2 * JPEG 2000 image decoder
3 * Copyright (c) 2007 Kamil Nowosad
4 * Copyright (c) 2013 Nicolas Bertrand <nicoinattendu@gmail.com>
6 * This file is part of Libav.
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * JPEG 2000 image decoder
28 #include "libavutil/common.h"
29 #include "libavutil/opt.h"
31 #include "bytestream.h"
36 #define JP2_SIG_TYPE 0x6A502020
37 #define JP2_SIG_VALUE 0x0D0A870A
38 #define JP2_CODESTREAM 0x6A703263
43 typedef struct Jpeg2000TilePart {
44 uint16_t tp_idx; // Tile-part index
45 uint8_t tile_index; // Tile index who refers the tile-part
46 uint32_t tp_len; // Length of tile-part
47 const uint8_t *tp_start_bstrm; // Start address bit stream in tile-part
48 const uint8_t *tp_end_bstrm; // End address of the bit stream tile part
51 /* RMK: For JPEG2000 DCINEMA 3 tile-parts in a tile
52 * one per component, so tile_part elements have a size of 3 */
53 typedef struct Jpeg2000Tile {
54 Jpeg2000Component *comp;
55 uint8_t properties[4];
56 Jpeg2000CodingStyle codsty[4];
57 Jpeg2000QuantStyle qntsty[4];
58 Jpeg2000TilePart tile_part[3];
61 typedef struct Jpeg2000DecoderContext {
63 AVCodecContext *avctx;
66 int image_offset_x, image_offset_y;
67 int tile_offset_x, tile_offset_y;
68 uint8_t cbps[4]; // bits per sample in particular components
69 uint8_t sgnd[4]; // if a component is signed
70 uint8_t properties[4];
74 int tile_width, tile_height;
75 int numXtiles, numYtiles;
78 Jpeg2000CodingStyle codsty[4];
79 Jpeg2000QuantStyle qntsty[4];
81 const uint8_t *buf_start;
83 const uint8_t *buf_end;
89 /*options parameters*/
91 int16_t reduction_factor;
92 } Jpeg2000DecoderContext;
94 /* get_bits functions for JPEG2000 packet bitstream
95 * It is a get_bit function with a bit-stuffing routine. If the value of the
96 * byte is 0xFF, the next byte includes an extra zero bit stuffed into the MSB.
97 * cf. ISO-15444-1:2002 / B.10.1 Bit-stuffing routine */
98 static int get_bits(Jpeg2000DecoderContext *s, int n)
101 if (s->buf_end - s->buf < ((n - s->bit_index) >> 8))
102 return AVERROR_INVALIDDATA;
105 if (s->bit_index == 0) {
106 s->bit_index = 7 + (*s->buf != 0xff);
110 res |= (*s->buf >> s->bit_index) & 1;
115 static void jpeg2000_flush(Jpeg2000DecoderContext *s)
123 /* decode the value stored in node */
124 static int tag_tree_decode(Jpeg2000DecoderContext *s, Jpeg2000TgtNode *node,
127 Jpeg2000TgtNode *stack[30];
128 int sp = -1, curval = 0;
130 while (node && !node->vis) {
138 curval = stack[sp]->val;
140 while (curval < threshold && sp >= 0) {
141 if (curval < stack[sp]->val)
142 curval = stack[sp]->val;
143 while (curval < threshold) {
145 if ((ret = get_bits(s, 1)) > 0) {
153 stack[sp]->val = curval;
159 /* marker segments */
160 /* get sizes and offsets of image, tiles; number of components */
161 static int get_siz(Jpeg2000DecoderContext *s)
165 if (s->buf_end - s->buf < 36)
166 return AVERROR_INVALIDDATA;
168 s->avctx->profile = bytestream_get_be16(&s->buf); // Rsiz
169 s->width = bytestream_get_be32(&s->buf); // Width
170 s->height = bytestream_get_be32(&s->buf); // Height
171 s->image_offset_x = bytestream_get_be32(&s->buf); // X0Siz
172 s->image_offset_y = bytestream_get_be32(&s->buf); // Y0Siz
173 s->tile_width = bytestream_get_be32(&s->buf); // XTSiz
174 s->tile_height = bytestream_get_be32(&s->buf); // YTSiz
175 s->tile_offset_x = bytestream_get_be32(&s->buf); // XT0Siz
176 s->tile_offset_y = bytestream_get_be32(&s->buf); // YT0Siz
177 s->ncomponents = bytestream_get_be16(&s->buf); // CSiz
179 if (s->buf_end - s->buf < 2 * s->ncomponents)
180 return AVERROR_INVALIDDATA;
182 for (i = 0; i < s->ncomponents; i++) { // Ssiz_i XRsiz_i, YRsiz_i
183 uint8_t x = bytestream_get_byte(&s->buf);
184 s->cbps[i] = (x & 0x7f) + 1;
185 s->precision = FFMAX(s->cbps[i], s->precision);
186 s->sgnd[i] = (x & 0x80) == 1;
187 s->cdx[i] = bytestream_get_byte(&s->buf);
188 s->cdy[i] = bytestream_get_byte(&s->buf);
191 s->numXtiles = ff_jpeg2000_ceildiv(s->width - s->tile_offset_x, s->tile_width);
192 s->numYtiles = ff_jpeg2000_ceildiv(s->height - s->tile_offset_y, s->tile_height);
194 s->tile = av_mallocz(s->numXtiles * s->numYtiles * sizeof(*s->tile));
196 return AVERROR(ENOMEM);
198 for (i = 0; i < s->numXtiles * s->numYtiles; i++) {
199 Jpeg2000Tile *tile = s->tile + i;
201 tile->comp = av_mallocz(s->ncomponents * sizeof(*tile->comp));
203 return AVERROR(ENOMEM);
206 /* compute image size with reduction factor */
207 s->avctx->width = ff_jpeg2000_ceildivpow2(s->width - s->image_offset_x,
208 s->reduction_factor);
209 s->avctx->height = ff_jpeg2000_ceildivpow2(s->height - s->image_offset_y,
210 s->reduction_factor);
212 switch (s->avctx->profile) {
213 case FF_PROFILE_JPEG2000_DCINEMA_2K:
214 case FF_PROFILE_JPEG2000_DCINEMA_4K:
215 /* XYZ color-space for digital cinema profiles */
216 s->avctx->pix_fmt = AV_PIX_FMT_XYZ12;
219 /* For other profiles selects color-space according number of
220 * components and bit depth precision. */
221 switch (s->ncomponents) {
223 if (s->precision > 8)
224 s->avctx->pix_fmt = AV_PIX_FMT_GRAY16;
226 s->avctx->pix_fmt = AV_PIX_FMT_GRAY8;
229 if (s->precision > 8)
230 s->avctx->pix_fmt = AV_PIX_FMT_RGB48;
232 s->avctx->pix_fmt = AV_PIX_FMT_RGB24;
235 s->avctx->pix_fmt = AV_PIX_FMT_BGRA;
238 /* pixel format can not be identified */
239 s->avctx->pix_fmt = AV_PIX_FMT_NONE;
247 /* get common part for COD and COC segments */
248 static int get_cox(Jpeg2000DecoderContext *s, Jpeg2000CodingStyle *c)
252 if (s->buf_end - s->buf < 5)
253 return AVERROR(EINVAL);
254 /* nreslevels = number of resolution levels
255 = number of decomposition level +1 */
256 c->nreslevels = bytestream_get_byte(&s->buf) + 1;
258 if (c->nreslevels > JPEG2000_MAX_RESLEVELS)
259 return AVERROR_INVALIDDATA;
261 /* compute number of resolution levels to decode */
262 if (c->nreslevels < s->reduction_factor)
263 c->nreslevels2decode = 1;
265 c->nreslevels2decode = c->nreslevels - s->reduction_factor;
267 c->log2_cblk_width = bytestream_get_byte(&s->buf) + 2; // cblk width
268 c->log2_cblk_height = bytestream_get_byte(&s->buf) + 2; // cblk height
270 if (c->log2_cblk_width > 10 || c->log2_cblk_height > 10 ||
271 c->log2_cblk_width + c->log2_cblk_height > 12) {
272 av_log(s->avctx, AV_LOG_ERROR, "cblk size invalid\n");
273 return AVERROR_INVALIDDATA;
276 c->cblk_style = bytestream_get_byte(&s->buf);
277 if (c->cblk_style != 0) { // cblk style
278 avpriv_request_sample(s->avctx, "Support for extra cblk styles");
279 return AVERROR_PATCHWELCOME;
281 c->transform = bytestream_get_byte(&s->buf); // DWT transformation type
282 /* set integer 9/7 DWT in case of BITEXACT flag */
283 if ((s->avctx->flags & CODEC_FLAG_BITEXACT) && (c->transform == FF_DWT97))
284 c->transform = FF_DWT97_INT;
286 if (c->csty & JPEG2000_CSTY_PREC) {
288 for (i = 0; i < c->nreslevels; i++) {
289 byte = bytestream_get_byte(&s->buf);
290 c->log2_prec_widths[i] = byte & 0x0F; // precinct PPx
291 c->log2_prec_heights[i] = (byte >> 4) & 0x0F; // precinct PPy
297 /* get coding parameters for a particular tile or whole image*/
298 static int get_cod(Jpeg2000DecoderContext *s, Jpeg2000CodingStyle *c,
301 Jpeg2000CodingStyle tmp;
304 if (s->buf_end - s->buf < 5)
305 return AVERROR_INVALIDDATA;
307 tmp.log2_prec_width =
308 tmp.log2_prec_height = 15;
310 tmp.csty = bytestream_get_byte(&s->buf);
312 // get progression order
313 tmp.prog_order = bytestream_get_byte(&s->buf);
315 tmp.nlayers = bytestream_get_be16(&s->buf);
316 tmp.mct = bytestream_get_byte(&s->buf); // multiple component transformation
319 for (compno = 0; compno < s->ncomponents; compno++)
320 if (!(properties[compno] & HAD_COC))
321 memcpy(c + compno, &tmp, sizeof(tmp));
325 /* Get coding parameters for a component in the whole image or a
326 * particular tile. */
327 static int get_coc(Jpeg2000DecoderContext *s, Jpeg2000CodingStyle *c,
332 if (s->buf_end - s->buf < 2)
333 return AVERROR_INVALIDDATA;
335 compno = bytestream_get_byte(&s->buf);
338 c->csty = bytestream_get_byte(&s->buf);
341 properties[compno] |= HAD_COC;
345 /* Get common part for QCD and QCC segments. */
346 static int get_qcx(Jpeg2000DecoderContext *s, int n, Jpeg2000QuantStyle *q)
350 if (s->buf_end - s->buf < 1)
351 return AVERROR_INVALIDDATA;
353 x = bytestream_get_byte(&s->buf); // Sqcd
355 q->nguardbits = x >> 5;
356 q->quantsty = x & 0x1f;
358 if (q->quantsty == JPEG2000_QSTY_NONE) {
360 if (s->buf_end - s->buf < n)
361 return AVERROR_INVALIDDATA;
362 for (i = 0; i < n; i++)
363 q->expn[i] = bytestream_get_byte(&s->buf) >> 3;
364 } else if (q->quantsty == JPEG2000_QSTY_SI) {
365 if (s->buf_end - s->buf < 2)
366 return AVERROR_INVALIDDATA;
367 x = bytestream_get_be16(&s->buf);
368 q->expn[0] = x >> 11;
369 q->mant[0] = x & 0x7ff;
370 for (i = 1; i < JPEG2000_MAX_DECLEVELS * 3; i++) {
371 int curexpn = FFMAX(0, q->expn[0] - (i - 1) / 3);
372 q->expn[i] = curexpn;
373 q->mant[i] = q->mant[0];
377 if (s->buf_end - s->buf < n)
378 return AVERROR_INVALIDDATA;
379 for (i = 0; i < n; i++) {
380 x = bytestream_get_be16(&s->buf);
381 q->expn[i] = x >> 11;
382 q->mant[i] = x & 0x7ff;
388 /* Get quantization parameters for a particular tile or a whole image. */
389 static int get_qcd(Jpeg2000DecoderContext *s, int n, Jpeg2000QuantStyle *q,
392 Jpeg2000QuantStyle tmp;
395 if ((ret = get_qcx(s, n, &tmp)) < 0)
397 for (compno = 0; compno < s->ncomponents; compno++)
398 if (!(properties[compno] & HAD_QCC))
399 memcpy(q + compno, &tmp, sizeof(tmp));
403 /* Get quantization parameters for a component in the whole image
404 * on in a particular tile. */
405 static int get_qcc(Jpeg2000DecoderContext *s, int n, Jpeg2000QuantStyle *q,
410 if (s->buf_end - s->buf < 1)
411 return AVERROR_INVALIDDATA;
413 compno = bytestream_get_byte(&s->buf);
414 properties[compno] |= HAD_QCC;
415 return get_qcx(s, n - 1, q + compno);
418 /* Get start of tile segment. */
419 static int get_sot(Jpeg2000DecoderContext *s, int n)
421 Jpeg2000TilePart *tp;
426 if (s->buf_end - s->buf < 4)
427 return AVERROR_INVALIDDATA;
429 Isot = bytestream_get_be16(&s->buf); // Isot
431 avpriv_request_sample(s->avctx, "Support for more than one tile");
432 return AVERROR_PATCHWELCOME;
434 Psot = bytestream_get_be32(&s->buf); // Psot
435 TPsot = bytestream_get_byte(&s->buf); // TPsot
437 /* Read TNSot but not used */
438 bytestream_get_byte(&s->buf); // TNsot
440 tp = s->tile[s->curtileno].tile_part + TPsot;
441 tp->tile_index = Isot;
445 /* Start of bit stream. Pointer to SOD marker
446 * Check SOD marker is present. */
447 if (JPEG2000_SOD == bytestream_get_be16(&s->buf))
448 tp->tp_start_bstrm = s->buf;
450 av_log(s->avctx, AV_LOG_ERROR, "SOD marker not found \n");
451 return AVERROR_INVALIDDATA;
454 /* End address of bit stream =
455 * start address + (Psot - size of SOT HEADER(n)
456 * - size of SOT MARKER(2) - size of SOD marker(2) */
457 tp->tp_end_bstrm = s->buf + (tp->tp_len - n - 4);
459 // set buffer pointer to end of tile part header
460 s->buf = tp->tp_end_bstrm;
465 /* Tile-part lengths: see ISO 15444-1:2002, section A.7.1
466 * Used to know the number of tile parts and lengths.
467 * There may be multiple TLMs in the header.
468 * TODO: The function is not used for tile-parts management, nor anywhere else.
469 * It can be useful to allocate memory for tile parts, before managing the SOT
470 * markers. Parsing the TLM header is needed to increment the input header
472 * This marker is mandatory for DCI. */
473 static uint8_t get_tlm(Jpeg2000DecoderContext *s, int n)
475 uint8_t Stlm, ST, SP, tile_tlm, i;
476 bytestream_get_byte(&s->buf); /* Ztlm: skipped */
477 Stlm = bytestream_get_byte(&s->buf);
479 // too complex ? ST = ((Stlm >> 4) & 0x01) + ((Stlm >> 4) & 0x02);
480 ST = (Stlm >> 4) & 0x03;
481 // TODO: Manage case of ST = 0b11 --> raise error
482 SP = (Stlm >> 6) & 0x01;
483 tile_tlm = (n - 4) / ((SP + 1) * 2 + ST);
484 for (i = 0; i < tile_tlm; i++) {
489 bytestream_get_byte(&s->buf);
492 bytestream_get_be16(&s->buf);
495 bytestream_get_be32(&s->buf);
499 bytestream_get_be16(&s->buf);
501 bytestream_get_be32(&s->buf);
507 static int init_tile(Jpeg2000DecoderContext *s, int tileno)
510 int tilex = tileno % s->numXtiles;
511 int tiley = tileno / s->numXtiles;
512 Jpeg2000Tile *tile = s->tile + tileno;
513 Jpeg2000CodingStyle *codsty;
514 Jpeg2000QuantStyle *qntsty;
517 return AVERROR(ENOMEM);
519 /* copy codsty, qnsty to tile. TODO: Is it the best way?
520 * codsty, qnsty is an array of 4 structs Jpeg2000CodingStyle
521 * and Jpeg2000QuantStyle */
522 memcpy(tile->codsty, s->codsty, s->ncomponents * sizeof(*codsty));
523 memcpy(tile->qntsty, s->qntsty, s->ncomponents * sizeof(*qntsty));
525 for (compno = 0; compno < s->ncomponents; compno++) {
526 Jpeg2000Component *comp = tile->comp + compno;
527 int ret; // global bandno
528 codsty = tile->codsty + compno;
529 qntsty = tile->qntsty + compno;
531 comp->coord_o[0][0] = FFMAX(tilex * s->tile_width + s->tile_offset_x, s->image_offset_x);
532 comp->coord_o[0][1] = FFMIN((tilex + 1) * s->tile_width + s->tile_offset_x, s->width);
533 comp->coord_o[1][0] = FFMAX(tiley * s->tile_height + s->tile_offset_y, s->image_offset_y);
534 comp->coord_o[1][1] = FFMIN((tiley + 1) * s->tile_height + s->tile_offset_y, s->height);
536 // FIXME: add a dcinema profile check ?
537 // value is guaranteed by profile (orig=0, 1 tile)
538 comp->coord[0][0] = 0;
539 comp->coord[0][1] = s->avctx->width;
540 comp->coord[1][0] = 0;
541 comp->coord[1][1] = s->avctx->height;
543 if (ret = ff_jpeg2000_init_component(comp, codsty, qntsty,
544 s->cbps[compno], s->cdx[compno],
545 s->cdy[compno], s->avctx))
551 /* Read the number of coding passes. */
552 static int getnpasses(Jpeg2000DecoderContext *s)
559 if ((num = get_bits(s, 2)) != 3)
560 return num < 0 ? num : 3 + num;
561 if ((num = get_bits(s, 5)) != 31)
562 return num < 0 ? num : 6 + num;
563 num = get_bits(s, 7);
564 return num < 0 ? num : 37 + num;
567 static int getlblockinc(Jpeg2000DecoderContext *s)
570 while (ret = get_bits(s, 1)) {
578 static int jpeg2000_decode_packet(Jpeg2000DecoderContext *s,
579 Jpeg2000CodingStyle *codsty,
580 Jpeg2000ResLevel *rlevel, int precno,
581 int layno, uint8_t *expn, int numgbits)
583 int bandno, cblkno, ret, nb_code_blocks;
585 if (!(ret = get_bits(s, 1))) {
591 for (bandno = 0; bandno < rlevel->nbands; bandno++) {
592 Jpeg2000Band *band = rlevel->band + bandno;
593 Jpeg2000Prec *prec = band->prec + precno;
595 if (band->coord[0][0] == band->coord[0][1] ||
596 band->coord[1][0] == band->coord[1][1])
600 nb_code_blocks = prec->nb_codeblocks_height *
601 prec->nb_codeblocks_width;
602 for (cblkno = 0; cblkno < nb_code_blocks; cblkno++) {
603 Jpeg2000Cblk *cblk = prec->cblk + cblkno;
604 int incl, newpasses, llen;
607 incl = get_bits(s, 1);
609 incl = tag_tree_decode(s, prec->cblkincl + cblkno, layno + 1) == layno;
616 cblk->nonzerobits = expn[bandno] + numgbits - 1 -
617 tag_tree_decode(s, prec->zerobits + cblkno,
619 if ((newpasses = getnpasses(s)) < 0)
621 if ((llen = getlblockinc(s)) < 0)
623 cblk->lblock += llen;
624 if ((ret = get_bits(s, av_log2(newpasses) + cblk->lblock)) < 0)
626 cblk->lengthinc = ret;
627 cblk->npasses += newpasses;
632 if (codsty->csty & JPEG2000_CSTY_EPH) {
633 if (AV_RB16(s->buf) == JPEG2000_EPH)
636 av_log(s->avctx, AV_LOG_ERROR, "EPH marker not found.\n");
639 for (bandno = 0; bandno < rlevel->nbands; bandno++) {
640 Jpeg2000Band *band = rlevel->band + bandno;
641 Jpeg2000Prec *prec = band->prec + precno;
643 nb_code_blocks = prec->nb_codeblocks_height * prec->nb_codeblocks_width;
644 for (cblkno = 0; cblkno < nb_code_blocks; cblkno++) {
645 Jpeg2000Cblk *cblk = prec->cblk + cblkno;
646 if (s->buf_end - s->buf < cblk->lengthinc)
647 return AVERROR_INVALIDDATA;
648 /* Code-block data can be empty. In that case initialize data
650 if (cblk->lengthinc > 0) {
651 bytestream_get_buffer(&s->buf, cblk->data, cblk->lengthinc);
653 cblk->data[0] = 0xFF;
654 cblk->data[1] = 0xFF;
656 cblk->length += cblk->lengthinc;
663 static int jpeg2000_decode_packets(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile)
665 int layno, reslevelno, compno, precno, ok_reslevel, ret;
666 uint8_t prog_order = tile->codsty[0].prog_order;
671 switch (prog_order) {
672 case JPEG2000_PGOD_LRCP:
673 for (layno = 0; layno < tile->codsty[0].nlayers; layno++) {
675 for (reslevelno = 0; ok_reslevel; reslevelno++) {
677 for (compno = 0; compno < s->ncomponents; compno++) {
678 Jpeg2000CodingStyle *codsty = tile->codsty + compno;
679 Jpeg2000QuantStyle *qntsty = tile->qntsty + compno;
680 if (reslevelno < codsty->nreslevels) {
681 Jpeg2000ResLevel *rlevel = tile->comp[compno].reslevel +
684 for (precno = 0; precno < rlevel->num_precincts_x * rlevel->num_precincts_y; precno++)
685 if ((ret = jpeg2000_decode_packet(s,
688 qntsty->expn + (reslevelno ? 3 * (reslevelno - 1) + 1 : 0),
689 qntsty->nguardbits)) < 0)
697 case JPEG2000_PGOD_CPRL:
698 for (compno = 0; compno < s->ncomponents; compno++) {
699 Jpeg2000CodingStyle *codsty = tile->codsty + compno;
700 Jpeg2000QuantStyle *qntsty = tile->qntsty + compno;
702 /* Set bit stream buffer address according to tile-part.
703 * For DCinema one tile-part per component, so can be
704 * indexed by component. */
705 s->buf = tile->tile_part[compno].tp_start_bstrm;
707 /* Position loop (y axis)
708 * TODO: Automate computing of step 256.
709 * Fixed here, but to be computed before entering here. */
710 for (y = 0; y < s->height; y += 256) {
711 /* Position loop (y axis)
712 * TODO: automate computing of step 256.
713 * Fixed here, but to be computed before entering here. */
714 for (x = 0; x < s->width; x += 256) {
715 for (reslevelno = 0; reslevelno < codsty->nreslevels; reslevelno++) {
717 uint8_t reducedresno = codsty->nreslevels - 1 -reslevelno; // ==> N_L - r
718 Jpeg2000ResLevel *rlevel = tile->comp[compno].reslevel + reslevelno;
720 if (!((y % (1 << (rlevel->log2_prec_height + reducedresno)) == 0) ||
721 (y == 0))) // TODO: 2nd condition simplified as try0 always =0 for dcinema
724 if (!((x % (1 << (rlevel->log2_prec_width + reducedresno)) == 0) ||
725 (x == 0))) // TODO: 2nd condition simplified as try0 always =0 for dcinema
728 // check if a precinct exists
729 prcx = ff_jpeg2000_ceildivpow2(x, reducedresno) >> rlevel->log2_prec_width;
730 prcy = ff_jpeg2000_ceildivpow2(y, reducedresno) >> rlevel->log2_prec_height;
731 precno = prcx + rlevel->num_precincts_x * prcy;
732 for (layno = 0; layno < tile->codsty[0].nlayers; layno++) {
733 if ((ret = jpeg2000_decode_packet(s, codsty, rlevel,
735 qntsty->expn + (reslevelno ? 3 * (reslevelno - 1) + 1 : 0),
736 qntsty->nguardbits)) < 0)
749 /* EOC marker reached */
755 /* TIER-1 routines */
756 static void decode_sigpass(Jpeg2000T1Context *t1, int width, int height,
757 int bpno, int bandno)
759 int mask = 3 << (bpno - 1), y0, x, y;
761 for (y0 = 0; y0 < height; y0 += 4)
762 for (x = 0; x < width; x++)
763 for (y = y0; y < height && y < y0 + 4; y++)
764 if ((t1->flags[y + 1][x + 1] & JPEG2000_T1_SIG_NB)
765 && !(t1->flags[y + 1][x + 1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS))) {
766 if (ff_mqc_decode(&t1->mqc,
768 ff_jpeg2000_getsigctxno(t1->flags[y + 1][x + 1],
770 int xorbit, ctxno = ff_jpeg2000_getsgnctxno(t1->flags[y + 1][x + 1],
774 (ff_mqc_decode(&t1->mqc,
775 t1->mqc.cx_states + ctxno) ^ xorbit)
778 ff_jpeg2000_set_significance(t1, x, y,
781 t1->flags[y + 1][x + 1] |= JPEG2000_T1_VIS;
785 static void decode_refpass(Jpeg2000T1Context *t1, int width, int height,
791 phalf = 1 << (bpno - 1);
794 for (y0 = 0; y0 < height; y0 += 4)
795 for (x = 0; x < width; x++)
796 for (y = y0; y < height && y < y0 + 4; y++)
797 if ((t1->flags[y + 1][x + 1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS)) == JPEG2000_T1_SIG) {
798 int ctxno = ff_jpeg2000_getrefctxno(t1->flags[y + 1][x + 1]);
799 int r = ff_mqc_decode(&t1->mqc,
800 t1->mqc.cx_states + ctxno)
802 t1->data[y][x] += t1->data[y][x] < 0 ? -r : r;
803 t1->flags[y + 1][x + 1] |= JPEG2000_T1_REF;
807 static void decode_clnpass(Jpeg2000DecoderContext *s, Jpeg2000T1Context *t1,
808 int width, int height, int bpno, int bandno,
811 int mask = 3 << (bpno - 1), y0, x, y, runlen, dec;
813 for (y0 = 0; y0 < height; y0 += 4)
814 for (x = 0; x < width; x++) {
815 if (y0 + 3 < height &&
816 !((t1->flags[y0 + 1][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
817 (t1->flags[y0 + 2][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
818 (t1->flags[y0 + 3][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
819 (t1->flags[y0 + 4][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)))) {
820 if (!ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_RL))
822 runlen = ff_mqc_decode(&t1->mqc,
823 t1->mqc.cx_states + MQC_CX_UNI);
824 runlen = (runlen << 1) | ff_mqc_decode(&t1->mqc,
833 for (y = y0 + runlen; y < y0 + 4 && y < height; y++) {
835 if (!(t1->flags[y + 1][x + 1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS)))
836 dec = ff_mqc_decode(&t1->mqc,
838 ff_jpeg2000_getsigctxno(t1->flags[y + 1][x + 1],
843 int ctxno = ff_jpeg2000_getsgnctxno(t1->flags[y + 1][x + 1],
845 t1->data[y][x] = (ff_mqc_decode(&t1->mqc,
846 t1->mqc.cx_states + ctxno) ^
849 ff_jpeg2000_set_significance(t1, x, y, t1->data[y][x] < 0);
852 t1->flags[y + 1][x + 1] &= ~JPEG2000_T1_VIS;
857 val = ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_UNI);
858 val = (val << 1) + ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_UNI);
859 val = (val << 1) + ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_UNI);
860 val = (val << 1) + ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_UNI);
862 av_log(s->avctx, AV_LOG_ERROR,
863 "Segmentation symbol value incorrect\n");
867 static int decode_cblk(Jpeg2000DecoderContext *s, Jpeg2000CodingStyle *codsty,
868 Jpeg2000T1Context *t1, Jpeg2000Cblk *cblk,
869 int width, int height, int bandpos)
871 int passno = cblk->npasses, pass_t = 2, bpno = cblk->nonzerobits - 1, y;
873 for (y = 0; y < height; y++)
874 memset(t1->data[y], 0, width * sizeof(width));
875 /* If code-block contains no compressed data: nothing to do. */
878 for (y = 0; y < height + 2; y++)
879 memset(t1->flags[y], 0, (width + 2) * sizeof(width));
881 ff_mqc_initdec(&t1->mqc, cblk->data);
882 cblk->data[cblk->length] = 0xff;
883 cblk->data[cblk->length + 1] = 0xff;
888 decode_sigpass(t1, width, height, bpno + 1, bandpos);
891 decode_refpass(t1, width, height, bpno + 1);
894 decode_clnpass(s, t1, width, height, bpno + 1, bandpos,
895 codsty->cblk_style & JPEG2000_CBLK_SEGSYM);
908 /* TODO: Verify dequantization for lossless case
909 * comp->data can be float or int
910 * band->stepsize can be float or int
911 * depending on the type of DWT transformation.
912 * see ISO/IEC 15444-1:2002 A.6.1 */
914 /* Float dequantization of a codeblock.*/
915 static void dequantization_float(int x, int y, Jpeg2000Cblk *cblk,
916 Jpeg2000Component *comp,
917 Jpeg2000T1Context *t1, Jpeg2000Band *band)
920 float *datap = &comp->data[(comp->coord[0][1] - comp->coord[0][0]) * y + x];
921 for (j = 0; j < (cblk->coord[1][1] - cblk->coord[1][0]); ++j)
922 for (i = 0; i < (cblk->coord[0][1] - cblk->coord[0][0]); ++i) {
923 idx = (comp->coord[0][1] - comp->coord[0][0]) * j + i;
924 datap[idx] = (float)(t1->data[j][i]) * ((float)band->stepsize);
929 /* Integer dequantization of a codeblock.*/
930 static void dequantization_int(int x, int y, Jpeg2000Cblk *cblk,
931 Jpeg2000Component *comp,
932 Jpeg2000T1Context *t1, Jpeg2000Band *band)
936 (int32_t *) &comp->data[(comp->coord[0][1] - comp->coord[0][0]) * y + x];
937 for (j = 0; j < (cblk->coord[1][1] - cblk->coord[1][0]); ++j)
938 for (i = 0; i < (cblk->coord[0][1] - cblk->coord[0][0]); ++i) {
939 idx = (comp->coord[0][1] - comp->coord[0][0]) * j + i;
941 ((int32_t)(t1->data[j][i]) * ((int32_t)band->stepsize) + (1 << 15)) >> 16;
946 /* Inverse ICT parameters in float and integer.
947 * int value = (float value) * (1<<16) */
948 static const float f_ict_params[4] = {
954 static const int i_ict_params[4] = {
961 static int mct_decode(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile)
965 int32_t *src[3], i0, i1, i2;
966 float *srcf[3], i0f, i1f, i2f;
968 for (i = 0; i < 3; i++)
969 if (tile->codsty[0].transform == FF_DWT97)
970 srcf[i] = tile->comp[i].data;
972 src[i] = (int32_t *)tile->comp[i].data;
974 for (i = 0; i < 2; i++)
975 csize *= tile->comp[0].coord[i][1] - tile->comp[0].coord[i][0];
976 switch (tile->codsty[0].transform) {
978 for (i = 0; i < csize; i++) {
979 i0f = *srcf[0] + (f_ict_params[0] * *srcf[2]);
980 i1f = *srcf[0] - (f_ict_params[1] * *srcf[1])
981 - (f_ict_params[2] * *srcf[2]);
982 i2f = *srcf[0] + (f_ict_params[3] * *srcf[1]);
989 for (i = 0; i < csize; i++) {
990 i0 = *src[0] + (((i_ict_params[0] * *src[2]) + (1 << 15)) >> 16);
991 i1 = *src[0] - (((i_ict_params[1] * *src[1]) + (1 << 15)) >> 16)
992 - (((i_ict_params[2] * *src[2]) + (1 << 15)) >> 16);
993 i2 = *src[0] + (((i_ict_params[3] * *src[1]) + (1 << 15)) >> 16);
1000 for (i = 0; i < csize; i++) {
1001 i1 = *src[0] - (*src[2] + *src[1] >> 2);
1013 static int jpeg2000_decode_tile(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile,
1016 int compno, reslevelno, bandno;
1020 Jpeg2000T1Context t1;
1021 /* Loop on tile components */
1023 for (compno = 0; compno < s->ncomponents; compno++) {
1024 Jpeg2000Component *comp = tile->comp + compno;
1025 Jpeg2000CodingStyle *codsty = tile->codsty + compno;
1026 /* Loop on resolution levels */
1027 for (reslevelno = 0; reslevelno < codsty->nreslevels2decode; reslevelno++) {
1028 Jpeg2000ResLevel *rlevel = comp->reslevel + reslevelno;
1030 for (bandno = 0; bandno < rlevel->nbands; bandno++) {
1031 uint16_t nb_precincts, precno;
1032 Jpeg2000Band *band = rlevel->band + bandno;
1033 int cblkno = 0, bandpos;
1034 bandpos = bandno + (reslevelno > 0);
1036 nb_precincts = rlevel->num_precincts_x * rlevel->num_precincts_y;
1037 /* Loop on precincts */
1038 for (precno = 0; precno < nb_precincts; precno++) {
1039 Jpeg2000Prec *prec = band->prec + precno;
1041 /* Loop on codeblocks */
1042 for (cblkno = 0; cblkno < prec->nb_codeblocks_width * prec->nb_codeblocks_height; cblkno++) {
1044 Jpeg2000Cblk *cblk = prec->cblk + cblkno;
1045 decode_cblk(s, codsty, &t1, cblk,
1046 cblk->coord[0][1] - cblk->coord[0][0],
1047 cblk->coord[1][1] - cblk->coord[1][0],
1050 /* Manage band offsets */
1051 x = cblk->coord[0][0];
1052 y = cblk->coord[1][0];
1053 if ((reslevelno > 0) && ((bandno + 1) & 1)) {
1054 Jpeg2000ResLevel *pres = comp->reslevel + (reslevelno - 1);
1055 x += pres->coord[0][1] - pres->coord[0][0];
1057 if ((reslevelno > 0) && ((bandno + 1) & 2)) {
1058 Jpeg2000ResLevel *pres = comp->reslevel + (reslevelno - 1);
1059 y += pres->coord[1][1] - pres->coord[1][0];
1062 if (s->avctx->flags & CODEC_FLAG_BITEXACT)
1063 dequantization_int(x, y, cblk, comp, &t1, band);
1065 dequantization_float(x, y, cblk, comp, &t1, band);
1069 } /* end reslevel */
1072 ff_dwt_decode(&comp->dwt, comp->data);
1075 /* inverse MCT transformation */
1076 if (tile->codsty[0].mct)
1077 mct_decode(s, tile);
1079 if (s->avctx->pix_fmt == AV_PIX_FMT_BGRA) // RGBA -> BGRA
1080 FFSWAP(float *, tile->comp[0].data, tile->comp[2].data);
1082 if (s->precision <= 8) {
1083 for (compno = 0; compno < s->ncomponents; compno++) {
1084 Jpeg2000Component *comp = tile->comp + compno;
1085 int32_t *datap = (int32_t *)comp->data;
1086 y = tile->comp[compno].coord[1][0] - s->image_offset_y;
1087 line = picture->data[0] + y * picture->linesize[0];
1088 for (; y < tile->comp[compno].coord[1][1] - s->image_offset_y; y += s->cdy[compno]) {
1091 x = tile->comp[compno].coord[0][0] - s->image_offset_x;
1092 dst = line + x * s->ncomponents + compno;
1094 for (; x < tile->comp[compno].coord[0][1] - s->image_offset_x; x += s->cdx[compno]) {
1095 *datap += 1 << (s->cbps[compno] - 1);
1098 else if (*datap >= (1 << s->cbps[compno]))
1099 *datap = (1 << s->cbps[compno]) - 1;
1101 dst += s->ncomponents;
1103 line += picture->linesize[0];
1107 for (compno = 0; compno < s->ncomponents; compno++) {
1108 Jpeg2000Component *comp = tile->comp + compno;
1109 float *datap = comp->data;
1110 int32_t *i_datap = (int32_t *) comp->data;
1113 y = tile->comp[compno].coord[1][0] - s->image_offset_y;
1114 linel = (uint16_t *)picture->data[0] + y * (picture->linesize[0] >> 1);
1115 for (; y < tile->comp[compno].coord[1][1] - s->image_offset_y; y += s->cdy[compno]) {
1117 x = tile->comp[compno].coord[0][0] - s->image_offset_x;
1118 dst = linel + (x * s->ncomponents + compno);
1119 for (; x < s->avctx->width; x += s->cdx[compno]) {
1121 /* DC level shift and clip see ISO 15444-1:2002 G.1.2 */
1122 if (s->avctx->flags & CODEC_FLAG_BITEXACT)
1123 val = *i_datap + (1 << (s->cbps[compno] - 1));
1125 val = lrintf(*datap) + (1 << (s->cbps[compno] - 1));
1126 val = av_clip(val, 0, (1 << s->cbps[compno]) - 1);
1127 /* align 12 bit values in little-endian mode */
1131 dst += s->ncomponents;
1133 linel += picture->linesize[0] >> 1;
1140 static void jpeg2000_dec_cleanup(Jpeg2000DecoderContext *s)
1143 for (tileno = 0; tileno < s->numXtiles * s->numYtiles; tileno++) {
1144 for (compno = 0; compno < s->ncomponents; compno++) {
1145 Jpeg2000Component *comp = s->tile[tileno].comp + compno;
1146 Jpeg2000CodingStyle *codsty = s->tile[tileno].codsty + compno;
1148 ff_jpeg2000_cleanup(comp, codsty);
1150 av_freep(&s->tile[tileno].comp);
1155 static int jpeg2000_read_main_headers(Jpeg2000DecoderContext *s)
1157 Jpeg2000CodingStyle *codsty = s->codsty;
1158 Jpeg2000QuantStyle *qntsty = s->qntsty;
1159 uint8_t *properties = s->properties;
1164 const uint8_t *oldbuf;
1166 if (s->buf_end - s->buf < 2) {
1167 av_log(s->avctx, AV_LOG_ERROR, "Missing EOC\n");
1171 marker = bytestream_get_be16(&s->buf);
1174 if (marker == JPEG2000_EOC)
1177 if (s->buf_end - s->buf < 2)
1178 return AVERROR_INVALIDDATA;
1179 len = bytestream_get_be16(&s->buf);
1185 ret = get_coc(s, codsty, properties);
1188 ret = get_cod(s, codsty, properties);
1191 ret = get_qcc(s, len, qntsty, properties);
1194 ret = get_qcd(s, len, qntsty, properties);
1197 ret = get_sot(s, len);
1200 // the comment is ignored
1204 // Tile-part lengths
1205 ret = get_tlm(s, len);
1208 av_log(s->avctx, AV_LOG_ERROR,
1209 "unsupported marker 0x%.4X at pos 0x%tX\n",
1210 marker, s->buf - s->buf_start - 4);
1214 if (((s->buf - oldbuf != len) && (marker != JPEG2000_SOT)) || ret) {
1215 av_log(s->avctx, AV_LOG_ERROR,
1216 "error during processing marker segment %.4x\n", marker);
1217 return ret ? ret : -1;
1223 /* Read bit stream packets --> T2 operation. */
1224 static int jpeg2000_read_bitstream_packets(Jpeg2000DecoderContext *s)
1227 Jpeg2000Tile *tile = s->tile + s->curtileno;
1229 if (ret = init_tile(s, s->curtileno))
1231 if (ret = jpeg2000_decode_packets(s, tile))
1237 static int jp2_find_codestream(Jpeg2000DecoderContext *s)
1240 int found_codestream = 0, search_range = 10;
1242 // Skip JPEG 2000 signature atom.
1245 while (!found_codestream && search_range) {
1246 atom_size = AV_RB32(s->buf);
1247 if (AV_RB32(s->buf + 4) == JP2_CODESTREAM) {
1248 found_codestream = 1;
1251 s->buf += atom_size;
1256 if (found_codestream)
1261 static int jpeg2000_decode_frame(AVCodecContext *avctx, void *data,
1262 int *got_frame, AVPacket *avpkt)
1264 Jpeg2000DecoderContext *s = avctx->priv_data;
1265 ThreadFrame frame = { .f = data };
1266 AVFrame *picture = data;
1270 s->buf = s->buf_start = avpkt->data;
1271 s->buf_end = s->buf_start + avpkt->size;
1272 s->curtileno = 0; // TODO: only one tile in DCI JP2K. to implement for more tiles
1274 // reduction factor, i.e number of resolution levels to skip
1275 s->reduction_factor = s->lowres;
1277 if (s->buf_end - s->buf < 2)
1278 return AVERROR_INVALIDDATA;
1280 // check if the image is in jp2 format
1281 if ((AV_RB32(s->buf) == 12) &&
1282 (AV_RB32(s->buf + 4) == JP2_SIG_TYPE) &&
1283 (AV_RB32(s->buf + 8) == JP2_SIG_VALUE)) {
1284 if (!jp2_find_codestream(s)) {
1285 av_log(avctx, AV_LOG_ERROR,
1286 "Could not find Jpeg2000 codestream atom.\n");
1287 return AVERROR_INVALIDDATA;
1291 if (bytestream_get_be16(&s->buf) != JPEG2000_SOC) {
1292 av_log(avctx, AV_LOG_ERROR, "SOC marker not present\n");
1293 return AVERROR_INVALIDDATA;
1295 if (ret = jpeg2000_read_main_headers(s))
1298 /* get picture buffer */
1299 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0) {
1300 av_log(avctx, AV_LOG_ERROR, "ff_thread_get_buffer() failed.\n");
1303 picture->pict_type = AV_PICTURE_TYPE_I;
1304 picture->key_frame = 1;
1306 if (ret = jpeg2000_read_bitstream_packets(s))
1308 for (tileno = 0; tileno < s->numXtiles * s->numYtiles; tileno++)
1309 if (ret = jpeg2000_decode_tile(s, s->tile + tileno, picture))
1315 jpeg2000_dec_cleanup(s);
1316 return ret ? ret : s->buf - s->buf_start;
1319 static void jpeg2000_init_static_data(AVCodec *codec)
1321 ff_jpeg2000_init_tier1_luts();
1324 #define OFFSET(x) offsetof(Jpeg2000DecoderContext, x)
1325 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1327 static const AVOption options[] = {
1328 { "lowres", "Lower the decoding resolution by a power of two",
1329 OFFSET(lowres), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, JPEG2000_MAX_RESLEVELS - 1, VD },
1333 static const AVProfile profiles[] = {
1334 { FF_PROFILE_JPEG2000_CSTREAM_RESTRICTION_0, "JPEG 2000 codestream restriction 0" },
1335 { FF_PROFILE_JPEG2000_CSTREAM_RESTRICTION_1, "JPEG 2000 codestream restriction 1" },
1336 { FF_PROFILE_JPEG2000_CSTREAM_NO_RESTRICTION, "JPEG 2000 no codestream restrictions" },
1337 { FF_PROFILE_JPEG2000_DCINEMA_2K, "JPEG 2000 digital cinema 2K" },
1338 { FF_PROFILE_JPEG2000_DCINEMA_4K, "JPEG 2000 digital cinema 4K" },
1339 { FF_PROFILE_UNKNOWN },
1342 static const AVClass class = {
1343 .class_name = "jpeg2000",
1344 .item_name = av_default_item_name,
1346 .version = LIBAVUTIL_VERSION_INT,
1349 AVCodec ff_jpeg2000_decoder = {
1351 .long_name = NULL_IF_CONFIG_SMALL("JPEG 2000"),
1352 .type = AVMEDIA_TYPE_VIDEO,
1353 .id = AV_CODEC_ID_JPEG2000,
1354 .capabilities = CODEC_CAP_FRAME_THREADS,
1355 .priv_data_size = sizeof(Jpeg2000DecoderContext),
1356 .init_static_data = jpeg2000_init_static_data,
1357 .decode = jpeg2000_decode_frame,
1358 .priv_class = &class,
1359 .pix_fmts = (enum AVPixelFormat[]) { AV_PIX_FMT_XYZ12,
1362 .profiles = NULL_IF_CONFIG_SMALL(profiles)