
cerr = NULL; |
200 |
|
} |
201 |
|
if (n) |
206 |
|
return; |
207 |
|
} |
208 |
|
N = n; |
209 |
< |
bzero((char *)cerr, 3*N*sizeof(short)); |
209 |
> |
memset((char *)cerr, '\0', 3*N*sizeof(short)); |
210 |
|
} |
211 |
|
err[0] = err[1] = err[2] = 0; |
212 |
|
for (x = 0; x < n; x++) { |
231 |
|
} |
232 |
|
|
233 |
|
/* The following was adapted and modified from the original (GW) */ |
234 |
+ |
|
235 |
+ |
/* cheater definitions (GW) */ |
236 |
+ |
#define thepicture thesamples |
237 |
+ |
#define lengthcount (nsamples*3) |
238 |
+ |
#define samplefac 1 |
239 |
+ |
|
240 |
|
/*----------------------------------------------------------------------*/ |
241 |
|
/* */ |
242 |
|
/* NeuQuant */ |
243 |
|
/* -------- */ |
244 |
|
/* */ |
245 |
< |
/* Copyright: Anthony Dekker, June 1994 */ |
245 |
> |
/* Copyright: Anthony Dekker, November 1994 */ |
246 |
|
/* */ |
247 |
|
/* This program performs colour quantization of graphics images (SUN */ |
248 |
|
/* raster files). It uses a Kohonen Neural Network. It produces */ |
275 |
|
/* Email: [email protected] */ |
276 |
|
/*----------------------------------------------------------------------*/ |
277 |
|
|
278 |
< |
#define bool int |
279 |
< |
#define false 0 |
280 |
< |
#define true 1 |
278 |
> |
#define bool int |
279 |
> |
#define false 0 |
280 |
> |
#define true 1 |
281 |
|
|
282 |
< |
#define initrad 32 |
283 |
< |
#define radiusdec 30 |
284 |
< |
#define alphadec; 30 |
282 |
> |
/* network defs */ |
283 |
> |
#define netsize clrtabsiz /* number of colours - can change this */ |
284 |
> |
#define maxnetpos (netsize-1) |
285 |
> |
#define netbiasshift 4 /* bias for colour values */ |
286 |
> |
#define ncycles 100 /* no. of learning cycles */ |
287 |
|
|
288 |
|
/* defs for freq and bias */ |
289 |
< |
#define gammashift 10 |
290 |
< |
#define betashift gammashift |
291 |
< |
#define intbiasshift 16 |
292 |
< |
#define intbias (1<<intbiasshift) |
293 |
< |
#define gamma (1<<gammashift) |
294 |
< |
#define beta (intbias>>betashift) |
289 |
> |
#define intbiasshift 16 /* bias for fractions */ |
290 |
> |
#define intbias (((int) 1)<<intbiasshift) |
291 |
> |
#define gammashift 10 /* gamma = 1024 */ |
292 |
> |
#define gamma (((int) 1)<<gammashift) |
293 |
> |
#define betashift 10 |
294 |
> |
#define beta (intbias>>betashift) /* beta = 1/1024 */ |
295 |
|
#define betagamma (intbias<<(gammashift-betashift)) |
265 |
– |
#define gammaphi (intbias<<(gammashift-8)) |
296 |
|
|
297 |
< |
/* defs for rad and alpha */ |
298 |
< |
#define maxrad (initrad+1) |
299 |
< |
#define radiusbiasshift 6 |
300 |
< |
#define radiusbias (1<<radiusbiasshift) |
301 |
< |
#define initradius ((int) (initrad*radiusbias)) |
302 |
< |
#define alphabiasshift 10 |
303 |
< |
#define initalpha (1<<alphabiasshift) |
297 |
> |
/* defs for decreasing radius factor */ |
298 |
> |
#define initrad (256>>3) /* for 256 cols, radius starts */ |
299 |
> |
#define radiusbiasshift 6 /* at 32.0 biased by 6 bits */ |
300 |
> |
#define radiusbias (((int) 1)<<radiusbiasshift) |
301 |
> |
#define initradius (initrad*radiusbias) /* and decreases by a */ |
302 |
> |
#define radiusdec 30 /* factor of 1/30 each cycle */ |
303 |
> |
|
304 |
> |
/* defs for decreasing alpha factor */ |
305 |
> |
#define alphabiasshift 10 /* alpha starts at 1.0 */ |
306 |
> |
#define initalpha (((int) 1)<<alphabiasshift) |
307 |
> |
int alphadec; /* biased by 10 bits */ |
308 |
> |
|
309 |
> |
/* radbias and alpharadbias used for radpower calculation */ |
310 |
|
#define radbiasshift 8 |
311 |
< |
#define radbias (1<<radbiasshift) |
311 |
> |
#define radbias (((int) 1)<<radbiasshift) |
312 |
|
#define alpharadbshift (alphabiasshift+radbiasshift) |
313 |
< |
#define alpharadbias (1<<alpharadbshift) |
313 |
> |
#define alpharadbias (((int) 1)<<alpharadbshift) |
314 |
|
|
315 |
< |
/* other defs */ |
316 |
< |
#define netbiasshift 4 |
317 |
< |
#define funnyshift (intbiasshift-netbiasshift) |
318 |
< |
#define maxnetval ((256<<netbiasshift)-1) |
319 |
< |
#define ncycles 100 |
320 |
< |
#define jump1 499 /* prime */ |
285 |
< |
#define jump2 491 /* prime */ |
286 |
< |
#define jump3 487 /* any pic whose size was divisible by all */ |
287 |
< |
#define jump4 503 /* four primes would be simply enormous */ |
315 |
> |
/* four primes near 500 - assume no image has a length so large */ |
316 |
> |
/* that it is divisible by all four primes */ |
317 |
> |
#define prime1 499 |
318 |
> |
#define prime2 491 |
319 |
> |
#define prime3 487 |
320 |
> |
#define prime4 503 |
321 |
|
|
289 |
– |
/* cheater definitions (GW) */ |
290 |
– |
#define thepicture thesamples |
291 |
– |
#define lengthcount (nsamples*3) |
292 |
– |
#define samplefac 1 |
293 |
– |
|
322 |
|
typedef int pixel[4]; /* BGRc */ |
323 |
+ |
pixel network[256]; |
324 |
|
|
325 |
< |
static pixel network[256]; |
325 |
> |
int netindex[256]; /* for network lookup - really 256 */ |
326 |
|
|
327 |
< |
static int netindex[256]; |
327 |
> |
int bias [256]; /* bias and freq arrays for learning */ |
328 |
> |
int freq [256]; |
329 |
> |
int radpower[initrad]; /* radpower for precomputation */ |
330 |
|
|
300 |
– |
static int bias [256]; |
301 |
– |
static int freq [256]; |
302 |
– |
static int radpower[256]; /* actually need only go up to maxrad */ |
331 |
|
|
332 |
< |
/* fixed space overhead 256*4+256+256+256+256 words = 256*8 = 8kB */ |
332 |
> |
/* initialise network in range (0,0,0) to (255,255,255) */ |
333 |
|
|
334 |
< |
|
335 |
< |
static |
308 |
< |
initnet() |
334 |
> |
static void |
335 |
> |
initnet(void) |
336 |
|
{ |
337 |
|
register int i; |
338 |
|
register int *p; |
339 |
|
|
340 |
< |
for (i=0; i<clrtabsiz; i++) { |
340 |
> |
for (i=0; i<netsize; i++) { |
341 |
|
p = network[i]; |
342 |
< |
p[0] = |
343 |
< |
p[1] = |
317 |
< |
p[2] = (i<<8) / clrtabsiz; |
318 |
< |
freq[i] = intbias/clrtabsiz; /* 1/256 */ |
342 |
> |
p[0] = p[1] = p[2] = (i << (netbiasshift+8))/netsize; |
343 |
> |
freq[i] = intbias/netsize; /* 1/netsize */ |
344 |
|
bias[i] = 0; |
345 |
|
} |
346 |
|
} |
347 |
|
|
348 |
|
|
349 |
< |
static |
350 |
< |
inxbuild() |
349 |
> |
/* do after unbias - insertion sort of network and build netindex[0..255] */ |
350 |
> |
|
351 |
> |
static void |
352 |
> |
inxbuild(void) |
353 |
|
{ |
354 |
|
register int i,j,smallpos,smallval; |
355 |
|
register int *p,*q; |
356 |
< |
int start,previous; |
356 |
> |
int previouscol,startpos; |
357 |
|
|
358 |
< |
previous = 0; |
359 |
< |
start = 0; |
360 |
< |
for (i=0; i<clrtabsiz; i++) { |
358 |
> |
previouscol = 0; |
359 |
> |
startpos = 0; |
360 |
> |
for (i=0; i<netsize; i++) { |
361 |
|
p = network[i]; |
362 |
|
smallpos = i; |
363 |
|
smallval = p[1]; /* index on g */ |
364 |
< |
/* find smallest in i+1..clrtabsiz-1 */ |
365 |
< |
for (j=i+1; j<clrtabsiz; j++) { |
364 |
> |
/* find smallest in i..netsize-1 */ |
365 |
> |
for (j=i+1; j<netsize; j++) { |
366 |
|
q = network[j]; |
367 |
|
if (q[1] < smallval) { /* index on g */ |
368 |
|
smallpos = j; |
370 |
|
} |
371 |
|
} |
372 |
|
q = network[smallpos]; |
373 |
+ |
/* swap p (i) and q (smallpos) entries */ |
374 |
|
if (i != smallpos) { |
375 |
|
j = q[0]; q[0] = p[0]; p[0] = j; |
376 |
|
j = q[1]; q[1] = p[1]; p[1] = j; |
378 |
|
j = q[3]; q[3] = p[3]; p[3] = j; |
379 |
|
} |
380 |
|
/* smallval entry is now in position i */ |
381 |
< |
if (smallval != previous) { |
382 |
< |
netindex[previous] = (start+i)>>1; |
383 |
< |
for (j=previous+1; j<smallval; j++) netindex[j] = i; |
384 |
< |
previous = smallval; |
385 |
< |
start = i; |
381 |
> |
if (smallval != previouscol) { |
382 |
> |
netindex[previouscol] = (startpos+i)>>1; |
383 |
> |
for (j=previouscol+1; j<smallval; j++) netindex[j] = i; |
384 |
> |
previouscol = smallval; |
385 |
> |
startpos = i; |
386 |
|
} |
387 |
|
} |
388 |
< |
netindex[previous] = (start+clrtabsiz-1)>>1; |
389 |
< |
for (j=previous+1; j<clrtabsiz; j++) netindex[j] = clrtabsiz-1; |
388 |
> |
netindex[previouscol] = (startpos+maxnetpos)>>1; |
389 |
> |
for (j=previouscol+1; j<256; j++) netindex[j] = maxnetpos; /* really 256 */ |
390 |
|
} |
391 |
|
|
392 |
|
|
393 |
|
static int |
394 |
< |
inxsearch(b,g,r) /* accepts real BGR values after net is unbiased */ |
395 |
< |
register int b,g,r; |
394 |
> |
inxsearch( /* accepts real BGR values after net is unbiased */ |
395 |
> |
register int b, |
396 |
> |
register int g, |
397 |
> |
register int r |
398 |
> |
) |
399 |
|
{ |
400 |
< |
register int i,j,best,x,y,bestd; |
400 |
> |
register int i,j,dist,a,bestd; |
401 |
|
register int *p; |
402 |
+ |
int best; |
403 |
|
|
404 |
|
bestd = 1000; /* biggest possible dist is 256*3 */ |
405 |
|
best = -1; |
406 |
|
i = netindex[g]; /* index on g */ |
407 |
< |
j = i-1; |
407 |
> |
j = i-1; /* start at netindex[g] and work outwards */ |
408 |
|
|
409 |
< |
while ((i<clrtabsiz) || (j>=0)) { |
410 |
< |
if (i<clrtabsiz) { |
409 |
> |
while ((i<netsize) || (j>=0)) { |
410 |
> |
if (i<netsize) { |
411 |
|
p = network[i]; |
412 |
< |
x = p[1] - g; /* inx key */ |
413 |
< |
if (x >= bestd) i = clrtabsiz; /* stop iter */ |
412 |
> |
dist = p[1] - g; /* inx key */ |
413 |
> |
if (dist >= bestd) i = netsize; /* stop iter */ |
414 |
|
else { |
415 |
|
i++; |
416 |
< |
if (x<0) x = -x; |
417 |
< |
y = p[0] - b; |
418 |
< |
if (y<0) y = -y; |
419 |
< |
x += y; |
420 |
< |
if (x<bestd) { |
421 |
< |
y = p[2] - r; |
422 |
< |
if (y<0) y = -y; |
391 |
< |
x += y; /* x holds distance */ |
392 |
< |
if (x<bestd) {bestd=x; best=p[3];} |
416 |
> |
if (dist<0) dist = -dist; |
417 |
> |
a = p[0] - b; if (a<0) a = -a; |
418 |
> |
dist += a; |
419 |
> |
if (dist<bestd) { |
420 |
> |
a = p[2] - r; if (a<0) a = -a; |
421 |
> |
dist += a; |
422 |
> |
if (dist<bestd) {bestd=dist; best=p[3];} |
423 |
|
} |
424 |
|
} |
425 |
|
} |
426 |
|
if (j>=0) { |
427 |
|
p = network[j]; |
428 |
< |
x = g - p[1]; /* inx key - reverse dif */ |
429 |
< |
if (x >= bestd) j = -1; /* stop iter */ |
428 |
> |
dist = g - p[1]; /* inx key - reverse dif */ |
429 |
> |
if (dist >= bestd) j = -1; /* stop iter */ |
430 |
|
else { |
431 |
|
j--; |
432 |
< |
if (x<0) x = -x; |
433 |
< |
y = p[0] - b; |
434 |
< |
if (y<0) y = -y; |
435 |
< |
x += y; |
436 |
< |
if (x<bestd) { |
437 |
< |
y = p[2] - r; |
438 |
< |
if (y<0) y = -y; |
409 |
< |
x += y; /* x holds distance */ |
410 |
< |
if (x<bestd) {bestd=x; best=p[3];} |
432 |
> |
if (dist<0) dist = -dist; |
433 |
> |
a = p[0] - b; if (a<0) a = -a; |
434 |
> |
dist += a; |
435 |
> |
if (dist<bestd) { |
436 |
> |
a = p[2] - r; if (a<0) a = -a; |
437 |
> |
dist += a; |
438 |
> |
if (dist<bestd) {bestd=dist; best=p[3];} |
439 |
|
} |
440 |
|
} |
441 |
|
} |
444 |
|
} |
445 |
|
|
446 |
|
|
447 |
+ |
/* finds closest neuron (min dist) and updates freq */ |
448 |
+ |
/* finds best neuron (min dist-bias) and returns position */ |
449 |
+ |
/* for frequently chosen neurons, freq[i] is high and bias[i] is negative */ |
450 |
+ |
/* bias[i] = gamma*((1/netsize)-freq[i]) */ |
451 |
+ |
|
452 |
|
static int |
453 |
< |
contest(b,g,r) /* accepts biased BGR values */ |
454 |
< |
register int b,g,r; |
453 |
> |
contest( /* accepts biased BGR values */ |
454 |
> |
register int b, |
455 |
> |
register int g, |
456 |
> |
register int r |
457 |
> |
) |
458 |
|
{ |
459 |
< |
register int i,best,bestbias,x,y,bestd,bestbiasd; |
460 |
< |
register int *p,*q, *pp; |
459 |
> |
register int i,dist,a,biasdist,betafreq; |
460 |
> |
int bestpos,bestbiaspos,bestd,bestbiasd; |
461 |
> |
register int *p,*f, *n; |
462 |
|
|
463 |
< |
bestd = ~(1<<31); |
463 |
> |
bestd = ~(((int) 1)<<31); |
464 |
|
bestbiasd = bestd; |
465 |
< |
best = -1; |
466 |
< |
bestbias = best; |
467 |
< |
q = bias; |
468 |
< |
p = freq; |
469 |
< |
for (i=0; i<clrtabsiz; i++) { |
470 |
< |
pp = network[i]; |
471 |
< |
x = pp[0] - b; |
472 |
< |
if (x<0) x = -x; |
473 |
< |
y = pp[1] - g; |
474 |
< |
if (y<0) y = -y; |
475 |
< |
x += y; |
476 |
< |
y = pp[2] - r; |
477 |
< |
if (y<0) y = -y; |
478 |
< |
x += y; /* x holds distance */ |
479 |
< |
/* >> netbiasshift not needed if funnyshift used */ |
480 |
< |
if (x<bestd) {bestd=x; best=i;} |
481 |
< |
y = x - ((*q)>>funnyshift); /* y holds biasd */ |
482 |
< |
if (y<bestbiasd) {bestbiasd=y; bestbias=i;} |
446 |
< |
y = (*p >> betashift); /* y holds beta*freq */ |
447 |
< |
*p -= y; |
448 |
< |
*q += (y<<gammashift); |
449 |
< |
p++; |
450 |
< |
q++; |
465 |
> |
bestpos = -1; |
466 |
> |
bestbiaspos = bestpos; |
467 |
> |
p = bias; |
468 |
> |
f = freq; |
469 |
> |
|
470 |
> |
for (i=0; i<netsize; i++) { |
471 |
> |
n = network[i]; |
472 |
> |
dist = n[0] - b; if (dist<0) dist = -dist; |
473 |
> |
a = n[1] - g; if (a<0) a = -a; |
474 |
> |
dist += a; |
475 |
> |
a = n[2] - r; if (a<0) a = -a; |
476 |
> |
dist += a; |
477 |
> |
if (dist<bestd) {bestd=dist; bestpos=i;} |
478 |
> |
biasdist = dist - ((*p)>>(intbiasshift-netbiasshift)); |
479 |
> |
if (biasdist<bestbiasd) {bestbiasd=biasdist; bestbiaspos=i;} |
480 |
> |
betafreq = (*f >> betashift); |
481 |
> |
*f++ -= betafreq; |
482 |
> |
*p++ += (betafreq<<gammashift); |
483 |
|
} |
484 |
< |
freq[best] += beta; |
485 |
< |
bias[best] -= betagamma; |
486 |
< |
return(bestbias); |
484 |
> |
freq[bestpos] += beta; |
485 |
> |
bias[bestpos] -= betagamma; |
486 |
> |
return(bestbiaspos); |
487 |
|
} |
488 |
|
|
489 |
|
|
490 |
< |
static |
491 |
< |
alterneigh(rad,i,b,g,r) /* accepts biased BGR values */ |
492 |
< |
int rad,i; |
493 |
< |
register int b,g,r; |
490 |
> |
/* move neuron i towards (b,g,r) by factor alpha */ |
491 |
> |
|
492 |
> |
static void |
493 |
> |
altersingle( /* accepts biased BGR values */ |
494 |
> |
register int alpha, |
495 |
> |
register int i, |
496 |
> |
register int b, |
497 |
> |
register int g, |
498 |
> |
register int r |
499 |
> |
) |
500 |
|
{ |
501 |
+ |
register int *n; |
502 |
+ |
|
503 |
+ |
n = network[i]; /* alter hit neuron */ |
504 |
+ |
*n -= (alpha*(*n - b)) / initalpha; |
505 |
+ |
n++; |
506 |
+ |
*n -= (alpha*(*n - g)) / initalpha; |
507 |
+ |
n++; |
508 |
+ |
*n -= (alpha*(*n - r)) / initalpha; |
509 |
+ |
} |
510 |
+ |
|
511 |
+ |
|
512 |
+ |
/* move neurons adjacent to i towards (b,g,r) by factor */ |
513 |
+ |
/* alpha*(1-((i-j)^2/[r]^2)) precomputed as radpower[|i-j|]*/ |
514 |
+ |
|
515 |
+ |
static void |
516 |
+ |
alterneigh( /* accents biased BGR values */ |
517 |
+ |
int rad, |
518 |
+ |
int i, |
519 |
+ |
register int b, |
520 |
+ |
register int g, |
521 |
+ |
register int r |
522 |
+ |
) |
523 |
+ |
{ |
524 |
|
register int j,k,lo,hi,a; |
525 |
|
register int *p, *q; |
526 |
|
|
527 |
< |
lo = i-rad; |
528 |
< |
if (lo<-1) lo= -1; |
468 |
< |
hi = i+rad; |
469 |
< |
if (hi>clrtabsiz) hi=clrtabsiz; |
527 |
> |
lo = i-rad; if (lo<-1) lo= -1; |
528 |
> |
hi = i+rad; if (hi>netsize) hi=netsize; |
529 |
|
|
530 |
|
j = i+1; |
531 |
|
k = i-1; |
554 |
|
} |
555 |
|
|
556 |
|
|
557 |
< |
static |
558 |
< |
altersingle(alpha,j,b,g,r) /* accepts biased BGR values */ |
500 |
< |
register int alpha,j,b,g,r; |
557 |
> |
static void |
558 |
> |
learn(void) |
559 |
|
{ |
502 |
– |
register int *q; |
503 |
– |
|
504 |
– |
q = network[j]; /* alter hit neuron */ |
505 |
– |
*q -= (alpha*(*q - b)) / initalpha; |
506 |
– |
q++; |
507 |
– |
*q -= (alpha*(*q - g)) / initalpha; |
508 |
– |
q++; |
509 |
– |
*q -= (alpha*(*q - r)) / initalpha; |
510 |
– |
} |
511 |
– |
|
512 |
– |
|
513 |
– |
static |
514 |
– |
learn() |
515 |
– |
{ |
560 |
|
register int i,j,b,g,r; |
561 |
< |
int radius,rad,alpha,step,delta,upto; |
561 |
> |
int radius,rad,alpha,step,delta,samplepixels; |
562 |
|
register unsigned char *p; |
563 |
|
unsigned char *lim; |
564 |
|
|
565 |
< |
upto = lengthcount/(3*samplefac); |
522 |
< |
delta = upto/ncycles; |
523 |
< |
lim = thepicture + lengthcount; |
565 |
> |
alphadec = 30 + ((samplefac-1)/3); |
566 |
|
p = thepicture; |
567 |
+ |
lim = thepicture + lengthcount; |
568 |
+ |
samplepixels = lengthcount/(3*samplefac); |
569 |
+ |
delta = samplepixels/ncycles; |
570 |
|
alpha = initalpha; |
571 |
|
radius = initradius; |
572 |
+ |
|
573 |
|
rad = radius >> radiusbiasshift; |
574 |
|
if (rad <= 1) rad = 0; |
575 |
|
for (i=0; i<rad; i++) |
576 |
|
radpower[i] = alpha*(((rad*rad - i*i)*radbias)/(rad*rad)); |
577 |
< |
|
578 |
< |
if ((lengthcount%jump1) != 0) step = 3*jump1; |
577 |
> |
|
578 |
> |
if ((lengthcount%prime1) != 0) step = 3*prime1; |
579 |
|
else { |
580 |
< |
if ((lengthcount%jump2) !=0) step = 3*jump2; |
580 |
> |
if ((lengthcount%prime2) !=0) step = 3*prime2; |
581 |
|
else { |
582 |
< |
if ((lengthcount%jump3) !=0) step = 3*jump3; |
583 |
< |
else step = 3*jump4; |
582 |
> |
if ((lengthcount%prime3) !=0) step = 3*prime3; |
583 |
> |
else step = 3*prime4; |
584 |
|
} |
585 |
|
} |
586 |
+ |
|
587 |
|
i = 0; |
588 |
< |
while (i < upto) { |
588 |
> |
while (i < samplepixels) { |
589 |
|
b = p[0] << netbiasshift; |
590 |
|
g = p[1] << netbiasshift; |
591 |
|
r = p[2] << netbiasshift; |
592 |
|
j = contest(b,g,r); |
593 |
|
|
594 |
|
altersingle(alpha,j,b,g,r); |
595 |
< |
if (rad) alterneigh(rad,j,b,g,r); |
549 |
< |
/* alter neighbours */ |
595 |
> |
if (rad) alterneigh(rad,j,b,g,r); /* alter neighbours */ |
596 |
|
|
597 |
|
p += step; |
598 |
|
if (p >= lim) p -= lengthcount; |
609 |
|
} |
610 |
|
} |
611 |
|
|
612 |
< |
static |
613 |
< |
unbiasnet() |
612 |
> |
/* unbias network to give 0..255 entries */ |
613 |
> |
/* which can then be used for colour map */ |
614 |
> |
/* and record position i to prepare for sort */ |
615 |
> |
|
616 |
> |
static void |
617 |
> |
unbiasnet(void) |
618 |
|
{ |
619 |
|
int i,j; |
620 |
|
|
621 |
< |
for (i=0; i<clrtabsiz; i++) { |
621 |
> |
for (i=0; i<netsize; i++) { |
622 |
|
for (j=0; j<3; j++) |
623 |
|
network[i][j] >>= netbiasshift; |
624 |
|
network[i][3] = i; /* record colour no */ |
625 |
|
} |
626 |
|
} |
627 |
|
|
628 |
< |
/* Don't do this until the network has been unbiased */ |
628 |
> |
|
629 |
> |
/* Don't do this until the network has been unbiased (GW) */ |
630 |
|
|
631 |
< |
static |
632 |
< |
cpyclrtab() |
631 |
> |
static void |
632 |
> |
cpyclrtab(void) |
633 |
|
{ |
634 |
|
register int i,j,k; |
635 |
|
|
636 |
< |
for (j=0; j<clrtabsiz; j++) { |
636 |
> |
for (j=0; j<netsize; j++) { |
637 |
|
k = network[j][3]; |
638 |
|
for (i = 0; i < 3; i++) |
639 |
|
clrtab[k][i] = network[j][2-i]; |