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 FFmpeg.
8 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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(EINVAL);
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(EINVAL);
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(EINVAL);
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 c->nreslevels = bytestream_get_byte(&s->buf) + 1; // num of resolution levels - 1
256 /* compute number of resolution levels to decode */
257 if (c->nreslevels < s->reduction_factor)
258 c->nreslevels2decode = 1;
260 c->nreslevels2decode = c->nreslevels - s->reduction_factor;
262 c->log2_cblk_width = bytestream_get_byte(&s->buf) + 2; // cblk width
263 c->log2_cblk_height = bytestream_get_byte(&s->buf) + 2; // cblk height
265 c->cblk_style = bytestream_get_byte(&s->buf);
266 if (c->cblk_style != 0) { // cblk style
267 av_log(s->avctx, AV_LOG_ERROR, "no extra cblk styles supported\n");
270 c->transform = bytestream_get_byte(&s->buf); // DWT transformation type
271 /* set integer 9/7 DWT in case of BITEXACT flag */
272 if ((s->avctx->flags & CODEC_FLAG_BITEXACT) && (c->transform == FF_DWT97))
273 c->transform = FF_DWT97_INT;
275 if (c->csty & JPEG2000_CSTY_PREC) {
277 for (i = 0; i < c->nreslevels; i++) {
278 byte = bytestream_get_byte(&s->buf);
279 c->log2_prec_widths[i] = byte & 0x0F; // precinct PPx
280 c->log2_prec_heights[i] = (byte >> 4) & 0x0F; // precinct PPy
286 /* get coding parameters for a particular tile or whole image*/
287 static int get_cod(Jpeg2000DecoderContext *s, Jpeg2000CodingStyle *c,
290 Jpeg2000CodingStyle tmp;
293 if (s->buf_end - s->buf < 5)
294 return AVERROR(EINVAL);
296 tmp.log2_prec_width =
297 tmp.log2_prec_height = 15;
299 tmp.csty = bytestream_get_byte(&s->buf);
301 // get progression order
302 tmp.prog_order = bytestream_get_byte(&s->buf);
304 tmp.nlayers = bytestream_get_be16(&s->buf);
305 tmp.mct = bytestream_get_byte(&s->buf); // multiple component transformation
308 for (compno = 0; compno < s->ncomponents; compno++)
309 if (!(properties[compno] & HAD_COC))
310 memcpy(c + compno, &tmp, sizeof(tmp));
314 /* Get coding parameters for a component in the whole image or a
315 * particular tile. */
316 static int get_coc(Jpeg2000DecoderContext *s, Jpeg2000CodingStyle *c,
321 if (s->buf_end - s->buf < 2)
322 return AVERROR(EINVAL);
324 compno = bytestream_get_byte(&s->buf);
327 c->csty = bytestream_get_byte(&s->buf);
330 properties[compno] |= HAD_COC;
334 /* Get common part for QCD and QCC segments. */
335 static int get_qcx(Jpeg2000DecoderContext *s, int n, Jpeg2000QuantStyle *q)
339 if (s->buf_end - s->buf < 1)
340 return AVERROR(EINVAL);
342 x = bytestream_get_byte(&s->buf); // Sqcd
344 q->nguardbits = x >> 5;
345 q->quantsty = x & 0x1f;
347 if (q->quantsty == JPEG2000_QSTY_NONE) {
349 if (s->buf_end - s->buf < n)
350 return AVERROR(EINVAL);
351 for (i = 0; i < n; i++)
352 q->expn[i] = bytestream_get_byte(&s->buf) >> 3;
353 } else if (q->quantsty == JPEG2000_QSTY_SI) {
354 if (s->buf_end - s->buf < 2)
355 return AVERROR(EINVAL);
356 x = bytestream_get_be16(&s->buf);
357 q->expn[0] = x >> 11;
358 q->mant[0] = x & 0x7ff;
359 for (i = 1; i < 32 * 3; i++) {
360 int curexpn = FFMAX(0, q->expn[0] - (i - 1) / 3);
361 q->expn[i] = curexpn;
362 q->mant[i] = q->mant[0];
366 if (s->buf_end - s->buf < n)
367 return AVERROR(EINVAL);
368 for (i = 0; i < n; i++) {
369 x = bytestream_get_be16(&s->buf);
370 q->expn[i] = x >> 11;
371 q->mant[i] = x & 0x7ff;
377 /* Get quantization parameters for a particular tile or a whole image. */
378 static int get_qcd(Jpeg2000DecoderContext *s, int n, Jpeg2000QuantStyle *q,
381 Jpeg2000QuantStyle tmp;
384 if (get_qcx(s, n, &tmp))
386 for (compno = 0; compno < s->ncomponents; compno++)
387 if (!(properties[compno] & HAD_QCC))
388 memcpy(q + compno, &tmp, sizeof(tmp));
392 /* Get quantization parameters for a component in the whole image
393 * on in a particular tile. */
394 static int get_qcc(Jpeg2000DecoderContext *s, int n, Jpeg2000QuantStyle *q,
399 if (s->buf_end - s->buf < 1)
400 return AVERROR(EINVAL);
402 compno = bytestream_get_byte(&s->buf);
403 properties[compno] |= HAD_QCC;
404 return get_qcx(s, n - 1, q + compno);
407 /* Get start of tile segment. */
408 static uint8_t get_sot(Jpeg2000DecoderContext *s, int n)
410 Jpeg2000TilePart *tp;
415 if (s->buf_end - s->buf < 4)
416 return AVERROR(EINVAL);
418 Isot = bytestream_get_be16(&s->buf); // Isot
420 av_log(s->avctx, AV_LOG_ERROR,
421 "Not a DCINEMA JP2K file: more than one tile\n");
424 Psot = bytestream_get_be32(&s->buf); // Psot
425 TPsot = bytestream_get_byte(&s->buf); // TPsot
427 /* Read TNSot but not used */
428 bytestream_get_byte(&s->buf); // TNsot
430 tp = s->tile[s->curtileno].tile_part + TPsot;
431 tp->tile_index = Isot;
435 /* Start of bit stream. Pointer to SOD marker
436 * Check SOD marker is present. */
437 if (JPEG2000_SOD == bytestream_get_be16(&s->buf))
438 tp->tp_start_bstrm = s->buf;
440 av_log(s->avctx, AV_LOG_ERROR, "SOD marker not found \n");
444 /* End address of bit stream =
445 * start address + (Psot - size of SOT HEADER(n)
446 * - size of SOT MARKER(2) - size of SOD marker(2) */
447 tp->tp_end_bstrm = s->buf + (tp->tp_len - n - 4);
449 // set buffer pointer to end of tile part header
450 s->buf = tp->tp_end_bstrm;
455 /* Tile-part lengths: see ISO 15444-1:2002, section A.7.1
456 * Used to know the number of tile parts and lengths.
457 * There may be multiple TLMs in the header.
458 * TODO: The function is not used for tile-parts management, nor anywhere else.
459 * It can be useful to allocate memory for tile parts, before managing the SOT
460 * markers. Parsing the TLM header is needed to increment the input header
462 * This marker is mandatory for DCI. */
463 static uint8_t get_tlm(Jpeg2000DecoderContext *s, int n)
465 uint8_t Stlm, ST, SP, tile_tlm, i;
466 bytestream_get_byte(&s->buf); /* Ztlm: skipped */
467 Stlm = bytestream_get_byte(&s->buf);
469 // too complex ? ST = ((Stlm >> 4) & 0x01) + ((Stlm >> 4) & 0x02);
470 ST = (Stlm >> 4) & 0x03;
471 // TODO: Manage case of ST = 0b11 --> raise error
472 SP = (Stlm >> 6) & 0x01;
473 tile_tlm = (n - 4) / ((SP + 1) * 2 + ST);
474 for (i = 0; i < tile_tlm; i++) {
479 bytestream_get_byte(&s->buf);
482 bytestream_get_be16(&s->buf);
485 bytestream_get_be32(&s->buf);
489 bytestream_get_be16(&s->buf);
491 bytestream_get_be32(&s->buf);
497 static int init_tile(Jpeg2000DecoderContext *s, int tileno)
500 int tilex = tileno % s->numXtiles;
501 int tiley = tileno / s->numXtiles;
502 Jpeg2000Tile *tile = s->tile + tileno;
503 Jpeg2000CodingStyle *codsty;
504 Jpeg2000QuantStyle *qntsty;
507 return AVERROR(ENOMEM);
509 /* copy codsty, qnsty to tile. TODO: Is it the best way?
510 * codsty, qnsty is an array of 4 structs Jpeg2000CodingStyle
511 * and Jpeg2000QuantStyle */
512 memcpy(tile->codsty, s->codsty, s->ncomponents * sizeof(*codsty));
513 memcpy(tile->qntsty, s->qntsty, s->ncomponents * sizeof(*qntsty));
515 for (compno = 0; compno < s->ncomponents; compno++) {
516 Jpeg2000Component *comp = tile->comp + compno;
517 int ret; // global bandno
518 codsty = tile->codsty + compno;
519 qntsty = tile->qntsty + compno;
521 comp->coord_o[0][0] = FFMAX(tilex * s->tile_width + s->tile_offset_x, s->image_offset_x);
522 comp->coord_o[0][1] = FFMIN((tilex + 1) * s->tile_width + s->tile_offset_x, s->width);
523 comp->coord_o[1][0] = FFMAX(tiley * s->tile_height + s->tile_offset_y, s->image_offset_y);
524 comp->coord_o[1][1] = FFMIN((tiley + 1) * s->tile_height + s->tile_offset_y, s->height);
526 // FIXME: add a dcinema profile check ?
527 // value is guaranteed by profile (orig=0, 1 tile)
528 comp->coord[0][0] = 0;
529 comp->coord[0][1] = s->avctx->width;
530 comp->coord[1][0] = 0;
531 comp->coord[1][1] = s->avctx->height;
533 if (ret = ff_jpeg2000_init_component(comp, codsty, qntsty,
534 s->cbps[compno], s->cdx[compno],
535 s->cdy[compno], s->avctx))
541 /* Read the number of coding passes. */
542 static int getnpasses(Jpeg2000DecoderContext *s)
549 if ((num = get_bits(s, 2)) != 3)
550 return num < 0 ? num : 3 + num;
551 if ((num = get_bits(s, 5)) != 31)
552 return num < 0 ? num : 6 + num;
553 num = get_bits(s, 7);
554 return num < 0 ? num : 37 + num;
557 static int getlblockinc(Jpeg2000DecoderContext *s)
560 while (ret = get_bits(s, 1)) {
568 static int jpeg2000_decode_packet(Jpeg2000DecoderContext *s,
569 Jpeg2000CodingStyle *codsty,
570 Jpeg2000ResLevel *rlevel, int precno,
571 int layno, uint8_t *expn, int numgbits)
573 int bandno, cblkno, ret, nb_code_blocks;
575 if (!(ret = get_bits(s, 1))) {
581 for (bandno = 0; bandno < rlevel->nbands; bandno++) {
582 Jpeg2000Band *band = rlevel->band + bandno;
583 Jpeg2000Prec *prec = band->prec + precno;
585 if (band->coord[0][0] == band->coord[0][1] ||
586 band->coord[1][0] == band->coord[1][1])
590 nb_code_blocks = prec->nb_codeblocks_height *
591 prec->nb_codeblocks_width;
592 for (cblkno = 0; cblkno < nb_code_blocks; cblkno++) {
593 Jpeg2000Cblk *cblk = prec->cblk + cblkno;
594 int incl, newpasses, llen;
597 incl = get_bits(s, 1);
599 incl = tag_tree_decode(s, prec->cblkincl + cblkno, layno + 1) == layno;
606 cblk->nonzerobits = expn[bandno] + numgbits - 1 -
607 tag_tree_decode(s, prec->zerobits + cblkno,
609 if ((newpasses = getnpasses(s)) < 0)
611 if ((llen = getlblockinc(s)) < 0)
613 cblk->lblock += llen;
614 if ((ret = get_bits(s, av_log2(newpasses) + cblk->lblock)) < 0)
616 cblk->lengthinc = ret;
617 cblk->npasses += newpasses;
622 if (codsty->csty & JPEG2000_CSTY_EPH) {
623 if (AV_RB16(s->buf) == JPEG2000_EPH)
626 av_log(s->avctx, AV_LOG_ERROR, "EPH marker not found.\n");
629 for (bandno = 0; bandno < rlevel->nbands; bandno++) {
630 Jpeg2000Band *band = rlevel->band + bandno;
631 Jpeg2000Prec *prec = band->prec + precno;
633 nb_code_blocks = prec->nb_codeblocks_height * prec->nb_codeblocks_width;
634 for (cblkno = 0; cblkno < nb_code_blocks; cblkno++) {
635 Jpeg2000Cblk *cblk = prec->cblk + cblkno;
636 if (s->buf_end - s->buf < cblk->lengthinc)
637 return AVERROR(EINVAL);
638 bytestream_get_buffer(&s->buf, cblk->data, cblk->lengthinc);
639 cblk->length += cblk->lengthinc;
646 static int jpeg2000_decode_packets(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile)
648 int layno, reslevelno, compno, precno, ok_reslevel;
649 uint8_t prog_order = tile->codsty[0].prog_order;
654 switch (prog_order) {
655 case JPEG2000_PGOD_LRCP:
656 for (layno = 0; layno < tile->codsty[0].nlayers; layno++) {
658 for (reslevelno = 0; ok_reslevel; reslevelno++) {
660 for (compno = 0; compno < s->ncomponents; compno++) {
661 Jpeg2000CodingStyle *codsty = tile->codsty + compno;
662 Jpeg2000QuantStyle *qntsty = tile->qntsty + compno;
663 if (reslevelno < codsty->nreslevels) {
664 Jpeg2000ResLevel *rlevel = tile->comp[compno].reslevel +
667 for (precno = 0; precno < rlevel->num_precincts_x * rlevel->num_precincts_y; precno++)
668 if (jpeg2000_decode_packet(s,
671 qntsty->expn + (reslevelno ? 3 * (reslevelno - 1) + 1 : 0),
680 case JPEG2000_PGOD_CPRL:
681 for (compno = 0; compno < s->ncomponents; compno++) {
682 Jpeg2000CodingStyle *codsty = tile->codsty + compno;
683 Jpeg2000QuantStyle *qntsty = tile->qntsty + compno;
685 /* Set bit stream buffer address according to tile-part.
686 * For DCinema one tile-part per component, so can be
687 * indexed by component. */
688 s->buf = tile->tile_part[compno].tp_start_bstrm;
690 /* Position loop (y axis)
691 * TODO: Automate computing of step 256.
692 * Fixed here, but to be computed before entering here. */
693 for (y = 0; y < s->height; y += 256) {
694 /* Position loop (y axis)
695 * TODO: automate computing of step 256.
696 * Fixed here, but to be computed before entering here. */
697 for (x = 0; x < s->width; x += 256) {
698 for (reslevelno = 0; reslevelno < codsty->nreslevels; reslevelno++) {
700 uint8_t reducedresno = codsty->nreslevels - 1 -reslevelno; // ==> N_L - r
701 Jpeg2000ResLevel *rlevel = tile->comp[compno].reslevel + reslevelno;
703 if (!((y % (1 << (rlevel->log2_prec_height + reducedresno)) == 0) ||
704 (y == 0))) // TODO: 2nd condition simplified as try0 always =0 for dcinema
707 if (!((x % (1 << (rlevel->log2_prec_width + reducedresno)) == 0) ||
708 (x == 0))) // TODO: 2nd condition simplified as try0 always =0 for dcinema
711 // check if a precinct exists
712 prcx = ff_jpeg2000_ceildivpow2(x, reducedresno) >> rlevel->log2_prec_width;
713 prcy = ff_jpeg2000_ceildivpow2(y, reducedresno) >> rlevel->log2_prec_height;
714 precno = prcx + rlevel->num_precincts_x * prcy;
715 for (layno = 0; layno < tile->codsty[0].nlayers; layno++) {
716 if (jpeg2000_decode_packet(s, codsty, rlevel,
718 qntsty->expn + (reslevelno ? 3 * (reslevelno - 1) + 1 : 0),
732 /* EOC marker reached */
738 /* TIER-1 routines */
739 static void decode_sigpass(Jpeg2000T1Context *t1, int width, int height,
740 int bpno, int bandno)
742 int mask = 3 << (bpno - 1), y0, x, y;
744 for (y0 = 0; y0 < height; y0 += 4)
745 for (x = 0; x < width; x++)
746 for (y = y0; y < height && y < y0 + 4; y++)
747 if ((t1->flags[y + 1][x + 1] & JPEG2000_T1_SIG_NB)
748 && !(t1->flags[y + 1][x + 1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS))) {
749 if (ff_mqc_decode(&t1->mqc,
751 ff_jpeg2000_getsigctxno(t1->flags[y + 1][x + 1],
753 int xorbit, ctxno = ff_jpeg2000_getsgnctxno(t1->flags[y + 1][x + 1],
757 (ff_mqc_decode(&t1->mqc,
758 t1->mqc.cx_states + ctxno) ^ xorbit)
761 ff_jpeg2000_set_significance(t1, x, y,
764 t1->flags[y + 1][x + 1] |= JPEG2000_T1_VIS;
768 static void decode_refpass(Jpeg2000T1Context *t1, int width, int height,
774 phalf = 1 << (bpno - 1);
777 for (y0 = 0; y0 < height; y0 += 4)
778 for (x = 0; x < width; x++)
779 for (y = y0; y < height && y < y0 + 4; y++)
780 if ((t1->flags[y + 1][x + 1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS)) == JPEG2000_T1_SIG) {
781 int ctxno = ff_jpeg2000_getrefctxno(t1->flags[y + 1][x + 1]);
782 int r = ff_mqc_decode(&t1->mqc,
783 t1->mqc.cx_states + ctxno)
785 t1->data[y][x] += t1->data[y][x] < 0 ? -r : r;
786 t1->flags[y + 1][x + 1] |= JPEG2000_T1_REF;
790 static void decode_clnpass(Jpeg2000DecoderContext *s, Jpeg2000T1Context *t1,
791 int width, int height, int bpno, int bandno,
794 int mask = 3 << (bpno - 1), y0, x, y, runlen, dec;
796 for (y0 = 0; y0 < height; y0 += 4)
797 for (x = 0; x < width; x++) {
798 if (y0 + 3 < height &&
799 !((t1->flags[y0 + 1][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
800 (t1->flags[y0 + 2][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
801 (t1->flags[y0 + 3][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)) ||
802 (t1->flags[y0 + 4][x + 1] & (JPEG2000_T1_SIG_NB | JPEG2000_T1_VIS | JPEG2000_T1_SIG)))) {
803 if (!ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_RL))
805 runlen = ff_mqc_decode(&t1->mqc,
806 t1->mqc.cx_states + MQC_CX_UNI);
807 runlen = (runlen << 1) | ff_mqc_decode(&t1->mqc,
816 for (y = y0 + runlen; y < y0 + 4 && y < height; y++) {
818 if (!(t1->flags[y + 1][x + 1] & (JPEG2000_T1_SIG | JPEG2000_T1_VIS)))
819 dec = ff_mqc_decode(&t1->mqc,
821 ff_jpeg2000_getsigctxno(t1->flags[y + 1][x + 1],
826 int ctxno = ff_jpeg2000_getsgnctxno(t1->flags[y + 1][x + 1],
828 t1->data[y][x] = (ff_mqc_decode(&t1->mqc,
829 t1->mqc.cx_states + ctxno) ^
832 ff_jpeg2000_set_significance(t1, x, y, t1->data[y][x] < 0);
835 t1->flags[y + 1][x + 1] &= ~JPEG2000_T1_VIS;
840 val = ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_UNI);
841 val = (val << 1) + ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_UNI);
842 val = (val << 1) + ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_UNI);
843 val = (val << 1) + ff_mqc_decode(&t1->mqc, t1->mqc.cx_states + MQC_CX_UNI);
845 av_log(s->avctx, AV_LOG_ERROR,
846 "Segmentation symbol value incorrect\n");
850 static int decode_cblk(Jpeg2000DecoderContext *s, Jpeg2000CodingStyle *codsty,
851 Jpeg2000T1Context *t1, Jpeg2000Cblk *cblk,
852 int width, int height, int bandpos)
854 int passno = cblk->npasses, pass_t = 2, bpno = cblk->nonzerobits - 1, y;
856 for (y = 0; y < height + 2; y++)
857 memset(t1->flags[y], 0, (width + 2) * sizeof(width));
859 for (y = 0; y < height; y++)
860 memset(t1->data[y], 0, width * sizeof(width));
862 ff_mqc_initdec(&t1->mqc, cblk->data);
863 cblk->data[cblk->length] = 0xff;
864 cblk->data[cblk->length + 1] = 0xff;
869 decode_sigpass(t1, width, height, bpno + 1, bandpos);
872 decode_refpass(t1, width, height, bpno + 1);
875 decode_clnpass(s, t1, width, height, bpno + 1, bandpos,
876 codsty->cblk_style & JPEG2000_CBLK_SEGSYM);
889 /* TODO: Verify dequantization for lossless case
890 * comp->data can be float or int
891 * band->stepsize can be float or int
892 * depending on the type of DWT transformation.
893 * see ISO/IEC 15444-1:2002 A.6.1 */
895 /* Float dequantization of a codeblock.*/
896 static void dequantization_float(int x, int y, Jpeg2000Cblk *cblk,
897 Jpeg2000Component *comp,
898 Jpeg2000T1Context *t1, Jpeg2000Band *band)
901 float *datap = &comp->data[(comp->coord[0][1] - comp->coord[0][0]) * y + x];
902 for (j = 0; j < (cblk->coord[1][1] - cblk->coord[1][0]); ++j)
903 for (i = 0; i < (cblk->coord[0][1] - cblk->coord[0][0]); ++i) {
904 idx = (comp->coord[0][1] - comp->coord[0][0]) * j + i;
905 datap[idx] = (float)(t1->data[j][i]) * ((float)band->stepsize);
910 /* Integer dequantization of a codeblock.*/
911 static void dequantization_int(int x, int y, Jpeg2000Cblk *cblk,
912 Jpeg2000Component *comp,
913 Jpeg2000T1Context *t1, Jpeg2000Band *band)
917 (int32_t *) &comp->data[(comp->coord[0][1] - comp->coord[0][0]) * y + x];
918 for (j = 0; j < (cblk->coord[1][1] - cblk->coord[1][0]); ++j)
919 for (i = 0; i < (cblk->coord[0][1] - cblk->coord[0][0]); ++i) {
920 idx = (comp->coord[0][1] - comp->coord[0][0]) * j + i;
922 ((int32_t)(t1->data[j][i]) * ((int32_t)band->stepsize) + (1 << 15)) >> 16;
927 /* Inverse ICT parameters in float and integer.
928 * int value = (float value) * (1<<16) */
929 static const float f_ict_params[4] = {
935 static const int i_ict_params[4] = {
942 static int mct_decode(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile)
946 int32_t *src[3], i0, i1, i2;
947 float *srcf[3], i0f, i1f, i2f;
949 for (i = 0; i < 3; i++)
950 if (tile->codsty[0].transform == FF_DWT97)
951 srcf[i] = tile->comp[i].data;
953 src[i] = (int32_t *)tile->comp[i].data;
955 for (i = 0; i < 2; i++)
956 csize *= tile->comp[0].coord[i][1] - tile->comp[0].coord[i][0];
957 switch (tile->codsty[0].transform) {
959 for (i = 0; i < csize; i++) {
960 i0f = *srcf[0] + (f_ict_params[0] * *srcf[2]);
961 i1f = *srcf[0] - (f_ict_params[1] * *srcf[1])
962 - (f_ict_params[2] * *srcf[2]);
963 i2f = *srcf[0] + (f_ict_params[3] * *srcf[1]);
970 for (i = 0; i < csize; i++) {
971 i0 = *src[0] + (((i_ict_params[0] * *src[2]) + (1 << 15)) >> 16);
972 i1 = *src[0] - (((i_ict_params[1] * *src[1]) + (1 << 15)) >> 16)
973 - (((i_ict_params[2] * *src[2]) + (1 << 15)) >> 16);
974 i2 = *src[0] + (((i_ict_params[3] * *src[1]) + (1 << 15)) >> 16);
981 for (i = 0; i < csize; i++) {
982 i1 = *src[0] - (*src[2] + *src[1] >> 2);
994 static int jpeg2000_decode_tile(Jpeg2000DecoderContext *s, Jpeg2000Tile *tile,
997 int compno, reslevelno, bandno;
1001 Jpeg2000T1Context t1;
1002 /* Loop on tile components */
1004 for (compno = 0; compno < s->ncomponents; compno++) {
1005 Jpeg2000Component *comp = tile->comp + compno;
1006 Jpeg2000CodingStyle *codsty = tile->codsty + compno;
1007 /* Loop on resolution levels */
1008 for (reslevelno = 0; reslevelno < codsty->nreslevels2decode; reslevelno++) {
1009 Jpeg2000ResLevel *rlevel = comp->reslevel + reslevelno;
1011 for (bandno = 0; bandno < rlevel->nbands; bandno++) {
1012 uint16_t nb_precincts, precno;
1013 Jpeg2000Band *band = rlevel->band + bandno;
1014 int cblkno = 0, bandpos;
1015 bandpos = bandno + (reslevelno > 0);
1017 nb_precincts = rlevel->num_precincts_x * rlevel->num_precincts_y;
1018 /* Loop on precincts */
1019 for (precno = 0; precno < nb_precincts; precno++) {
1020 Jpeg2000Prec *prec = band->prec + precno;
1022 /* Loop on codeblocks */
1023 for (cblkno = 0; cblkno < prec->nb_codeblocks_width * prec->nb_codeblocks_height; cblkno++) {
1025 Jpeg2000Cblk *cblk = prec->cblk + cblkno;
1026 decode_cblk(s, codsty, &t1, cblk,
1027 cblk->coord[0][1] - cblk->coord[0][0],
1028 cblk->coord[1][1] - cblk->coord[1][0],
1031 /* Manage band offsets */
1032 x = cblk->coord[0][0];
1033 y = cblk->coord[1][0];
1034 if ((reslevelno > 0) && ((bandno + 1) & 1)) {
1035 Jpeg2000ResLevel *pres = comp->reslevel + (reslevelno - 1);
1036 x += pres->coord[0][1] - pres->coord[0][0];
1038 if ((reslevelno > 0) && ((bandno + 1) & 2)) {
1039 Jpeg2000ResLevel *pres = comp->reslevel + (reslevelno - 1);
1040 y += pres->coord[1][1] - pres->coord[1][0];
1043 if (s->avctx->flags & CODEC_FLAG_BITEXACT)
1044 dequantization_int(x, y, cblk, comp, &t1, band);
1046 dequantization_float(x, y, cblk, comp, &t1, band);
1050 } /* end reslevel */
1053 ff_dwt_decode(&comp->dwt, comp->data);
1056 /* inverse MCT transformation */
1057 if (tile->codsty[0].mct)
1058 mct_decode(s, tile);
1060 if (s->avctx->pix_fmt == PIX_FMT_BGRA) // RGBA -> BGRA
1061 FFSWAP(float *, tile->comp[0].data, tile->comp[2].data);
1063 if (s->precision <= 8) {
1064 for (compno = 0; compno < s->ncomponents; compno++) {
1065 Jpeg2000Component *comp = tile->comp + compno;
1066 int32_t *datap = (int32_t *)comp->data;
1067 y = tile->comp[compno].coord[1][0] - s->image_offset_y;
1068 line = picture->data[0] + y * picture->linesize[0];
1069 for (; y < tile->comp[compno].coord[1][1] - s->image_offset_y; y += s->cdy[compno]) {
1072 x = tile->comp[compno].coord[0][0] - s->image_offset_x;
1073 dst = line + x * s->ncomponents + compno;
1075 for (; x < tile->comp[compno].coord[0][1] - s->image_offset_x; x += s->cdx[compno]) {
1076 *datap += 1 << (s->cbps[compno] - 1);
1079 else if (*datap >= (1 << s->cbps[compno]))
1080 *datap = (1 << s->cbps[compno]) - 1;
1082 dst += s->ncomponents;
1084 line += picture->linesize[0];
1088 for (compno = 0; compno < s->ncomponents; compno++) {
1089 Jpeg2000Component *comp = tile->comp + compno;
1090 float *datap = comp->data;
1091 int32_t *i_datap = (int32_t *) comp->data;
1094 y = tile->comp[compno].coord[1][0] - s->image_offset_y;
1095 linel = (uint16_t *)picture->data[0] + y * (picture->linesize[0] >> 1);
1096 for (; y < tile->comp[compno].coord[1][1] - s->image_offset_y; y += s->cdy[compno]) {
1098 x = tile->comp[compno].coord[0][0] - s->image_offset_x;
1099 dst = linel + (x * s->ncomponents + compno);
1100 for (; x < s->avctx->width; x += s->cdx[compno]) {
1102 /* DC level shift and clip see ISO 15444-1:2002 G.1.2 */
1103 if (s->avctx->flags & CODEC_FLAG_BITEXACT)
1104 val = *i_datap + (1 << (s->cbps[compno] - 1));
1106 val = lrintf(*datap) + (1 << (s->cbps[compno] - 1));
1107 val = av_clip(val, 0, (1 << s->cbps[compno]) - 1);
1108 /* align 12 bit values in little-endian mode */
1112 dst += s->ncomponents;
1114 linel += picture->linesize[0] >> 1;
1121 static void jpeg2000_dec_cleanup(Jpeg2000DecoderContext *s)
1124 for (tileno = 0; tileno < s->numXtiles * s->numYtiles; tileno++) {
1125 for (compno = 0; compno < s->ncomponents; compno++) {
1126 Jpeg2000Component *comp = s->tile[tileno].comp + compno;
1127 Jpeg2000CodingStyle *codsty = s->tile[tileno].codsty + compno;
1129 ff_jpeg2000_cleanup(comp, codsty);
1131 av_freep(&s->tile[tileno].comp);
1136 static int jpeg2000_read_main_headers(Jpeg2000DecoderContext *s)
1138 Jpeg2000CodingStyle *codsty = s->codsty;
1139 Jpeg2000QuantStyle *qntsty = s->qntsty;
1140 uint8_t *properties = s->properties;
1145 const uint8_t *oldbuf;
1147 if (s->buf_end - s->buf < 2) {
1148 av_log(s->avctx, AV_LOG_ERROR, "Missing EOC\n");
1152 marker = bytestream_get_be16(&s->buf);
1155 if (marker == JPEG2000_EOC)
1158 if (s->buf_end - s->buf < 2)
1159 return AVERROR(EINVAL);
1160 len = bytestream_get_be16(&s->buf);
1166 ret = get_coc(s, codsty, properties);
1169 ret = get_cod(s, codsty, properties);
1172 ret = get_qcc(s, len, qntsty, properties);
1175 ret = get_qcd(s, len, qntsty, properties);
1178 ret = get_sot(s, len);
1181 // the comment is ignored
1185 // Tile-part lengths
1186 ret = get_tlm(s, len);
1189 av_log(s->avctx, AV_LOG_ERROR,
1190 "unsupported marker 0x%.4X at pos 0x%lX\n",
1191 marker, (uint64_t)(s->buf - s->buf_start - 4));
1195 if (((s->buf - oldbuf != len) && (marker != JPEG2000_SOT)) || ret) {
1196 av_log(s->avctx, AV_LOG_ERROR,
1197 "error during processing marker segment %.4x\n", marker);
1198 return ret ? ret : -1;
1204 /* Read bit stream packets --> T2 operation. */
1205 static int jpeg2000_read_bitstream_packets(Jpeg2000DecoderContext *s)
1208 Jpeg2000Tile *tile = s->tile + s->curtileno;
1210 if (ret = init_tile(s, s->curtileno))
1212 if (ret = jpeg2000_decode_packets(s, tile))
1218 static int jp2_find_codestream(Jpeg2000DecoderContext *s)
1221 int found_codestream = 0, search_range = 10;
1223 // Skip JPEG 2000 signature atom.
1226 while (!found_codestream && search_range) {
1227 atom_size = AV_RB32(s->buf);
1228 if (AV_RB32(s->buf + 4) == JP2_CODESTREAM) {
1229 found_codestream = 1;
1232 s->buf += atom_size;
1237 if (found_codestream)
1242 static int jpeg2000_decode_frame(AVCodecContext *avctx, void *data,
1243 int *got_frame, AVPacket *avpkt)
1245 Jpeg2000DecoderContext *s = avctx->priv_data;
1246 ThreadFrame frame = { .f = data };
1247 AVFrame *picture = data;
1251 s->buf = s->buf_start = avpkt->data;
1252 s->buf_end = s->buf_start + avpkt->size;
1253 s->curtileno = 0; // TODO: only one tile in DCI JP2K. to implement for more tiles
1255 // reduction factor, i.e number of resolution levels to skip
1256 s->reduction_factor = s->lowres;
1258 ff_jpeg2000_init_tier1_luts();
1260 if (s->buf_end - s->buf < 2)
1261 return AVERROR(EINVAL);
1263 // check if the image is in jp2 format
1264 if ((AV_RB32(s->buf) == 12) &&
1265 (AV_RB32(s->buf + 4) == JP2_SIG_TYPE) &&
1266 (AV_RB32(s->buf + 8) == JP2_SIG_VALUE)) {
1267 if (!jp2_find_codestream(s)) {
1268 av_log(avctx, AV_LOG_ERROR,
1269 "couldn't find jpeg2k codestream atom\n");
1272 } else if (AV_RB16(s->buf) != JPEG2000_SOC && AV_RB32(s->buf + 4) == JP2_CODESTREAM) {
1276 if (bytestream_get_be16(&s->buf) != JPEG2000_SOC) {
1277 av_log(avctx, AV_LOG_ERROR, "SOC marker not present\n");
1280 if (ret = jpeg2000_read_main_headers(s))
1283 /* get picture buffer */
1284 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0) {
1285 av_log(avctx, AV_LOG_ERROR, "ff_thread_get_buffer() failed.\n");
1288 picture->pict_type = AV_PICTURE_TYPE_I;
1289 picture->key_frame = 1;
1291 if (ret = jpeg2000_read_bitstream_packets(s))
1293 for (tileno = 0; tileno < s->numXtiles * s->numYtiles; tileno++)
1294 if (ret = jpeg2000_decode_tile(s, s->tile + tileno, picture))
1296 jpeg2000_dec_cleanup(s);
1300 return s->buf - s->buf_start;
1303 #define OFFSET(x) offsetof(Jpeg2000DecoderContext, x)
1304 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1306 static const AVOption options[] = {
1307 { "lowres", "Lower the decoding resolution by a power of two",
1308 OFFSET(lowres), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, JPEG2000_MAX_RESLEVELS - 1, VD },
1312 static const AVProfile profiles[] = {
1313 { FF_PROFILE_JPEG2000_CSTREAM_RESTRICTION_0, "JPEG 2000 codestream restriction 0" },
1314 { FF_PROFILE_JPEG2000_CSTREAM_RESTRICTION_1, "JPEG 2000 codestream restriction 1" },
1315 { FF_PROFILE_JPEG2000_CSTREAM_NO_RESTRICTION, "JPEG 2000 no codestream restrictions" },
1316 { FF_PROFILE_JPEG2000_DCINEMA_2K, "JPEG 2000 digital cinema 2K" },
1317 { FF_PROFILE_JPEG2000_DCINEMA_4K, "JPEG 2000 digital cinema 4K" },
1318 { FF_PROFILE_UNKNOWN },
1321 static const AVClass class = {
1322 .class_name = "jpeg2000",
1323 .item_name = av_default_item_name,
1325 .version = LIBAVUTIL_VERSION_INT,
1328 AVCodec ff_jpeg2000_decoder = {
1330 .long_name = NULL_IF_CONFIG_SMALL("JPEG 2000"),
1331 .type = AVMEDIA_TYPE_VIDEO,
1332 .id = AV_CODEC_ID_JPEG2000,
1333 .capabilities = CODEC_CAP_FRAME_THREADS,
1334 .priv_data_size = sizeof(Jpeg2000DecoderContext),
1335 .decode = jpeg2000_decode_frame,
1336 .priv_class = &class,
1337 .pix_fmts = (enum PixelFormat[]) { AV_PIX_FMT_XYZ12,
1340 .profiles = NULL_IF_CONFIG_SMALL(profiles)