Mercurial > hg > wm

comparison Meerwald-dir/dct.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/dct.c@bd669312f068
children
comparison
equal deleted inserted replaced
23:71dd4b96221b 24:9f20bce6184e
1 #include "wm.h"
2 #include "dct.h"
3
4 #define INVROOT2 0.7071067814
5 #define SWAP(A, B) {double t = A; A = B; B = t;}
6
7 int N;
8 int M;
9
10 double *dct_NxN_tmp = NULL;
11 double *dct_NxN_costable = NULL;
12 int dct_NxN_log2N = 0;
13
14 static const unsigned int JPEG_lumin_quant_table[NJPEG][NJPEG] = {
15 {16, 11, 10, 16, 24, 40, 51, 61},
16 {12, 12, 14, 19, 26, 58, 60, 55},
17 {14, 13, 16, 24, 40, 57, 69, 56},
18 {14, 17, 22, 29, 51, 87, 80, 62},
19 {18, 22, 37, 56, 68, 109, 103, 77},
20 {24, 35, 55, 64, 81, 104, 113, 92},
21 {49, 64, 78, 87, 103, 121, 120, 101},
22 {72, 92, 95, 98, 112, 100, 103, 99}};
23
24 static void initcosarray()
25 {
26 int i,group,base,item,nitems,halfN;
27 double factor;
28
29 dct_NxN_log2N = -1;
30 do{
31 dct_NxN_log2N++;
32 if ((1<<dct_NxN_log2N)>N){
33 fprintf(stderr, "dct_NxN: %d not a power of 2\n", N);
34 exit(1);
35 }
36 }while((1<<dct_NxN_log2N)<N);
37 if (dct_NxN_costable) free(dct_NxN_costable);
38 dct_NxN_costable = malloc(N * sizeof(double));
39 #ifdef DEBUG
40 if(!dct_NxN_costable){
41 fprintf(stderr, "Unable to allocate C array\n");
42 exit(1);
43 }
44 #endif
45 halfN=N/2;
46 for(i=0;i<=halfN-1;i++) dct_NxN_costable[halfN+i]=4*i+1;
47 for(group=1;group<=dct_NxN_log2N-1;group++){
48 base= 1<<(group-1);
49 nitems=base;
50 factor = 1.0*(1<<(dct_NxN_log2N-group));
51 for(item=1; item<=nitems;item++) dct_NxN_costable[base+item-1]=factor*dct_NxN_costable[halfN+item-1];
52 }
53
54 for(i=1;i<=N-1;i++) dct_NxN_costable[i] = 1.0/(2.0*cos(dct_NxN_costable[i]*M_PI/(2.0*N)));
55 }
56
57 void init_dct_NxN(int width, int height) {
58 #ifdef DEBUG
59 if (width != height || width <= 0) {
60 fprintf(stderr, "init_dct_NxN(): dimensions out of range\n");
61 exit(1);
62 }
63 #endif
64
65 if (dct_NxN_tmp && M != height)
66 free(dct_NxN_tmp);
67
68 N = width;
69 M = height;
70
71 dct_NxN_tmp = malloc(height * sizeof(double));
72 #ifdef DEBUG
73 if (!dct_NxN_tmp) {
74 fprintf(stderr, "init_dct_NxN(): failed to allocate memory\n");
75 exit(1);
76 }
77 #endif
78
79 initcosarray();
80 }
81
82 static void bitrev(double *f, int len)
83 {
84 int i,j,m;
85
86 if (len<=2) return; /* No action necessary if n=1 or n=2 */
87 j=1;
88 for(i=1; i<=len; i++){
89 if(i<j)
90 SWAP(f[j-1], f[i-1]);
91 m = len>>1;
92 while(j>m){
93 j=j-m;
94 m=(m+1)>>1;
95 }
96 j=j+m;
97 }
98 }
99
100 static void inv_sums(double *f)
101 {
102 int stepsize,stage,curptr,nthreads,thread,step,nsteps;
103
104 for(stage=1; stage <=dct_NxN_log2N-1; stage++){
105 nthreads = 1<<(stage-1);
106 stepsize = nthreads<<1;
107 nsteps = (1<<(dct_NxN_log2N-stage)) - 1;
108 for(thread=1; thread<=nthreads; thread++){
109 curptr=N-thread;
110 for(step=1; step<=nsteps; step++){
111 f[curptr] += f[curptr-stepsize];
112 curptr -= stepsize;
113 }
114 }
115 }
116 }
117
118 static void fwd_sums(double *f)
119 {
120 int stepsize,stage,curptr,nthreads,thread,step,nsteps;
121
122 for(stage=dct_NxN_log2N-1; stage >=1; stage--){
123 nthreads = 1<<(stage-1);
124 stepsize = nthreads<<1;
125 nsteps = (1<<(dct_NxN_log2N-stage)) - 1;
126 for(thread=1; thread<=nthreads; thread++){
127 curptr=nthreads +thread-1;
128 for(step=1; step<=nsteps; step++){
129 f[curptr] += f[curptr+stepsize];
130 curptr += stepsize;
131 }
132 }
133 }
134 }
135
136 static void scramble(double *f,int len){
137 int i,ii1,ii2;
138
139 bitrev(f,len);
140 bitrev(&f[0], len>>1);
141 bitrev(&f[len>>1], len>>1);
142 ii1=len-1;
143 ii2=len>>1;
144 for(i=0; i<(len>>2); i++){
145 SWAP(f[ii1], f[ii2]);
146 ii1--;
147 ii2++;
148 }
149 }
150
151 static void unscramble(double *f,int len)
152 {
153 int i,ii1,ii2;
154
155 ii1 = len-1;
156 ii2 = len>>1;
157 for(i=0; i<(len>>2); i++){
158 SWAP(f[ii1], f[ii2]);
159 ii1--;
160 ii2++;
161 }
162 bitrev(&f[0], len>>1);
163 bitrev(&f[len>>1], len>>1);
164 bitrev(f,len);
165 }
166
167 static void inv_butterflies(double *f)
168 {
169 int stage,ii1,ii2,butterfly,ngroups,group,wingspan,increment,baseptr;
170 double Cfac,T;
171
172 for(stage=1; stage<=dct_NxN_log2N;stage++){
173 ngroups=1<<(dct_NxN_log2N-stage);
174 wingspan=1<<(stage-1);
175 increment=wingspan<<1;
176 for(butterfly=1; butterfly<=wingspan; butterfly++){
177 Cfac = dct_NxN_costable[wingspan+butterfly-1];
178 baseptr=0;
179 for(group=1; group<=ngroups; group++){
180 ii1=baseptr+butterfly-1;
181 ii2=ii1+wingspan;
182 T=Cfac * f[ii2];
183 f[ii2]=f[ii1]-T;
184 f[ii1]=f[ii1]+T;
185 baseptr += increment;
186 }
187 }
188 }
189 }
190
191 static void fwd_butterflies(double *f)
192 {
193 int stage,ii1,ii2,butterfly,ngroups,group,wingspan,increment,baseptr;
194 double Cfac,T;
195
196 for(stage=dct_NxN_log2N; stage>=1;stage--){
197 ngroups=1<<(dct_NxN_log2N-stage);
198 wingspan=1<<(stage-1);
199 increment=wingspan<<1;
200 for(butterfly=1; butterfly<=wingspan; butterfly++){
201 Cfac = dct_NxN_costable[wingspan+butterfly-1];
202 baseptr=0;
203 for(group=1; group<=ngroups; group++){
204 ii1=baseptr+butterfly-1;
205 ii2=ii1+wingspan;
206 T= f[ii2];
207 f[ii2]=Cfac *(f[ii1]-T);
208 f[ii1]=f[ii1]+T;
209 baseptr += increment;
210 }
211 }
212 }
213 }
214
215 static void ifct_noscale(double *f)
216 {
217 f[0] *= INVROOT2;
218 inv_sums(f);
219 bitrev(f,N);
220 inv_butterflies(f);
221 unscramble(f,N);
222 }
223
224 static void fct_noscale(double *f)
225 {
226 scramble(f,N);
227 fwd_butterflies(f);
228 bitrev(f,N);
229 fwd_sums(f);
230 f[0] *= INVROOT2;
231 }
232
233 void fdct_NxN(gray **pixels, double **dcts) {
234 int u,v;
235 double two_over_sqrtncolsnrows = 2.0/sqrt((double) N*M);
236
237 for (u=0; u < N; u++)
238 for (v=0; v < M; v++)
239 dcts[u][v] = ((int) pixels[u][v] - 128);
240
241 for (u=0; u<=M-1; u++){
242 fct_noscale(dcts[u]);
243 }
244 for (v=0; v<=N-1; v++){
245 for (u=0; u<=M-1; u++){
246 dct_NxN_tmp[u] = dcts[u][v];
247 }
248 fct_noscale(dct_NxN_tmp);
249 for (u=0; u<=M-1; u++){
250 dcts[u][v] = dct_NxN_tmp[u]*two_over_sqrtncolsnrows;
251 }
252 }
253 }
254
255 void idct_NxN(double **dcts, gray **pixels) {
256 int u,v;
257 double two_over_sqrtncolsnrows = 2.0/sqrt((double) N*M);
258
259 double **tmp;
260
261 tmp = alloc_coeffs(N, N);
262 for (u=0;u<N;u++)
263 for (v=0;v<M;v++)
264 tmp[u][v] = dcts[u][v];
265
266 for (u=0; u<=M-1; u++){
267 ifct_noscale(tmp[u]);
268 }
269 for (v=0; v<=N-1; v++){
270 for (u=0; u<=M-1; u++){
271 dct_NxN_tmp[u] = tmp[u][v];
272 }
273 ifct_noscale(dct_NxN_tmp);
274 for (u=0; u<=M-1; u++){
275 tmp[u][v] = dct_NxN_tmp[u]*two_over_sqrtncolsnrows;
276 }
277 }
278
279 for (u=0;u<N;u++)
280 for (v=0;v<M;v++)
281 pixels[u][v] = PIXELRANGE(tmp[u][v] + 128.5);
282 free(tmp);
283 }
284
285 void fdct_inplace_NxN(double **coeffs) {
286 int u,v;
287 double two_over_sqrtncolsnrows = 2.0/sqrt((double) N*M);
288
289 for (u=0; u<=M-1; u++)
290 fct_noscale(coeffs[u]);
291
292 for (v=0; v<=N-1; v++){
293 for (u=0; u<=M-1; u++)
294 dct_NxN_tmp[u] = coeffs[u][v];
295
296 fct_noscale(dct_NxN_tmp);
297 for (u=0; u<=M-1; u++)
298 coeffs[u][v] = dct_NxN_tmp[u]*two_over_sqrtncolsnrows;
299 }
300 }
301
302 void idct_inplace_NxN(double **coeffs) {
303 int u,v;
304 double two_over_sqrtncolsnrows = 2.0/sqrt((double) N*M);
305
306 for (u=0; u<=M-1; u++)
307 ifct_noscale(coeffs[u]);
308
309 for (v=0; v<=N-1; v++) {
310 for (u=0; u<=M-1; u++)
311 dct_NxN_tmp[u] = coeffs[u][v];
312
313 ifct_noscale(dct_NxN_tmp);
314 for (u=0; u<=M-1; u++)
315 coeffs[u][v] = dct_NxN_tmp[u]*two_over_sqrtncolsnrows;
316 }
317
318 }
319
320 double **dct_NxM_costable_x = NULL;
321 double **dct_NxM_costable_y = NULL;
322
323 void init_dct_NxM(int cols, int rows) {
324 int i, j;
325 double cx = sqrt(2.0 / cols);
326 double cy = sqrt(2.0 / rows);
327
328 #ifdef DEBUG
329 if (cols <= 0 || rows <= 0) {
330 fprintf(stderr, "init_dct_NxM(): dimensions out of range\n");
331 exit(1);
332 }
333 #endif
334
335 if (dct_NxM_costable_x && N != cols) {
336 free_coeffs(dct_NxM_costable_x);
337 dct_NxM_costable_x = NULL;
338 }
339
340 if (dct_NxM_costable_y && M != rows) {
341 free_coeffs(dct_NxM_costable_y);
342 dct_NxM_costable_y = NULL;
343 }
344
345 if (!dct_NxM_costable_x)
346 dct_NxM_costable_x = alloc_coeffs(cols, cols);
347 if (!dct_NxM_costable_y)
348 dct_NxM_costable_y = alloc_coeffs(rows, rows);
349
350 N = cols;
351 M = rows;
352
353 for (i = 0; i < cols; i++) {
354 for (j = 0; j < cols; j++) {
355 dct_NxM_costable_x[i][j] = cx * cos((M_PI * ((2*i + 1) * j)) / (double) (2 * N));
356 }
357 }
358
359 for (i = 0; i < rows; i++) {
360 for (j = 0; j < rows; j++) {
361 dct_NxM_costable_y[i][j] = cy * cos((M_PI * ((2*i + 1) * j)) / (double) (2 * M));
362 }
363 }
364 }
365
366 void fdct_NxM(gray **pixels, double **dcts) {
367 int x, y;
368 int i, j;
369 double t;
370 double cx0 = sqrt(1.0 / N);
371 double cy0 = sqrt(1.0 / M);
372
373 t = 0.0;
374 for (x = 0; x < N; x++)
375 for (y = 0; y < M; y++)
376 t += ((int) pixels[y][x] - 128);
377 dcts[0][0] = cx0 * cy0 * t;
378
379 for (i = 1; i < N; i++) {
380 t = 0.0;
381 for (x = 0; x < N; x++) {
382 double s = 0.0;
383 for (y = 0; y < M; y++) {
384 s += ((int) pixels[y][x] - 128);
385 }
386 t += s * dct_NxM_costable_x[x][i];
387 }
388 dcts[0][i] = cy0 * t;
389 }
390
391 for (j = 1; j < M; j++) {
392 t = 0.0;
393 for (y = 0; y < M; y++) {
394 double s = 0.0;
395 for (x = 0; x < N; x++) {
396 s += ((int) pixels[y][x] - 128);
397 }
398 t += s * dct_NxM_costable_y[y][j];
399 }
400 dcts[j][0] = cx0 * t;
401 }
402
403 for (i = 1; i < N; i++) {
404 for (j = 1; j < M; j++) {
405 t = 0.0;
406 for (x = 0; x < N; x++) {
407 double s = 0;
408 for (y = 0; y < M; y++) {
409 s += ((int) pixels[y][x] - 128) * dct_NxM_costable_y[y][j];
410 }
411 t += s * dct_NxM_costable_x[x][i];
412 }
413 dcts[j][i] = t;
414 }
415 }
416 }
417
418 void idct_NxM(double **dcts, gray **pixels) {
419 int x, y;
420 int i, j;
421 double cx0 = sqrt(1.0 / N);
422 double cy0 = sqrt(1.0 / M);
423 double t;
424
425 for (x = 0; x < N; x++) {
426 for (y = 0; y < M; y++) {
427
428 t = cx0 * cy0 * dcts[0][0];
429
430 for (i = 1; i < N; i++)
431 t += cy0 * dcts[0][i] * dct_NxM_costable_x[x][i];
432
433 for (j = 1; j < M; j++)
434 t += cx0 * dcts[j][0] * dct_NxM_costable_y[y][j];
435
436 for (i = 1; i < N; i++) {
437 double s = 0.0;
438 for (j = 1; j < M; j++) {
439 s += dcts[j][i] * dct_NxM_costable_y[y][j];
440 }
441 t += s * dct_NxM_costable_x[x][i];
442 }
443
444 pixels[y][x] = PIXELRANGE((int) (t + 128.5));
445 }
446 }
447 }
448
449 double C[NJPEG][NJPEG];
450 double Ct[NJPEG][NJPEG];
451 int Quantum[NJPEG][NJPEG];
452
453 void init_quantum_8x8(int quality) {
454 int i;
455 int j;
456
457 for (i = 0; i < NJPEG; i++)
458 for ( j = 0 ; j < NJPEG ; j++ )
459 Quantum[ i ][ j ] = 1 + ( ( 1 + i + j ) * quality );
460 }
461
462 void init_quantum_JPEG_lumin(int quality) {
463 int i;
464 int j;
465
466 if (quality < 50)
467 quality = 5000 / quality;
468 else
469 quality = 200 - quality * 2;
470
471 for (i = 0; i < NJPEG; i++)
472 for (j = 0 ; j < NJPEG ; j++)
473 if (quality)
474 Quantum[i][j] = (JPEG_lumin_quant_table[i][j] * quality + 50) / 100;
475 else
476 Quantum[i][j] = JPEG_lumin_quant_table[i][j];
477 }
478
479 void init_quantum_JPEG_chromin(int quality) {
480 int i;
481 int j;
482
483 if (quality < 50)
484 quality = 5000 / quality;
485 else
486 quality = 200 - quality * 2;
487
488 for (i = 0; i < NJPEG; i++)
489 for (j = 0 ; j < NJPEG ; j++)
490 if (quality)
491 Quantum[i][j] = (JPEG_lumin_quant_table[i][j] * quality + 50) / 100;
492 else
493 Quantum[i][j] = JPEG_lumin_quant_table[i][j];
494 }
495
496 void quantize_8x8(double **transform) {
497 int i;
498 int j;
499
500 for (i = 0; i < NJPEG; i++)
501 for (j = 0; j < NJPEG; j++)
502 transform[i][j] = ROUND(transform[i][j] / Quantum[i][j]);
503 }
504
505 void dequantize_8x8(double **transform) {
506 int i;
507 int j;
508
509 for (i = 0; i < NJPEG; i++)
510 for (j = 0; j < NJPEG; j++)
511 transform[i][j] = ROUND(transform[i][j] * Quantum[i][j]);
512 }
513
514 void init_dct_8x8() {
515 int i;
516 int j;
517 double pi = atan( 1.0 ) * 4.0;
518
519 for ( j = 0 ; j < NJPEG ; j++ ) {
520 C[ 0 ][ j ] = 1.0 / sqrt( (double) NJPEG );
521 Ct[ j ][ 0 ] = C[ 0 ][ j ];
522 }
523
524 for ( i = 1 ; i < NJPEG ; i++ )
525 for ( j = 0 ; j < NJPEG ; j++ ) {
526 C[ i ][ j ] = sqrt( 2.0 / NJPEG ) * cos( pi * ( 2 * j + 1 ) * i / ( 2.0 * NJPEG ) );
527 Ct[ j ][ i ] = C[ i ][ j ];
528 }
529 }
530
531 /*
532 * The Forward DCT routine implements the matrix function:
533 *
534 * DCT = C * pixels * Ct
535 */
536
537 void fdct_8x8(gray **input, double **output) {
538 double temp[NJPEG][NJPEG];
539 double temp1;
540 int i;
541 int j;
542 int k;
543
544 /* MatrixMultiply( temp, input, Ct ); */
545 for ( i = 0 ; i < NJPEG ; i++ ) {
546 for ( j = 0 ; j < NJPEG ; j++ ) {
547 temp[ i ][ j ] = 0.0;
548 for ( k = 0 ; k < NJPEG ; k++ )
549 temp[ i ][ j ] += ( (int) input[ i ][ k ] - 128 ) *
550 Ct[ k ][ j ];
551 }
552 }
553
554 /* MatrixMultiply( output, C, temp ); */
555 for ( i = 0 ; i < NJPEG ; i++ ) {
556 for ( j = 0 ; j < NJPEG ; j++ ) {
557 temp1 = 0.0;
558 for ( k = 0 ; k < NJPEG ; k++ )
559 temp1 += C[ i ][ k ] * temp[ k ][ j ];
560 output[ i ][ j ] = temp1;
561 }
562 }
563 }
564
565 void fdct_block_8x8(gray **input, int col, int row, double **output) {
566 int i;
567 gray *input_array[NJPEG];
568
569 for (i = 0; i < NJPEG; i++)
570 input_array[i] = &input[row + i][col];
571
572 fdct_8x8(input_array, output);
573 }
574
575 /*
576 * The Inverse DCT routine implements the matrix function:
577 *
578 * pixels = C * DCT * Ct
579 */
580
581 void idct_8x8(double **input, gray **output) {
582 double temp[ NJPEG ][ NJPEG ];
583 double temp1;
584 int i;
585 int j;
586 int k;
587
588 /* MatrixMultiply( temp, input, C ); */
589 for ( i = 0 ; i < NJPEG ; i++ ) {
590 for ( j = 0 ; j < NJPEG ; j++ ) {
591 temp[ i ][ j ] = 0.0;
592 for ( k = 0 ; k < NJPEG ; k++ )
593 temp[ i ][ j ] += input[ i ][ k ] * C[ k ][ j ];
594 }
595 }
596
597 /* MatrixMultiply( output, Ct, temp ); */
598 for ( i = 0 ; i < NJPEG ; i++ ) {
599 for ( j = 0 ; j < NJPEG ; j++ ) {
600 temp1 = 0.0;
601 for ( k = 0 ; k < NJPEG ; k++ )
602 temp1 += Ct[ i ][ k ] * temp[ k ][ j ];
603 temp1 += 128.0;
604 output[i][j] = PIXELRANGE(ROUND(temp1));
605 }
606 }
607 }
608
609 void idct_block_8x8(double **input, gray **output, int col, int row) {
610 int i;
611 gray *output_array[NJPEG];
612
613 for (i = 0; i < NJPEG; i++)
614 output_array[i] = &output[row + i][col];
615
616 idct_8x8(input, output_array);
617 }
618
619 int is_middle_frequency_coeff_8x8(int coeff) {
620 switch (coeff) {
621 case 3:
622 case 10:
623 case 17:
624 case 24:
625 return 1;
626 case 4:
627 case 11:
628 case 18:
629 case 25:
630 case 32:
631 return 2;
632 case 5:
633 case 12:
634 case 19:
635 case 26:
636 case 33:
637 case 40:
638 return 3;
639 case 13:
640 case 20:
641 case 27:
642 case 34:
643 case 41:
644 return 4;
645 case 28:
646 case 35:
647 return 5;
648 default:
649 return 0;
650 }
651 }

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