2 * Copyright (c) 2001-2003, David Janssens
3 * Copyright (c) 2002-2003, Yannick Verschueren
4 * Copyright (c) 2003-2005, Francois Devaux and Antonin Descampe
5 * Copyright (c) 2005, Hervé Drolon, FreeImage Team
6 * Copyright (c) 2002-2005, Communications and remote sensing Laboratory, Universite catholique de Louvain, Belgium
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS `AS IS'
19 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
22 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGE.
33 * This is a modified version of the openjpeg dwt.c file.
34 * Average speed improvement compared to the original file (measured on
35 * my own machine, a P4 running at 3.0 GHz):
36 * 5x3 wavelets about 2 times faster
37 * 9x7 wavelets about 3 times faster
38 * for both, encoding and decoding.
40 * The better performance is caused by doing the 1-dimensional DWT
41 * within a temporary buffer where the data can be accessed sequential
42 * for both directions, horizontal and vertical. The 2d vertical DWT was
43 * the major bottleneck in the former version.
45 * I have also removed the "Add Patrick" part because it is not longer
49 * -Ive (aka Reiner Wahler)
50 * mail: ive@lilysoft.com
53 #include "opj_includes.h"
55 /** @defgroup DWT DWT - Implementation of a discrete wavelet transform */
58 /** @name Local static functions */
62 Forward lazy transform (horizontal)
64 static void dwt_deinterleave_h(int *a, int *b, int dn, int sn, int cas);
66 Forward lazy transform (vertical)
68 static void dwt_deinterleave_v(int *a, int *b, int dn, int sn, int x, int cas);
70 Inverse lazy transform (horizontal)
72 static void dwt_interleave_h(int *a, int *b, int dn, int sn, int cas);
74 Inverse lazy transform (vertical)
76 static void dwt_interleave_v(int *a, int *b, int dn, int sn, int x, int cas);
78 Forward 5-3 wavelet tranform in 1-D
80 static void dwt_encode_1(int *a, int dn, int sn, int cas);
82 Inverse 5-3 wavelet tranform in 1-D
84 static void dwt_decode_1(int *a, int dn, int sn, int cas);
86 Forward 9-7 wavelet transform in 1-D
88 static void dwt_encode_1_real(int *a, int dn, int sn, int cas);
90 Inverse 9-7 wavelet transform in 1-D
92 static void dwt_decode_1_real(int *a, int dn, int sn, int cas);
96 static void dwt_encode_stepsize(int stepsize, int numbps, opj_stepsize_t *bandno_stepsize);
102 #define S(i) a[(i)*2]
103 #define D(i) a[(1+(i)*2)]
104 #define S_(i) ((i)<0?S(0):((i)>=sn?S(sn-1):S(i)))
105 #define D_(i) ((i)<0?D(0):((i)>=dn?D(dn-1):D(i)))
107 #define SS_(i) ((i)<0?S(0):((i)>=dn?S(dn-1):S(i)))
108 #define DD_(i) ((i)<0?D(0):((i)>=sn?D(sn-1):D(i)))
111 /* This table contains the norms of the 5-3 wavelets for different bands. */
113 static const double dwt_norms[4][10] = {
114 {1.000, 1.500, 2.750, 5.375, 10.68, 21.34, 42.67, 85.33, 170.7, 341.3},
115 {1.038, 1.592, 2.919, 5.703, 11.33, 22.64, 45.25, 90.48, 180.9},
116 {1.038, 1.592, 2.919, 5.703, 11.33, 22.64, 45.25, 90.48, 180.9},
117 {.7186, .9218, 1.586, 3.043, 6.019, 12.01, 24.00, 47.97, 95.93}
121 /* This table contains the norms of the 9-7 wavelets for different bands. */
123 static const double dwt_norms_real[4][10] = {
124 {1.000, 1.965, 4.177, 8.403, 16.90, 33.84, 67.69, 135.3, 270.6, 540.9},
125 {2.022, 3.989, 8.355, 17.04, 34.27, 68.63, 137.3, 274.6, 549.0},
126 {2.022, 3.989, 8.355, 17.04, 34.27, 68.63, 137.3, 274.6, 549.0},
127 {2.080, 3.865, 8.307, 17.18, 34.71, 69.59, 139.3, 278.6, 557.2}
131 ==========================================================
133 ==========================================================
137 /* Forward lazy transform (horizontal). */
139 static void dwt_deinterleave_h(int *a, int *b, int dn, int sn, int cas) {
141 for (i=0; i<sn; i++) b[i]=a[2*i+cas];
142 for (i=0; i<dn; i++) b[sn+i]=a[(2*i+1-cas)];
146 /* Forward lazy transform (vertical). */
148 static void dwt_deinterleave_v(int *a, int *b, int dn, int sn, int x, int cas) {
150 for (i=0; i<sn; i++) b[i*x]=a[2*i+cas];
151 for (i=0; i<dn; i++) b[(sn+i)*x]=a[(2*i+1-cas)];
155 /* Inverse lazy transform (horizontal). */
157 static void dwt_interleave_h(int *a, int *b, int dn, int sn, int cas) {
163 for (i = 0; i < sn; i++) {
170 for (i = 0; i < dn; i++) {
178 /* Inverse lazy transform (vertical). */
180 static void dwt_interleave_v(int *a, int *b, int dn, int sn, int x, int cas) {
186 for (i = 0; i < sn; i++) {
193 for (i = 0; i < dn; i++) {
202 /* Forward 5-3 wavelet tranform in 1-D. */
204 static void dwt_encode_1(int *a, int dn, int sn, int cas) {
208 if ((dn > 0) || (sn > 1)) { /* NEW : CASE ONE ELEMENT */
209 for (i = 0; i < dn; i++) D(i) -= (S_(i) + S_(i + 1)) >> 1;
210 for (i = 0; i < sn; i++) S(i) += (D_(i - 1) + D_(i) + 2) >> 2;
213 if (!sn && dn == 1) /* NEW : CASE ONE ELEMENT */
216 for (i = 0; i < dn; i++) S(i) -= (DD_(i) + DD_(i - 1)) >> 1;
217 for (i = 0; i < sn; i++) D(i) += (SS_(i) + SS_(i + 1) + 2) >> 2;
223 /* Inverse 5-3 wavelet tranform in 1-D. */
225 static void dwt_decode_1(int *a, int dn, int sn, int cas) {
229 if ((dn > 0) || (sn > 1)) { /* NEW : CASE ONE ELEMENT */
230 for (i = 0; i < sn; i++) S(i) -= (D_(i - 1) + D_(i) + 2) >> 2;
231 for (i = 0; i < dn; i++) D(i) += (S_(i) + S_(i + 1)) >> 1;
234 if (!sn && dn == 1) /* NEW : CASE ONE ELEMENT */
237 for (i = 0; i < sn; i++) D(i) -= (SS_(i) + SS_(i + 1) + 2) >> 2;
238 for (i = 0; i < dn; i++) S(i) += (DD_(i) + DD_(i - 1)) >> 1;
244 /* Forward 9-7 wavelet transform in 1-D. */
246 static void dwt_encode_1_real(int *a, int dn, int sn, int cas) {
249 if ((dn > 0) || (sn > 1)) { /* NEW : CASE ONE ELEMENT */
250 for (i = 0; i < dn; i++)
251 D(i) -= fix_mul(S_(i) + S_(i + 1), 12993);
252 for (i = 0; i < sn; i++)
253 S(i) -= fix_mul(D_(i - 1) + D_(i), 434);
254 for (i = 0; i < dn; i++)
255 D(i) += fix_mul(S_(i) + S_(i + 1), 7233);
256 for (i = 0; i < sn; i++)
257 S(i) += fix_mul(D_(i - 1) + D_(i), 3633);
258 for (i = 0; i < dn; i++)
259 D(i) = fix_mul(D(i), 5038); /*5038 */
260 for (i = 0; i < sn; i++)
261 S(i) = fix_mul(S(i), 6659); /*6660 */
264 if ((sn > 0) || (dn > 1)) { /* NEW : CASE ONE ELEMENT */
265 for (i = 0; i < dn; i++)
266 S(i) -= fix_mul(DD_(i) + DD_(i - 1), 12993);
267 for (i = 0; i < sn; i++)
268 D(i) -= fix_mul(SS_(i) + SS_(i + 1), 434);
269 for (i = 0; i < dn; i++)
270 S(i) += fix_mul(DD_(i) + DD_(i - 1), 7233);
271 for (i = 0; i < sn; i++)
272 D(i) += fix_mul(SS_(i) + SS_(i + 1), 3633);
273 for (i = 0; i < dn; i++)
274 S(i) = fix_mul(S(i), 5038); /*5038 */
275 for (i = 0; i < sn; i++)
276 D(i) = fix_mul(D(i), 6659); /*6660 */
282 /* Inverse 9-7 wavelet transform in 1-D. */
284 static void dwt_decode_1_real(int *a, int dn, int sn, int cas) {
287 if ((dn > 0) || (sn > 1)) { /* NEW : CASE ONE ELEMENT */
288 for (i = 0; i < sn; i++)
289 S(i) = fix_mul(S(i), 10078); /* 10076 */
290 for (i = 0; i < dn; i++)
291 D(i) = fix_mul(D(i), 13318); /* 13320 */
292 for (i = 0; i < sn; i++)
293 S(i) -= fix_mul(D_(i - 1) + D_(i), 3633);
294 for (i = 0; i < dn; i++)
295 D(i) -= fix_mul(S_(i) + S_(i + 1), 7233);
296 for (i = 0; i < sn; i++)
297 S(i) += fix_mul(D_(i - 1) + D_(i), 434);
298 for (i = 0; i < dn; i++)
299 D(i) += fix_mul(S_(i) + S_(i + 1), 12994); /* 12993 */
302 if ((sn > 0) || (dn > 1)) { /* NEW : CASE ONE ELEMENT */
303 for (i = 0; i < sn; i++)
304 D(i) = fix_mul(D(i), 10078); /* 10076 */
305 for (i = 0; i < dn; i++)
306 S(i) = fix_mul(S(i), 13318); /* 13320 */
307 for (i = 0; i < sn; i++)
308 D(i) -= fix_mul(SS_(i) + SS_(i + 1), 3633);
309 for (i = 0; i < dn; i++)
310 S(i) -= fix_mul(DD_(i) + DD_(i - 1), 7233);
311 for (i = 0; i < sn; i++)
312 D(i) += fix_mul(SS_(i) + SS_(i + 1), 434);
313 for (i = 0; i < dn; i++)
314 S(i) += fix_mul(DD_(i) + DD_(i - 1), 12994); /* 12993 */
319 static void dwt_encode_stepsize(int stepsize, int numbps, opj_stepsize_t *bandno_stepsize) {
321 p = int_floorlog2(stepsize) - 13;
322 n = 11 - int_floorlog2(stepsize);
323 bandno_stepsize->mant = (n < 0 ? stepsize >> -n : stepsize << n) & 0x7ff;
324 bandno_stepsize->expn = numbps - p;
328 ==========================================================
330 ==========================================================
334 /* Forward 5-3 wavelet tranform in 2-D. */
336 void dwt_encode(opj_tcd_tilecomp_t * tilec) {
343 w = tilec->x1-tilec->x0;
344 l = tilec->numresolutions-1;
347 for (i = 0; i < l; i++) {
348 int rw; /* width of the resolution level computed */
349 int rh; /* heigth of the resolution level computed */
350 int rw1; /* width of the resolution level once lower than computed one */
351 int rh1; /* height of the resolution level once lower than computed one */
352 int cas_col; /* 0 = non inversion on horizontal filtering 1 = inversion between low-pass and high-pass filtering */
353 int cas_row; /* 0 = non inversion on vertical filtering 1 = inversion between low-pass and high-pass filtering */
356 rw = tilec->resolutions[l - i].x1 - tilec->resolutions[l - i].x0;
357 rh = tilec->resolutions[l - i].y1 - tilec->resolutions[l - i].y0;
358 rw1= tilec->resolutions[l - i - 1].x1 - tilec->resolutions[l - i - 1].x0;
359 rh1= tilec->resolutions[l - i - 1].y1 - tilec->resolutions[l - i - 1].y0;
361 cas_row = tilec->resolutions[l - i].x0 % 2;
362 cas_col = tilec->resolutions[l - i].y0 % 2;
366 bj = (int*)opj_malloc(rh * sizeof(int));
367 for (j = 0; j < rw; j++) {
369 for (k = 0; k < rh; k++) bj[k] = aj[k*w];
370 dwt_encode_1(bj, dn, sn, cas_col);
371 dwt_deinterleave_v(bj, aj, dn, sn, w, cas_col);
377 bj = (int*)opj_malloc(rw * sizeof(int));
378 for (j = 0; j < rh; j++) {
380 for (k = 0; k < rw; k++) bj[k] = aj[k];
381 dwt_encode_1(bj, dn, sn, cas_row);
382 dwt_deinterleave_h(bj, aj, dn, sn, cas_row);
390 /* Inverse 5-3 wavelet tranform in 2-D. */
392 void dwt_decode(opj_tcd_tilecomp_t * tilec, int stop) {
399 w = tilec->x1-tilec->x0;
400 l = tilec->numresolutions-1;
403 for (i = l - 1; i >= stop; i--) {
404 int rw; /* width of the resolution level computed */
405 int rh; /* heigth of the resolution level computed */
406 int rw1; /* width of the resolution level once lower than computed one */
407 int rh1; /* height of the resolution level once lower than computed one */
408 int cas_col; /* 0 = non inversion on horizontal filtering 1 = inversion between low-pass and high-pass filtering */
409 int cas_row; /* 0 = non inversion on vertical filtering 1 = inversion between low-pass and high-pass filtering */
412 rw = tilec->resolutions[l - i].x1 - tilec->resolutions[l - i].x0;
413 rh = tilec->resolutions[l - i].y1 - tilec->resolutions[l - i].y0;
414 rw1= tilec->resolutions[l - i - 1].x1 - tilec->resolutions[l - i - 1].x0;
415 rh1= tilec->resolutions[l - i - 1].y1 - tilec->resolutions[l - i - 1].y0;
417 cas_row = tilec->resolutions[l - i].x0 % 2;
418 cas_col = tilec->resolutions[l - i].y0 % 2;
422 bj = (int*)opj_malloc(rw * sizeof(int));
423 for (j = 0; j < rh; j++) {
425 dwt_interleave_h(aj, bj, dn, sn, cas_row);
426 dwt_decode_1(bj, dn, sn, cas_row);
427 for (k = 0; k < rw; k++) aj[k] = bj[k];
433 bj = (int*)opj_malloc(rh * sizeof(int));
434 for (j = 0; j < rw; j++) {
436 dwt_interleave_v(aj, bj, dn, sn, w, cas_col);
437 dwt_decode_1(bj, dn, sn, cas_col);
438 for (k = 0; k < rh; k++) aj[k * w] = bj[k];
446 /* Get gain of 5-3 wavelet transform. */
448 int dwt_getgain(int orient) {
451 if (orient == 1 || orient == 2)
457 /* Get norm of 5-3 wavelet. */
459 double dwt_getnorm(int level, int orient) {
460 return dwt_norms[orient][level];
464 /* Forward 9-7 wavelet transform in 2-D. */
467 void dwt_encode_real(opj_tcd_tilecomp_t * tilec) {
474 w = tilec->x1-tilec->x0;
475 l = tilec->numresolutions-1;
478 for (i = 0; i < l; i++) {
479 int rw; /* width of the resolution level computed */
480 int rh; /* heigth of the resolution level computed */
481 int rw1; /* width of the resolution level once lower than computed one */
482 int rh1; /* height of the resolution level once lower than computed one */
483 int cas_col; /* 0 = non inversion on horizontal filtering 1 = inversion between low-pass and high-pass filtering */
484 int cas_row; /* 0 = non inversion on vertical filtering 1 = inversion between low-pass and high-pass filtering */
487 rw = tilec->resolutions[l - i].x1 - tilec->resolutions[l - i].x0;
488 rh = tilec->resolutions[l - i].y1 - tilec->resolutions[l - i].y0;
489 rw1= tilec->resolutions[l - i - 1].x1 - tilec->resolutions[l - i - 1].x0;
490 rh1= tilec->resolutions[l - i - 1].y1 - tilec->resolutions[l - i - 1].y0;
492 cas_row = tilec->resolutions[l - i].x0 % 2;
493 cas_col = tilec->resolutions[l - i].y0 % 2;
497 bj = (int*)opj_malloc(rh * sizeof(int));
498 for (j = 0; j < rw; j++) {
500 for (k = 0; k < rh; k++) bj[k] = aj[k*w];
501 dwt_encode_1_real(bj, dn, sn, cas_col);
502 dwt_deinterleave_v(bj, aj, dn, sn, w, cas_col);
508 bj = (int*)opj_malloc(rw * sizeof(int));
509 for (j = 0; j < rh; j++) {
511 for (k = 0; k < rw; k++) bj[k] = aj[k];
512 dwt_encode_1_real(bj, dn, sn, cas_row);
513 dwt_deinterleave_h(bj, aj, dn, sn, cas_row);
521 /* Inverse 9-7 wavelet transform in 2-D. */
523 void dwt_decode_real(opj_tcd_tilecomp_t * tilec, int stop) {
530 w = tilec->x1-tilec->x0;
531 l = tilec->numresolutions-1;
534 for (i = l-1; i >= stop; i--) {
535 int rw; /* width of the resolution level computed */
536 int rh; /* heigth of the resolution level computed */
537 int rw1; /* width of the resolution level once lower than computed one */
538 int rh1; /* height of the resolution level once lower than computed one */
539 int cas_col; /* 0 = non inversion on horizontal filtering 1 = inversion between low-pass and high-pass filtering */
540 int cas_row; /* 0 = non inversion on vertical filtering 1 = inversion between low-pass and high-pass filtering */
543 rw = tilec->resolutions[l - i].x1 - tilec->resolutions[l - i].x0;
544 rh = tilec->resolutions[l - i].y1 - tilec->resolutions[l - i].y0;
545 rw1= tilec->resolutions[l - i - 1].x1 - tilec->resolutions[l - i - 1].x0;
546 rh1= tilec->resolutions[l - i - 1].y1 - tilec->resolutions[l - i - 1].y0;
548 cas_col = tilec->resolutions[l - i].x0 % 2; /* 0 = non inversion on horizontal filtering 1 = inversion between low-pass and high-pass filtering */
549 cas_row = tilec->resolutions[l - i].y0 % 2; /* 0 = non inversion on vertical filtering 1 = inversion between low-pass and high-pass filtering */
553 bj = (int*)opj_malloc(rw * sizeof(int));
554 for (j = 0; j < rh; j++) {
556 dwt_interleave_h(aj, bj, dn, sn, cas_col);
557 dwt_decode_1_real(bj, dn, sn, cas_col);
558 for (k = 0; k < rw; k++) aj[k] = bj[k];
564 bj = (int*)opj_malloc(rh * sizeof(int));
565 for (j = 0; j < rw; j++) {
567 dwt_interleave_v(aj, bj, dn, sn, w, cas_row);
568 dwt_decode_1_real(bj, dn, sn, cas_row);
569 for (k = 0; k < rh; k++) aj[k * w] = bj[k];
577 /* Get gain of 9-7 wavelet transform. */
579 int dwt_getgain_real(int orient) {
585 /* Get norm of 9-7 wavelet. */
587 double dwt_getnorm_real(int level, int orient) {
588 return dwt_norms_real[orient][level];
591 void dwt_calc_explicit_stepsizes(opj_tccp_t * tccp, int prec) {
592 int numbands, bandno;
593 numbands = 3 * tccp->numresolutions - 2;
594 for (bandno = 0; bandno < numbands; bandno++) {
596 int resno, level, orient, gain;
598 resno = (bandno == 0) ? 0 : ((bandno - 1) / 3 + 1);
599 orient = (bandno == 0) ? 0 : ((bandno - 1) % 3 + 1);
600 level = tccp->numresolutions - 1 - resno;
601 gain = (tccp->qmfbid == 0) ? 0 : ((orient == 0) ? 0 : (((orient == 1) || (orient == 2)) ? 1 : 2));
602 if (tccp->qntsty == J2K_CCP_QNTSTY_NOQNT) {
605 double norm = dwt_norms_real[orient][level];
606 stepsize = (1 << (gain + 1)) / norm;
608 dwt_encode_stepsize((int) floor(stepsize * 8192.0), prec + gain, &tccp->stepsizes[bandno]);