FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
pngenc.c
Go to the documentation of this file.
1 /*
2  * PNG image format
3  * Copyright (c) 2003 Fabrice Bellard
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "avcodec.h"
23 #include "internal.h"
24 #include "bytestream.h"
25 #include "huffyuvencdsp.h"
26 #include "png.h"
27 
28 #include "libavutil/avassert.h"
29 #include "libavutil/opt.h"
30 
31 #include <zlib.h>
32 
33 #define IOBUF_SIZE 4096
34 
35 typedef struct PNGEncContext {
36  AVClass *class;
38 
42 
44 
45  z_stream zstream;
47  int dpi; ///< Physical pixel density, in dots per inch, if set
48  int dpm; ///< Physical pixel density, in dots per meter, if set
50 
51 static void png_get_interlaced_row(uint8_t *dst, int row_size,
52  int bits_per_pixel, int pass,
53  const uint8_t *src, int width)
54 {
55  int x, mask, dst_x, j, b, bpp;
56  uint8_t *d;
57  const uint8_t *s;
58  static const int masks[] = {0x80, 0x08, 0x88, 0x22, 0xaa, 0x55, 0xff};
59 
60  mask = masks[pass];
61  switch (bits_per_pixel) {
62  case 1:
63  memset(dst, 0, row_size);
64  dst_x = 0;
65  for (x = 0; x < width; x++) {
66  j = (x & 7);
67  if ((mask << j) & 0x80) {
68  b = (src[x >> 3] >> (7 - j)) & 1;
69  dst[dst_x >> 3] |= b << (7 - (dst_x & 7));
70  dst_x++;
71  }
72  }
73  break;
74  default:
75  bpp = bits_per_pixel >> 3;
76  d = dst;
77  s = src;
78  for (x = 0; x < width; x++) {
79  j = x & 7;
80  if ((mask << j) & 0x80) {
81  memcpy(d, s, bpp);
82  d += bpp;
83  }
84  s += bpp;
85  }
86  break;
87  }
88 }
89 
91  int w, int bpp)
92 {
93  int i;
94  for (i = 0; i < w; i++) {
95  int a, b, c, p, pa, pb, pc;
96 
97  a = src[i - bpp];
98  b = top[i];
99  c = top[i - bpp];
100 
101  p = b - c;
102  pc = a - c;
103 
104  pa = abs(p);
105  pb = abs(pc);
106  pc = abs(p + pc);
107 
108  if (pa <= pb && pa <= pc)
109  p = a;
110  else if (pb <= pc)
111  p = b;
112  else
113  p = c;
114  dst[i] = src[i] - p;
115  }
116 }
117 
118 static void sub_left_prediction(PNGEncContext *c, uint8_t *dst, const uint8_t *src, int bpp, int size)
119 {
120  const uint8_t *src1 = src + bpp;
121  const uint8_t *src2 = src;
122  int x, unaligned_w;
123 
124  memcpy(dst, src, bpp);
125  dst += bpp;
126  size -= bpp;
127  unaligned_w = FFMIN(32 - bpp, size);
128  for (x = 0; x < unaligned_w; x++)
129  *dst++ = *src1++ - *src2++;
130  size -= unaligned_w;
131  c->hdsp.diff_bytes(dst, src1, src2, size);
132 }
133 
134 static void png_filter_row(PNGEncContext *c, uint8_t *dst, int filter_type,
135  uint8_t *src, uint8_t *top, int size, int bpp)
136 {
137  int i;
138 
139  switch (filter_type) {
141  memcpy(dst, src, size);
142  break;
144  sub_left_prediction(c, dst, src, bpp, size);
145  break;
146  case PNG_FILTER_VALUE_UP:
147  c->hdsp.diff_bytes(dst, src, top, size);
148  break;
150  for (i = 0; i < bpp; i++)
151  dst[i] = src[i] - (top[i] >> 1);
152  for (; i < size; i++)
153  dst[i] = src[i] - ((src[i - bpp] + top[i]) >> 1);
154  break;
156  for (i = 0; i < bpp; i++)
157  dst[i] = src[i] - top[i];
158  sub_png_paeth_prediction(dst + i, src + i, top + i, size - i, bpp);
159  break;
160  }
161 }
162 
164  uint8_t *src, uint8_t *top, int size, int bpp)
165 {
166  int pred = s->filter_type;
167  av_assert0(bpp || !pred);
168  if (!top && pred)
169  pred = PNG_FILTER_VALUE_SUB;
170  if (pred == PNG_FILTER_VALUE_MIXED) {
171  int i;
172  int cost, bcost = INT_MAX;
173  uint8_t *buf1 = dst, *buf2 = dst + size + 16;
174  for (pred = 0; pred < 5; pred++) {
175  png_filter_row(s, buf1 + 1, pred, src, top, size, bpp);
176  buf1[0] = pred;
177  cost = 0;
178  for (i = 0; i <= size; i++)
179  cost += abs((int8_t) buf1[i]);
180  if (cost < bcost) {
181  bcost = cost;
182  FFSWAP(uint8_t *, buf1, buf2);
183  }
184  }
185  return buf2;
186  } else {
187  png_filter_row(s, dst + 1, pred, src, top, size, bpp);
188  dst[0] = pred;
189  return dst;
190  }
191 }
192 
193 static void png_write_chunk(uint8_t **f, uint32_t tag,
194  const uint8_t *buf, int length)
195 {
196  uint32_t crc;
197  uint8_t tagbuf[4];
198 
199  bytestream_put_be32(f, length);
200  crc = crc32(0, Z_NULL, 0);
201  AV_WL32(tagbuf, tag);
202  crc = crc32(crc, tagbuf, 4);
203  bytestream_put_be32(f, av_bswap32(tag));
204  if (length > 0) {
205  crc = crc32(crc, buf, length);
206  memcpy(*f, buf, length);
207  *f += length;
208  }
209  bytestream_put_be32(f, crc);
210 }
211 
212 /* XXX: do filtering */
213 static int png_write_row(PNGEncContext *s, const uint8_t *data, int size)
214 {
215  int ret;
216 
217  s->zstream.avail_in = size;
218  s->zstream.next_in = data;
219  while (s->zstream.avail_in > 0) {
220  ret = deflate(&s->zstream, Z_NO_FLUSH);
221  if (ret != Z_OK)
222  return -1;
223  if (s->zstream.avail_out == 0) {
224  if (s->bytestream_end - s->bytestream > IOBUF_SIZE + 100)
226  MKTAG('I', 'D', 'A', 'T'), s->buf, IOBUF_SIZE);
227  s->zstream.avail_out = IOBUF_SIZE;
228  s->zstream.next_out = s->buf;
229  }
230  }
231  return 0;
232 }
233 
235  const AVFrame *pict, int *got_packet)
236 {
237  PNGEncContext *s = avctx->priv_data;
238  const AVFrame *const p = pict;
239  int bit_depth, color_type, y, len, row_size, ret, is_progressive;
240  int bits_per_pixel, pass_row_size, enc_row_size;
241  int64_t max_packet_size;
242  int compression_level;
243  uint8_t *ptr, *top, *crow_buf, *crow;
244  uint8_t *crow_base = NULL;
245  uint8_t *progressive_buf = NULL;
246  uint8_t *top_buf = NULL;
247 
248  is_progressive = !!(avctx->flags & CODEC_FLAG_INTERLACED_DCT);
249  switch (avctx->pix_fmt) {
250  case AV_PIX_FMT_RGBA64BE:
251  bit_depth = 16;
252  color_type = PNG_COLOR_TYPE_RGB_ALPHA;
253  break;
254  case AV_PIX_FMT_RGB48BE:
255  bit_depth = 16;
256  color_type = PNG_COLOR_TYPE_RGB;
257  break;
258  case AV_PIX_FMT_RGBA:
259  bit_depth = 8;
260  color_type = PNG_COLOR_TYPE_RGB_ALPHA;
261  break;
262  case AV_PIX_FMT_RGB24:
263  bit_depth = 8;
264  color_type = PNG_COLOR_TYPE_RGB;
265  break;
266  case AV_PIX_FMT_GRAY16BE:
267  bit_depth = 16;
268  color_type = PNG_COLOR_TYPE_GRAY;
269  break;
270  case AV_PIX_FMT_GRAY8:
271  bit_depth = 8;
272  color_type = PNG_COLOR_TYPE_GRAY;
273  break;
274  case AV_PIX_FMT_GRAY8A:
275  bit_depth = 8;
276  color_type = PNG_COLOR_TYPE_GRAY_ALPHA;
277  break;
279  bit_depth = 1;
280  color_type = PNG_COLOR_TYPE_GRAY;
281  break;
282  case AV_PIX_FMT_PAL8:
283  bit_depth = 8;
284  color_type = PNG_COLOR_TYPE_PALETTE;
285  break;
286  default:
287  return -1;
288  }
289  bits_per_pixel = ff_png_get_nb_channels(color_type) * bit_depth;
290  row_size = (avctx->width * bits_per_pixel + 7) >> 3;
291 
292  s->zstream.zalloc = ff_png_zalloc;
293  s->zstream.zfree = ff_png_zfree;
294  s->zstream.opaque = NULL;
295  compression_level = avctx->compression_level == FF_COMPRESSION_DEFAULT
296  ? Z_DEFAULT_COMPRESSION
297  : av_clip(avctx->compression_level, 0, 9);
298  ret = deflateInit2(&s->zstream, compression_level,
299  Z_DEFLATED, 15, 8, Z_DEFAULT_STRATEGY);
300  if (ret != Z_OK)
301  return -1;
302 
303  enc_row_size = deflateBound(&s->zstream, row_size);
304  max_packet_size = avctx->height * (int64_t)(enc_row_size +
305  ((enc_row_size + IOBUF_SIZE - 1) / IOBUF_SIZE) * 12)
307  if (max_packet_size > INT_MAX)
308  return AVERROR(ENOMEM);
309  if ((ret = ff_alloc_packet2(avctx, pkt, max_packet_size)) < 0)
310  return ret;
311 
312  s->bytestream_start =
313  s->bytestream = pkt->data;
314  s->bytestream_end = pkt->data + pkt->size;
315 
316  crow_base = av_malloc((row_size + 32) << (s->filter_type == PNG_FILTER_VALUE_MIXED));
317  if (!crow_base)
318  goto fail;
319  // pixel data should be aligned, but there's a control byte before it
320  crow_buf = crow_base + 15;
321  if (is_progressive) {
322  progressive_buf = av_malloc(row_size + 1);
323  if (!progressive_buf)
324  goto fail;
325  }
326  if (is_progressive) {
327  top_buf = av_malloc(row_size + 1);
328  if (!top_buf)
329  goto fail;
330  }
331 
332  /* write png header */
334  s->bytestream += 8;
335 
336  AV_WB32(s->buf, avctx->width);
337  AV_WB32(s->buf + 4, avctx->height);
338  s->buf[8] = bit_depth;
339  s->buf[9] = color_type;
340  s->buf[10] = 0; /* compression type */
341  s->buf[11] = 0; /* filter type */
342  s->buf[12] = is_progressive; /* interlace type */
343 
344  png_write_chunk(&s->bytestream, MKTAG('I', 'H', 'D', 'R'), s->buf, 13);
345 
346  if (s->dpm) {
347  AV_WB32(s->buf, s->dpm);
348  AV_WB32(s->buf + 4, s->dpm);
349  s->buf[8] = 1; /* unit specifier is meter */
350  } else {
351  AV_WB32(s->buf, avctx->sample_aspect_ratio.num);
352  AV_WB32(s->buf + 4, avctx->sample_aspect_ratio.den);
353  s->buf[8] = 0; /* unit specifier is unknown */
354  }
355  png_write_chunk(&s->bytestream, MKTAG('p', 'H', 'Y', 's'), s->buf, 9);
356 
357  /* put the palette if needed */
358  if (color_type == PNG_COLOR_TYPE_PALETTE) {
359  int has_alpha, alpha, i;
360  unsigned int v;
361  uint32_t *palette;
362  uint8_t *alpha_ptr;
363 
364  palette = (uint32_t *)p->data[1];
365  ptr = s->buf;
366  alpha_ptr = s->buf + 256 * 3;
367  has_alpha = 0;
368  for (i = 0; i < 256; i++) {
369  v = palette[i];
370  alpha = v >> 24;
371  if (alpha != 0xff)
372  has_alpha = 1;
373  *alpha_ptr++ = alpha;
374  bytestream_put_be24(&ptr, v);
375  }
377  MKTAG('P', 'L', 'T', 'E'), s->buf, 256 * 3);
378  if (has_alpha) {
380  MKTAG('t', 'R', 'N', 'S'), s->buf + 256 * 3, 256);
381  }
382  }
383 
384  /* now put each row */
385  s->zstream.avail_out = IOBUF_SIZE;
386  s->zstream.next_out = s->buf;
387  if (is_progressive) {
388  int pass;
389 
390  for (pass = 0; pass < NB_PASSES; pass++) {
391  /* NOTE: a pass is completely omitted if no pixels would be
392  * output */
393  pass_row_size = ff_png_pass_row_size(pass, bits_per_pixel, avctx->width);
394  if (pass_row_size > 0) {
395  top = NULL;
396  for (y = 0; y < avctx->height; y++)
397  if ((ff_png_pass_ymask[pass] << (y & 7)) & 0x80) {
398  ptr = p->data[0] + y * p->linesize[0];
399  FFSWAP(uint8_t *, progressive_buf, top_buf);
400  png_get_interlaced_row(progressive_buf, pass_row_size,
401  bits_per_pixel, pass,
402  ptr, avctx->width);
403  crow = png_choose_filter(s, crow_buf, progressive_buf,
404  top, pass_row_size, bits_per_pixel >> 3);
405  png_write_row(s, crow, pass_row_size + 1);
406  top = progressive_buf;
407  }
408  }
409  }
410  } else {
411  top = NULL;
412  for (y = 0; y < avctx->height; y++) {
413  ptr = p->data[0] + y * p->linesize[0];
414  crow = png_choose_filter(s, crow_buf, ptr, top,
415  row_size, bits_per_pixel >> 3);
416  png_write_row(s, crow, row_size + 1);
417  top = ptr;
418  }
419  }
420  /* compress last bytes */
421  for (;;) {
422  ret = deflate(&s->zstream, Z_FINISH);
423  if (ret == Z_OK || ret == Z_STREAM_END) {
424  len = IOBUF_SIZE - s->zstream.avail_out;
425  if (len > 0 && s->bytestream_end - s->bytestream > len + 100) {
426  png_write_chunk(&s->bytestream, MKTAG('I', 'D', 'A', 'T'), s->buf, len);
427  }
428  s->zstream.avail_out = IOBUF_SIZE;
429  s->zstream.next_out = s->buf;
430  if (ret == Z_STREAM_END)
431  break;
432  } else {
433  goto fail;
434  }
435  }
436  png_write_chunk(&s->bytestream, MKTAG('I', 'E', 'N', 'D'), NULL, 0);
437 
438  pkt->size = s->bytestream - s->bytestream_start;
439  pkt->flags |= AV_PKT_FLAG_KEY;
440  *got_packet = 1;
441  ret = 0;
442 
443 the_end:
444  av_free(crow_base);
445  av_free(progressive_buf);
446  av_free(top_buf);
447  deflateEnd(&s->zstream);
448  return ret;
449 fail:
450  ret = -1;
451  goto the_end;
452 }
453 
455 {
456  PNGEncContext *s = avctx->priv_data;
457 
458  switch (avctx->pix_fmt) {
459  case AV_PIX_FMT_RGBA:
460  avctx->bits_per_coded_sample = 32;
461  break;
462  case AV_PIX_FMT_RGB24:
463  avctx->bits_per_coded_sample = 24;
464  break;
465  case AV_PIX_FMT_GRAY8:
466  avctx->bits_per_coded_sample = 0x28;
467  break;
469  avctx->bits_per_coded_sample = 1;
470  break;
471  case AV_PIX_FMT_PAL8:
472  avctx->bits_per_coded_sample = 8;
473  }
474 
475  avctx->coded_frame = av_frame_alloc();
476  if (!avctx->coded_frame)
477  return AVERROR(ENOMEM);
478 
480  avctx->coded_frame->key_frame = 1;
481 
483 
484  s->filter_type = av_clip(avctx->prediction_method,
487  if (avctx->pix_fmt == AV_PIX_FMT_MONOBLACK)
489 
490  if (s->dpi && s->dpm) {
491  av_log(avctx, AV_LOG_ERROR, "Only one of 'dpi' or 'dpm' options should be set\n");
492  return AVERROR(EINVAL);
493  } else if (s->dpi) {
494  s->dpm = s->dpi * 10000 / 254;
495  }
496 
497  return 0;
498 }
499 
501 {
502  av_frame_free(&avctx->coded_frame);
503  return 0;
504 }
505 
506 #define OFFSET(x) offsetof(PNGEncContext, x)
507 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
508 static const AVOption options[] = {
509  {"dpi", "Set image resolution (in dots per inch)", OFFSET(dpi), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 0x10000, VE},
510  {"dpm", "Set image resolution (in dots per meter)", OFFSET(dpm), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 0x10000, VE},
511  { NULL }
512 };
513 
514 static const AVClass pngenc_class = {
515  .class_name = "PNG encoder",
516  .item_name = av_default_item_name,
517  .option = options,
518  .version = LIBAVUTIL_VERSION_INT,
519 };
520 
522  .name = "png",
523  .long_name = NULL_IF_CONFIG_SMALL("PNG (Portable Network Graphics) image"),
524  .type = AVMEDIA_TYPE_VIDEO,
525  .id = AV_CODEC_ID_PNG,
526  .priv_data_size = sizeof(PNGEncContext),
527  .init = png_enc_init,
528  .close = png_enc_close,
529  .encode2 = encode_frame,
531  .pix_fmts = (const enum AVPixelFormat[]) {
538  },
539  .priv_class = &pngenc_class,
540 };