Mercurial > hg > wm

comparison Meerwald-dir/wm_bruyn_d.c @ 24:9f20bce6184e v0.7

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
move directories, support netpbm 11
author Peter Meerwald-Stadler <pmeerw@pmeerw.net>
date Fri, 20 Dec 2024 13:08:59 +0100
parents Meerwald/wm_bruyn_d.c@bd669312f068
children
comparison
equal deleted inserted replaced
23:71dd4b96221b 24:9f20bce6184e
1 #include "wm.h"
2 #include "signature.h"
3 #include "coord.h"
4 #include "gray.h"
5 #include "sort.h"
6 #include "bruyn_common.h"
7 #include "netpbm/pgm.h"
8
9 char *progname;
10
11 // prints out program's parameters
12 void usage(void) {
13 fprintf(stderr, "usage: %s [-b n] [-h] [-k] [-n n] [-o file] [-pP n] [-q n] [-tT n] [-v n] -s file file\n", progname);
14 fprintf(stderr, "\t-b n\t\tblock size\n");
15 fprintf(stderr, "\t-h\t\tprint usage\n");
16 fprintf(stderr, "\t-k\t\tdisable block skipping\n");
17 fprintf(stderr, "\t-n n\t\tnumber of signature bits to detect\n");
18 fprintf(stderr, "\t-o file\t\textracted signature file\n");
19 fprintf(stderr, "\t-p n\t\tpattern type for zone 1\n");
20 fprintf(stderr, "\t-P n\t\tpattern type for zone 2\n");
21 fprintf(stderr, "\t-q n\t\tsignature strength\n");
22 fprintf(stderr, "\t-s file\t\tembedded signature\n");
23 fprintf(stderr, "\t-t n\t\tthreshold for noise\n");
24 fprintf(stderr, "\t-T n\t\tthreshold for slope\n");
25 fprintf(stderr, "\t-v n\t\tverbosity level\n");
26 exit(0);
27 }
28
29 int main(int argc, char *argv[]) {
30 FILE *in = stdin;
31 FILE *out = stdout;
32 FILE *sig = NULL;
33
34 gray** image;
35 gray **block;
36 gray **zone;
37 gray **category1, **category2;
38 gray maxval;
39 double *slope;
40 int rows, cols, format;
41 int c;
42 int i, j;
43 int n;
44 int row;
45 int n_block;
46
47 char signature_name[MAXPATHLEN];
48 char input_name[MAXPATHLEN] = "(stdin)";
49 char output_name[MAXPATHLEN] = "(stdout)";
50
51 double quality = 0.0;
52 double threshold_noise = 0.0;
53 double threshold_slope = 0.0;
54 int pattern1 = 0;
55 int pattern2 = 0;
56 int blocksize = 0;
57 int seed;
58
59 int verbose = 0;
60 int skipping = 0;
61
62 struct coords *coords;
63
64 progname = argv[0];
65
66 pgm_init(&argc, argv); wm_init2();
67
68 // parse command line and set options
69 while ((c = getopt(argc, argv, "b:h?n:o:p:P:q:s:t:T:v:k")) != EOF) {
70 switch (c) {
71 case 'h':
72 case '?':
73 usage();
74 break;
75 case 'k':
76 skipping = 1;
77 break;
78 case 'n':
79 nbit_signature = atoi(optarg);
80 if (nbit_signature <= 0 || nbit_signature > NBITSIGNATURE) {
81 fprintf(stderr, "%s: invalid signature length %d\n", progname, nbit_signature);
82 exit(1);
83 }
84 break;
85 case 'o':
86 if ((out = fopen(optarg, "wb")) == NULL) {
87 fprintf(stderr, "%s: unable to open output file %s\n", progname, optarg);
88 exit(1);
89 }
90 strcpy(output_name, optarg);
91 break;
92 case 'p':
93 pattern1 = atoi(optarg);
94 if (pattern1 <= 0 || pattern1 > NPATTERN) {
95 fprintf(stderr, "%s: pattern type out of range\n", progname);
96 exit(1);
97 }
98 break;
99 case 'P':
100 pattern2 = atoi(optarg);
101 if (pattern2 <= 0 || pattern2 > 3) {
102 fprintf(stderr, "%s: pattern type out of range\n", progname);
103 exit(1);
104 }
105 break;
106 case 'q':
107 quality = atof(optarg);
108 if (quality <= 0) {
109 fprintf(stderr, "%s: quality factor %f out of range\n", progname, quality);
110 }
111 break;
112 case 's':
113 if ((sig = fopen(optarg, "r")) == NULL) {
114 fprintf(stderr, "%s: unable to open signature file %s\n", progname, optarg);
115 exit(1);
116 }
117 strcpy(signature_name, optarg);
118 break;
119 case 't':
120 threshold_noise = atof(optarg);
121 if (threshold_noise <= 0) {
122 fprintf(stderr, "%s: noise threshold %f out of range\n", progname, threshold_noise);
123 }
124 break;
125 case 'T':
126 threshold_slope = atof(optarg);
127 if (threshold_slope <= 0) {
128 fprintf(stderr, "%s: slope threshold %f out of range\n", progname, threshold_slope);
129 }
130 break;
131 case 'v':
132 verbose = atoi(optarg);
133 if (verbose < 0) {
134 fprintf(stderr, "%s: verbosity level %d out of range\n",progname, verbose);
135 exit(1);
136 }
137 break;
138 }
139 }
140
141 argc -= optind;
142 argv += optind;
143
144 if (argc > 1) {
145 usage();
146 exit(1);
147 }
148
149 // open input image file or read from stdin
150 if (argc == 1 && *argv[0] != '-') {
151 if ((in = fopen(argv[0], "rb")) == NULL) {
152 fprintf(stderr, "%s: unable to open input file %s\n", progname, argv[0]);
153 exit(1);
154 }
155 else
156 strcpy(input_name, argv[0]);
157 }
158
159 // read signature file and set options
160 // command line options override signature file options
161 if (sig) {
162 char line[1024];
163 fgets(line, sizeof(line), sig);
164 if (strspn(line, "BRSG") >= 4) {
165 if (nbit_signature == 0)
166 fscanf(sig, "%d\n", &nbit_signature);
167 else
168 fscanf(sig, "%*d\n");
169 if (skipping == 0)
170 fscanf(sig, "%d\n", &skipping);
171 else
172 fscanf(sig, "%*d\n");
173 if (pattern1 == 0)
174 fscanf(sig, "%d\n", &pattern1);
175 else
176 fscanf(sig, "%*d\n");
177 if (pattern2 == 0)
178 fscanf(sig, "%d\n", &pattern2);
179 else
180 fscanf(sig, "%*d\n");
181 if (quality == 0.0)
182 fscanf(sig, "%lf\n", &quality);
183 else
184 fscanf(sig, "%*f\n");
185 if (threshold_noise == 0.0)
186 fscanf(sig, "%lf\n", &threshold_noise);
187 else
188 fscanf(sig, "%*f\n");
189 if (threshold_slope == 0.0)
190 fscanf(sig, "%lf\n", &threshold_slope);
191 else
192 fscanf(sig, "%*f\n");
193 if (blocksize == 0)
194 fscanf(sig, "%d\n", &blocksize);
195 else
196 fscanf(sig, "%*d\n");
197 fscanf(sig, "%d\n", &seed);
198 srandom(seed);
199 n_signature = NBITSTOBYTES(nbit_signature);
200 fread(signature, sizeof(char), n_signature, sig);
201 init_signature_bits();
202 fscanf(sig, "\n");
203 }
204 else {
205 fprintf(stderr, "%s: invalid signature file %s\n", progname, signature_name);
206 exit(1);
207 }
208 fclose(sig);
209 }
210 else {
211 fprintf(stderr, "%s: signature file not specified, use -s file option\n", progname);
212 exit(1);
213 }
214
215 if (pattern1 <= 0 || pattern2 <= 0 || pattern1 > NPATTERN || pattern2 > NPATTERN) {
216 fprintf(stderr, "%s: invalid pattern type specified\n", progname);
217 exit(1);
218 }
219
220 // read dimensions of input image file
221 pgm_readpgminit(in, &cols, &rows, &maxval, &format);
222
223 // see if we can extract all signature bits
224 // we want at least half of the blocks untouched
225 if (((rows / blocksize) * (cols / blocksize)) < nbit_signature / 2) {
226 fprintf(stderr, "%s: image not large enough to contain %d bits of signature\n", progname, nbit_signature);
227 exit(1);
228 }
229 n_block = blocksize * blocksize;
230
231 // allocate structure to remember which blocks we already touched,
232 // allow plenty of room to skip over blocks
233 if ((coords = alloc_coords(nbit_signature * 16)) == NULL) {
234 fprintf(stderr, "%s: unable to allocate memory\n", progname);
235 exit(1);
236 }
237
238 // read in input image file
239 image = pgm_allocarray(cols, rows);
240 for (row = 0; row < rows; row++)
241 pgm_readpgmrow(in, image[row], cols, maxval, format);
242
243 fclose(in);
244
245 row = 0;
246
247 // allocate memory for one block
248 block = alloc_grays(blocksize, blocksize);
249
250 // allocate memory for zone classification
251 zone = alloc_grays(blocksize, blocksize);
252
253 // allocate memory for category classification
254 category1 = alloc_grays(blocksize, blocksize);
255 category2 = alloc_grays(blocksize, blocksize);
256
257 // set up category classification array according to
258 // pattern type parameter
259 for (i = 0; i < blocksize; i++)
260 for (j = 0; j < blocksize; j++) {
261 category1[j][i] = lookup_pattern(pattern1, i, j);
262 category2[j][i] = lookup_pattern(pattern2, i, j);
263 }
264
265 // allocate memory for slope calculation
266 slope = malloc(sizeof(double) * n_block);
267
268 fprintf(out, "BRWM\n");
269 fprintf(out, "%d\n", nbit_signature);
270
271 // extract all the signature bits, one by one
272 n = 0;
273 while (n < nbit_signature) {
274 int xb;
275 int yb;
276 int blocktype;
277 double smax;
278 int alpha, beta_minus, beta_plus;
279 double mean_1A, mean_1B, mean_2A, mean_2B, mean_1, mean_2;
280 int n_1A, n_1B, n_2A, n_2B, n_1, n_2;
281 double sigma, sigma_1, sigma_2;
282 int zone1_ok, zone2_ok;
283
284 // find an unused block randomly, depending on seed
285 do {
286 xb = random() % (cols / blocksize);
287 yb = random() % (rows / blocksize);
288 } while (add_coord(coords, xb, yb) < 0);
289
290 // copy image block
291 fprintf(stderr, "XXX1 %d %d %d\n", xb*blocksize, yb*blocksize, blocksize);
292 copy_grays_to_block(block, image, xb * blocksize, yb * blocksize, blocksize, blocksize);
293 fprintf(stderr, "XXX2\n");
294
295 if (verbose > 0)
296 fprintf(stderr, "detecting bit #%d in block at (%d/%d)\n", n, xb * blocksize, yb * blocksize);
297
298 // sort luminance values in block to represent increasing function F
299 sort_grays(block[0], n_block);
300
301 // calculate slopes of F and determine smax, the max. slope of F
302 // the index where smax occures is called alpha
303 alpha = 0;
304 smax = 0.0;
305 for (i = 0; i < n_block - 1; i++) {
306 slope[i] = block[0][i + 1] - block[0][i];
307 if (slope[i] > smax) {
308 smax = slope[i];
309 alpha = i;
310 }
311 }
312 slope[n_block - 1] = 0;
313
314 // block type classification
315 blocktype = BLOCKTYPE_UNKNOWN;
316
317 if (smax < threshold_noise) {
318 // block has noise contrast
319 beta_minus = beta_plus = alpha;
320 blocktype = BLOCKTYPE_NOISE;
321 }
322 else {
323 // block has progressive or hard contrast, let's find out...
324
325 beta_minus = alpha - 1;
326 while (beta_minus >= 0 && smax - slope[beta_minus] <= threshold_slope)
327 beta_minus--;
328
329 beta_plus = alpha + 1;
330 while (beta_plus < n_block && smax - slope[beta_plus] <= threshold_slope)
331 beta_plus++;
332
333 if (beta_minus + 1 == alpha && beta_plus - 1 == alpha)
334 blocktype = BLOCKTYPE_HARD;
335 else
336 blocktype = BLOCKTYPE_PROGRESSIVE;
337 }
338
339 if (verbose > 1) {
340 fprintf(stderr, "blocktype: %d\n", blocktype);
341 fprintf(stderr, "Smax = %lf, alpha = %d, beta- = %d, beta+ = %d\n", smax, alpha, beta_minus, beta_plus);
342 }
343
344 // block pixel classification
345 for (i = 0; i < blocksize; i++)
346 for (j = 0; j < blocksize; j++) {
347 gray pixel = image[yb * blocksize + j][xb * blocksize + i];
348 zone[j][i] = ZONE_VOID;
349 switch (blocktype) {
350 case BLOCKTYPE_PROGRESSIVE:
351 case BLOCKTYPE_HARD:
352 if (pixel < block[0][beta_minus])
353 zone[j][i] = ZONE_1;
354 else if (pixel > block[0][beta_plus])
355 zone[j][i] = ZONE_2;
356 break;
357 case BLOCKTYPE_NOISE:
358 if (pixel < block[0][n_block / 2])
359 zone[j][i] = ZONE_1;
360 else if (pixel > block[0][n_block / 2])
361 zone[j][i] = ZONE_2;
362 break;
363 default:
364 fprintf(stderr, "%s: invalid block type\n", progname);
365 break;
366 }
367 }
368
369 // calculate mean values for zone/categories
370 mean_1A = mean_1B = mean_2A = mean_2B = mean_1 = mean_2 = 0.0;
371 n_1A = n_1B = n_2A = n_2B = n_1 = n_2 = 0;
372 for (i = 0; i < blocksize; i++)
373 for (j = 0; j < blocksize; j++) {
374 gray pixel = image[yb * blocksize + j][xb * blocksize + i];
375 int pixel_zone = zone[j][i];
376 int pixel_category = CATEGORY_VOID;
377 if (pixel_zone == ZONE_1)
378 pixel_category = category1[j][i];
379 else if (pixel_zone == ZONE_2)
380 pixel_category = category2[j][i];
381
382 switch (pixel_zone | pixel_category) {
383 case CLASSIFICATION_1A:
384 n_1++;
385 n_1A++;
386 mean_1A += pixel;
387 mean_1 += pixel;
388 break;
389 case CLASSIFICATION_1B:
390 n_1++;
391 n_1B++;
392 mean_1B += pixel;
393 mean_1 += pixel;
394 break;
395 case CLASSIFICATION_2A:
396 n_2++;
397 n_2A++;
398 mean_2A += pixel;
399 mean_2 += pixel;
400 break;
401 case CLASSIFICATION_2B:
402 n_2++;
403 n_2B++;
404 mean_2B += pixel;
405 mean_2 += pixel;
406 break;
407 }
408 }
409
410 if (n_1 && n_1A && n_1B) {
411 mean_1 /= (double) n_1;
412 mean_1A /= (double) n_1A;
413 mean_1B /= (double) n_1B;
414 zone1_ok = 1;
415 }
416 else {
417 mean_1 = mean_1A = mean_1B = 0.0;
418 zone1_ok = 0;
419 if (verbose > 0)
420 fprintf(stderr, "zone 1 unusable\n");
421 }
422
423 if (n_2 && n_2A && n_2B) {
424 mean_2 /= (double) n_2;
425 mean_2A /= (double) n_2A;
426 mean_2B /= (double) n_2B;
427 zone2_ok = 1;
428 }
429 else {
430 mean_2 = mean_2A = mean_2B = 0.0;
431 zone2_ok = 0;
432 if (verbose > 0)
433 fprintf(stderr, "zone 2 unusable\n");
434 }
435
436 // bit extraction
437 if (zone1_ok && zone2_ok) {
438 sigma_1 = mean_1A - mean_1B;
439 sigma_2 = mean_2A - mean_2B;
440
441 if (verbose > 2) {
442 fprintf(stderr, "m_1A = %lf, m_1B = %lf\n", mean_1A, mean_1B);
443 fprintf(stderr, "m_2A = %lf, m_2B = %lf\n", mean_2A, mean_2B);
444 fprintf(stderr, "sigma1 = %lf, sigma2 = %lf\n", sigma_1, sigma_2);
445 }
446
447 #define EPSILON 0.001
448 if (fabs(sigma_1 * sigma_2) < EPSILON) {
449 // case 3
450 sigma = MAX(fabs(sigma_1), fabs(sigma_2));
451 set_signature_bit(n, sigma > 0.0);
452 if (verbose > 0)
453 fprintf(stderr, "case 3, bit #%d = %d\n", n, sigma > 0.0);
454 }
455 else if (sigma_1 * sigma_2 > 0.0) {
456 // case 1
457 set_signature_bit(n, sigma_1 > 0.0);
458 if (verbose > 0)
459 fprintf(stderr, "case 1, bit #%d = %d\n", n, sigma_1 > 0.0);
460 }
461 else if (sigma_1 * sigma_2 < 0.0) {
462 // case 2
463 sigma = (double) (n_1A + n_1B) * sigma_1 + (double) (n_2A + n_2B) * sigma_2;
464 set_signature_bit(n, sigma > 0.0);
465 if (verbose > 0)
466 fprintf(stderr, "case 2, bit #%d = %d\n", n, sigma > 0.0);
467 }
468 }
469 else if (zone1_ok) {
470 set_signature_bit(n, mean_1A > mean_1B);
471 if (verbose > 0)
472 fprintf(stderr, "case 4, bit #%d = %d\n", n, mean_1A > mean_1B);
473 }
474 else if (zone2_ok) {
475 set_signature_bit(n, mean_2A > mean_2B);
476 if (verbose > 0)
477 fprintf(stderr, "case 5, bit #%d = %d\n", n, mean_2A > mean_2B);
478 }
479 else {
480 // pathological case - can it ever happen?
481 if (verbose > 0)
482 fprintf(stderr, "block skipped\n");
483 if (!skipping) continue;
484 }
485
486 n++;
487 }
488
489 free_grays(category2);
490 free_grays(category1);
491 free_grays(zone);
492 free_grays(block);
493
494 // write extracted signature
495
496 fwrite(signature, sizeof(char), n_signature, out);
497 fclose(out);
498
499 pgm_freearray(image, rows);
500
501 exit(0);
502 }

Repositories maintained by Peter Meerwald, pmeerw@pmeerw.net.