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