1 /* $Id: scan.c,v 1.52 2005/11/25 15:05:25 titer Exp $
3 This file is part of the HandBrake source code.
4 Homepage: <http://handbrake.fr/>.
5 It may be used under the terms of the GNU General Public License. */
8 #include "a52dec/a52.h"
17 hb_list_t * list_title;
24 static void ScanFunc( void * );
25 static int DecodePreviews( hb_scan_t *, hb_title_t * title );
26 static void LookForAudio( hb_title_t * title, hb_buffer_t * b );
27 static int AllAudioOK( hb_title_t * title );
29 static const char *aspect_to_string( double aspect )
31 switch ( (int)(aspect * 9.) )
33 case 9 * 4 / 3: return "4:3";
34 case 9 * 16 / 9: return "16:9";
36 static char arstr[32];
37 sprintf( arstr, aspect >= 1.? "%.2f:1" : "1:%.2f", aspect );
41 hb_thread_t * hb_scan_init( hb_handle_t * handle, const char * path,
42 int title_index, hb_list_t * list_title )
44 hb_scan_t * data = calloc( sizeof( hb_scan_t ), 1 );
47 data->path = strdup( path );
48 data->title_index = title_index;
49 data->list_title = list_title;
51 return hb_thread_init( "scan", ScanFunc, data, HB_NORMAL_PRIORITY );
54 static void ScanFunc( void * _data )
56 hb_scan_t * data = (hb_scan_t *) _data;
63 /* Try to open the path as a DVD. If it fails, try as a file */
64 hb_log( "scan: trying to open with libdvdread" );
65 if( ( data->dvd = hb_dvd_init( data->path ) ) )
67 hb_log( "scan: DVD has %d title(s)",
68 hb_dvd_title_count( data->dvd ) );
69 if( data->title_index )
71 /* Scan this title only */
72 hb_list_add( data->list_title, hb_dvd_title_scan( data->dvd,
73 data->title_index ) );
78 for( i = 0; i < hb_dvd_title_count( data->dvd ); i++ )
80 hb_list_add( data->list_title,
81 hb_dvd_title_scan( data->dvd, i + 1 ) );
85 else if ( (data->stream = hb_stream_open( data->path, 0 ) ) != NULL )
87 hb_list_add( data->list_title, hb_stream_title_scan( data->stream ) );
91 hb_log( "scan: unrecognized file type" );
95 for( i = 0; i < hb_list_count( data->list_title ); )
100 hb_title_t * title_tmp = NULL;
102 title = hb_list_item( data->list_title, i );
104 /* I've seen a DVD with strictly identical titles. Check this
105 here and ignore it if redundant */
106 for( j = 0; j < i; j++ )
108 title_tmp = hb_list_item( data->list_title, j );
109 if( title->vts == title_tmp->vts &&
110 title->block_start == title_tmp->block_start &&
111 title->block_end == title_tmp->block_end &&
112 title->block_count == title_tmp->block_count )
123 hb_log( "scan: title %d is duplicate with title %d",
124 title->index, title_tmp->index );
125 hb_list_rem( data->list_title, title );
126 free( title ); /* This _will_ leak! */
130 #define p state.param.scanning
132 state.state = HB_STATE_SCANNING;
133 p.title_cur = title->index;
134 p.title_count = data->dvd ? hb_dvd_title_count( data->dvd ) : hb_list_count(data->list_title);
135 hb_set_state( data->h, &state );
138 /* Decode previews */
139 /* this will also detect more AC3 / DTS information */
140 if( !DecodePreviews( data, title ) )
142 /* TODO: free things */
143 hb_list_rem( data->list_title, title );
147 /* Make sure we found audio rates and bitrates */
148 for( j = 0; j < hb_list_count( title->list_audio ); )
150 audio = hb_list_item( title->list_audio, j );
151 if( !audio->config.in.bitrate )
153 hb_log( "scan: removing audio 0x%x because no bitrate found",
155 hb_list_rem( title->list_audio, audio );
162 /* If we don't have any audio streams left, remove the title */
163 if( !hb_list_count( title->list_audio ) )
165 hb_log("scan: ignoring title %i, no audio tracks found", title->index );
166 hb_list_rem( data->list_title, title );
174 /* Init jobs templates */
175 for( i = 0; i < hb_list_count( data->list_title ); i++ )
179 title = hb_list_item( data->list_title, i );
180 job = calloc( sizeof( hb_job_t ), 1 );
185 /* Set defaults settings */
186 job->chapter_start = 1;
187 job->chapter_end = hb_list_count( title->list_chapter );
189 /* Autocrop by default. Gnark gnark */
190 memcpy( job->crop, title->crop, 4 * sizeof( int ) );
192 /* Preserve a source's pixel aspect, if it's available. */
193 if( title->pixel_aspect_width && title->pixel_aspect_height )
195 job->pixel_aspect_width = title->pixel_aspect_width;
196 job->pixel_aspect_height = title->pixel_aspect_height;
199 if( title->aspect != 0 && title->aspect != 1. &&
200 !job->pixel_aspect_width && !job->pixel_aspect_height)
202 hb_reduce( &job->pixel_aspect_width, &job->pixel_aspect_height,
203 (int)(title->aspect * title->height), title->width );
206 job->width = title->width - job->crop[2] - job->crop[3];
207 hb_fix_aspect( job, HB_KEEP_WIDTH );
208 if( job->height > title->height - job->crop[0] - job->crop[1] )
210 job->height = title->height - job->crop[0] - job->crop[1];
211 hb_fix_aspect( job, HB_KEEP_HEIGHT );
214 hb_log( "scan: title (%d) job->width:%d, job->height:%d",
215 i, job->width, job->height );
219 job->vcodec = HB_VCODEC_FFMPEG;
220 job->vquality = -1.0;
221 job->vbitrate = 1000;
223 job->vrate = title->rate;
224 job->vrate_base = title->rate_base;
226 job->list_audio = hb_list_init();
230 job->mux = HB_MUX_MP4;
235 hb_dvd_close( &data->dvd );
239 hb_stream_close(&data->stream);
246 // -----------------------------------------------
247 // stuff related to cropping
251 static inline int absdiff( int x, int y )
253 return x < y ? y - x : x - y;
256 static inline int clampBlack( int x )
258 // luma 'black' is 16 and anything less should be clamped at 16
259 return x < 16 ? 16 : x;
262 static int row_all_dark( hb_title_t *title, uint8_t* luma, int row )
264 luma += title->width * row;
266 // compute the average luma value of the row
268 for ( i = 0; i < title->width; ++i )
270 avg += clampBlack( luma[i] );
276 // since we're trying to detect smooth borders, only take the row if
277 // all pixels are within +-16 of the average (this range is fairly coarse
278 // but there's a lot of quantization noise for luma values near black
279 // so anything less will fail to crop because of the noise).
280 for ( i = 0; i < title->width; ++i )
282 if ( absdiff( avg, clampBlack( luma[i] ) ) > 16 )
288 static int column_all_dark( hb_title_t *title, uint8_t* luma, int top, int bottom,
291 int stride = title->width;
292 int height = title->height - top - bottom;
293 luma += stride * top + col;
295 // compute the average value of the column
296 int i = height, avg = 0, row = 0;
297 for ( ; --i >= 0; row += stride )
299 avg += clampBlack( luma[row] );
305 // since we're trying to detect smooth borders, only take the column if
306 // all pixels are within +-16 of the average.
308 for ( ; --i >= 0; row += stride )
310 if ( absdiff( avg, clampBlack( luma[row] ) ) > 16 )
325 static void record_crop( crop_record_t *crops, int t, int b, int l, int r )
327 crops->t[crops->n] = t;
328 crops->b[crops->n] = b;
329 crops->l[crops->n] = l;
330 crops->r[crops->n] = r;
334 static int compare_int( const void *a, const void *b )
336 return *(const int *)a - *(const int *)b;
339 static void sort_crops( crop_record_t *crops )
341 qsort( crops->t, crops->n, sizeof(crops->t[0]), compare_int );
342 qsort( crops->b, crops->n, sizeof(crops->t[0]), compare_int );
343 qsort( crops->l, crops->n, sizeof(crops->t[0]), compare_int );
344 qsort( crops->r, crops->n, sizeof(crops->t[0]), compare_int );
347 // -----------------------------------------------
348 // stuff related to title width/height/aspect info
351 int count; /* number of times we've seen this info entry */
352 hb_work_info_t info; /* copy of info entry */
355 static void remember_info( info_list_t *info_list, hb_work_info_t *info )
357 for ( ; info_list->count; ++info_list )
359 if ( memcmp( &info_list->info, info, sizeof(*info) ) == 0 )
361 // we found a match - bump its count
366 // no match found - add new entry to list (info_list points to
367 // the first free slot). NB - we assume that info_list was allocated
368 // so that it's big enough even if there are no dups. I.e., 10 slots
369 // allocated if there are 10 previews.
370 info_list->count = 1;
371 info_list->info = *info;
374 static void most_common_info( info_list_t *info_list, hb_work_info_t *info )
377 for ( i = 1; info_list[i].count; ++i )
379 if ( info_list[i].count > info_list[biggest].count )
382 *info = info_list[biggest].info;
386 /***********************************************************************
388 ***********************************************************************
389 * Decode 10 pictures for the given title.
390 * It assumes that data->reader and data->vts have successfully been
391 * DVDOpen()ed and ifoOpen()ed.
392 **********************************************************************/
393 static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
395 int i, npreviews = 0;
396 hb_buffer_t * buf_ps, * buf_es;
398 int progressive_count = 0;
399 int interlaced_preview_count = 0;
400 info_list_t * info_list = calloc( 10+1, sizeof(*info_list) );
401 crop_record_t *crops = calloc( 1, sizeof(*crops) );
403 buf_ps = hb_buffer_init( HB_DVD_READ_BUFFER_SIZE );
404 list_es = hb_list_init();
406 hb_log( "scan: decoding previews for title %d", title->index );
409 hb_dvd_start( data->dvd, title->index, 1 );
411 for( i = 0; i < 10; i++ )
419 if( !hb_dvd_seek( data->dvd, (float) ( i + 1 ) / 11.0 ) )
424 else if (data->stream)
426 /* we start reading streams at zero rather than 1/11 because
427 * short streams may have only one sequence header in the entire
428 * file and we need it to decode any previews. */
429 if (!hb_stream_seek(data->stream, (float) i / 11.0 ) )
435 hb_deep_log( 2, "scan: preview %d", i + 1 );
437 int vcodec = title->video_codec? title->video_codec : WORK_DECMPEG2;
438 hb_work_object_t *vid_decoder = hb_get_work( vcodec );
439 vid_decoder->codec_param = title->video_codec_param;
440 vid_decoder->init( vid_decoder, NULL );
441 hb_buffer_t * vid_buf = NULL;
443 for( j = 0; j < 10240 ; j++ )
447 if( !hb_dvd_read( data->dvd, buf_ps ) )
449 hb_log( "Warning: Could not read data for preview %d, skipped", i + 1 );
453 else if (data->stream)
455 if ( !hb_stream_read(data->stream,buf_ps) )
457 hb_log( "Warning: Could not read data for preview %d, skipped", i + 1 );
461 if ( title->demuxer == HB_NULL_DEMUXER )
463 hb_demux_null( buf_ps, list_es, 0 );
467 hb_demux_ps( buf_ps, list_es, 0 );
470 while( ( buf_es = hb_list_item( list_es, 0 ) ) )
472 hb_list_rem( list_es, buf_es );
473 if( buf_es->id == title->video_id && vid_buf == NULL )
475 vid_decoder->work( vid_decoder, &buf_es, &vid_buf );
477 else if( ! AllAudioOK( title ) )
479 LookForAudio( title, buf_es );
482 hb_buffer_close( &buf_es );
485 if( vid_buf && AllAudioOK( title ) )
491 hb_log( "scan: could not get a decoded picture" );
495 /* Get size and rate infos */
497 hb_work_info_t vid_info;
498 if( !vid_decoder->info( vid_decoder, &vid_info ) )
501 * Could not fill vid_info, don't continue and try to use vid_info
504 vid_decoder->close( vid_decoder );
508 vid_decoder->close( vid_decoder );
511 remember_info( info_list, &vid_info );
513 title->video_codec_name = strdup( vid_info.name );
514 title->width = vid_info.width;
515 title->height = vid_info.height;
516 title->rate = vid_info.rate;
517 title->rate_base = vid_info.rate_base;
518 title->video_bitrate = vid_info.bitrate;
520 if( title->rate_base == 1126125 )
522 /* Frame FPS is 23.976 (meaning it's progressive), so
523 start keeping track of how many are reporting at
524 that speed. When enough show up that way, we want
525 to make that the overall title FPS.
529 if( progressive_count < 6 )
531 /* Not enough frames are reporting as progressive,
532 which means we should be conservative and use
533 29.97 as the title's FPS for now.
535 title->rate_base = 900900;
539 /* A majority of the scan frames are progressive. Make that
540 the title's FPS, and announce it once to the log.
542 if( progressive_count == 6 )
544 hb_deep_log( 2, "Title's mostly NTSC Film, setting fps to 23.976");
546 title->rate_base = 1126125;
549 else if( title->rate_base == 900900 && progressive_count >= 6 )
552 * We've already deduced that the frame rate is 23.976, so set it
555 title->rate_base = 1126125;
558 while( ( buf_es = hb_list_item( list_es, 0 ) ) )
560 hb_list_rem( list_es, buf_es );
561 hb_buffer_close( &buf_es );
564 /* Check preview for interlacing artifacts */
565 if( hb_detect_comb( vid_buf, title->width, title->height, 10, 30, 9, 10, 30, 9 ) )
567 hb_deep_log( 2, "Interlacing detected in preview frame %i", i+1);
568 interlaced_preview_count++;
571 hb_get_tempory_filename( data->h, filename, "%x%d",
572 (intptr_t)title, i );
574 file_preview = fopen( filename, "w" );
577 fwrite( vid_buf->data, title->width * title->height * 3 / 2,
579 fclose( file_preview );
583 hb_log( "scan: fopen failed (%s)", filename );
586 /* Detect black borders */
588 #define Y vid_buf->data
589 int top, bottom, left, right;
590 int h4 = title->height / 4, w4 = title->width / 4;
592 // When widescreen content is matted to 16:9 or 4:3 there's sometimes
593 // a thin border on the outer edge of the matte. On TV content it can be
594 // "line 21" VBI data that's normally hidden in the overscan. For HD
595 // content it can just be a diagnostic added in post production so that
596 // the frame borders are visible. We try to ignore these borders so
597 // we can crop the matte. The border width depends on the resolution
598 // (12 pixels on 1080i looks visually the same as 4 pixels on 480i)
599 // so we allow the border to be up to 1% of the frame height.
600 const int border = title->height / 100;
602 for ( top = border; top < h4; ++top )
604 if ( ! row_all_dark( title, Y, top ) )
609 // we never made it past the border region - see if the rows we
610 // didn't check are dark or if we shouldn't crop at all.
611 for ( top = 0; top < border; ++top )
613 if ( ! row_all_dark( title, Y, top ) )
621 for ( bottom = border; bottom < h4; ++bottom )
623 if ( ! row_all_dark( title, Y, title->height - 1 - bottom ) )
626 if ( bottom <= border )
628 for ( bottom = 0; bottom < border; ++bottom )
630 if ( ! row_all_dark( title, Y, title->height - 1 - bottom ) )
633 if ( bottom >= border )
638 for ( left = 0; left < w4; ++left )
640 if ( ! column_all_dark( title, Y, top, bottom, left ) )
643 for ( right = 0; right < w4; ++right )
645 if ( ! column_all_dark( title, Y, top, bottom, title->width - 1 - right ) )
649 // only record the result if all the crops are less than a quarter of
650 // the frame otherwise we can get fooled by frames with a lot of black
651 // like titles, credits & fade-thru-black transitions.
652 if ( top < h4 && bottom < h4 && left < w4 && right < w4 )
654 record_crop( crops, top, bottom, left, right );
660 hb_buffer_close( &vid_buf );
665 // use the most common frame info for our final title dimensions
666 hb_work_info_t vid_info;
667 most_common_info( info_list, &vid_info );
669 title->width = vid_info.width;
670 title->height = vid_info.height;
671 title->pixel_aspect_width = vid_info.pixel_aspect_width;
672 title->pixel_aspect_height = vid_info.pixel_aspect_height;
674 // compute the aspect ratio based on the storage dimensions and the
675 // pixel aspect ratio (if supplied) or just storage dimensions if no PAR.
676 title->aspect = (double)title->width / (double)title->height;
677 if( title->pixel_aspect_width && title->pixel_aspect_height )
679 title->aspect *= (double)title->pixel_aspect_width /
680 (double)title->pixel_aspect_height;
682 // For unknown reasons some French PAL DVDs put the original
683 // content's aspect ratio into the mpeg PAR even though it's
684 // the wrong PAR for the DVD. Apparently they rely on the fact
685 // that DVD players ignore the content PAR and just use the
686 // aspect ratio from the DVD metadata. So, if the aspect computed
687 // from the PAR is different from the container's aspect we use
688 // the container's aspect & recompute the PAR from it.
689 if( title->container_aspect && title->aspect != title->container_aspect )
691 hb_log("scan: content PAR gives wrong aspect %.2f; "
692 "using container aspect %.2f", title->aspect,
693 title->container_aspect );
694 title->aspect = title->container_aspect;
695 hb_reduce( &title->pixel_aspect_width, &title->pixel_aspect_height,
696 (int)(title->aspect * title->height), title->width );
700 // don't try to crop unless we got at least 3 previews
704 // The next line selects median cropping - at least
705 // 50% of the frames will have their borders removed.
706 // Other possible choices are loose cropping (i = 0) where
707 // no non-black pixels will be cropped from any frame and a
708 // tight cropping (i = crops->n - (crops->n >> 2)) where at
709 // least 75% of the frames will have their borders removed.
711 title->crop[0] = EVEN( crops->t[i] );
712 title->crop[1] = EVEN( crops->b[i] );
713 title->crop[2] = EVEN( crops->l[i] );
714 title->crop[3] = EVEN( crops->r[i] );
718 hb_log( "scan: %d previews, %dx%d, %.3f fps, autocrop = %d/%d/%d/%d, "
719 "aspect %s, PAR %d:%d",
720 npreviews, title->width, title->height, (float) title->rate /
721 (float) title->rate_base,
722 title->crop[0], title->crop[1], title->crop[2], title->crop[3],
723 aspect_to_string( title->aspect ), title->pixel_aspect_width,
724 title->pixel_aspect_height );
726 if( interlaced_preview_count >= ( npreviews / 2 ) )
728 hb_log("Title is likely interlaced or telecined (%i out of %i previews). You should do something about that.",
729 interlaced_preview_count, npreviews);
730 title->detected_interlacing = 1;
734 title->detected_interlacing = 0;
738 hb_buffer_close( &buf_ps );
739 while( ( buf_es = hb_list_item( list_es, 0 ) ) )
741 hb_list_rem( list_es, buf_es );
742 hb_buffer_close( &buf_es );
744 hb_list_close( &list_es );
746 hb_dvd_stop( data->dvd );
752 * This routine is called for every frame from a non-video elementary stream.
753 * These are a mix of audio & subtitle streams, some of which we want & some
754 * we're ignoring. This routine checks the frame against all our audio streams
755 * to see if it's one we want and haven't identified yet. If yes, it passes the
756 * frame to a codec-specific id routine which is responsible for filling in
757 * the sample rate, bit rate, channels & other audio parameters.
759 * Since a sample rate is essential for further audio processing, any audio
760 * stream which isn't successfully id'd by is deleted at the end of the scan.
761 * This is necessary to avoid ambiguities where things that might be audio
762 * aren't (e.g., some European DVD Teletext streams use the same IDs as US ATSC
765 static void LookForAudio( hb_title_t * title, hb_buffer_t * b )
769 hb_audio_t * audio = NULL;
770 for( i = 0; i < hb_list_count( title->list_audio ); i++ )
772 audio = hb_list_item( title->list_audio, i );
773 /* check if this elementary stream is one we want */
774 if ( audio->id == b->id )
783 if( !audio || audio->config.in.bitrate != 0 )
785 /* not found or already done */
789 hb_work_object_t *w = hb_codec_decoder( audio->config.in.codec );
791 if ( w == NULL || w->bsinfo == NULL )
793 hb_log( "Internal error in scan: unhandled audio type %d for id 0x%x",
794 audio->config.in.codec, audio->id );
800 w->codec_param = audio->config.in.codec_param;
801 int ret = w->bsinfo( w, b, &info );
804 hb_log( "no info on audio type %d/0x%x for id 0x%x",
805 audio->config.in.codec, audio->config.in.codec_param,
811 /* didn't find any info */
814 audio->config.in.samplerate = info.rate;
815 audio->config.in.bitrate = info.bitrate;
816 audio->config.in.channel_layout = info.channel_layout;
817 audio->config.in.version = info.version;
818 audio->config.in.mode = info.mode;
819 audio->config.flags.ac3 = info.flags;
821 // update the audio description string based on the info we found
822 if ( audio->config.flags.ac3 & AUDIO_F_DOLBY )
824 strcat( audio->config.lang.description, " (Dolby Surround)" );
828 int layout = audio->config.in.channel_layout;
829 char *desc = audio->config.lang.description +
830 strlen( audio->config.lang.description );
831 sprintf( desc, " (%d.%d ch)",
832 HB_INPUT_CH_LAYOUT_GET_DISCRETE_FRONT_COUNT(layout) +
833 HB_INPUT_CH_LAYOUT_GET_DISCRETE_REAR_COUNT(layout),
834 HB_INPUT_CH_LAYOUT_GET_DISCRETE_LFE_COUNT(layout) );
837 hb_log( "scan: audio 0x%x: %s, rate=%dHz, bitrate=%d %s", audio->id,
838 info.name, audio->config.in.samplerate, audio->config.in.bitrate,
839 audio->config.lang.description );
844 // We get here if there's no hope of finding info on an audio bitstream,
845 // either because we don't have a decoder (or a decoder with a bitstream
846 // info proc) or because the decoder's info proc said that the stream
847 // wasn't something it could handle. Delete the item from the title's
848 // audio list so we won't keep reading packets while trying to get its
854 hb_list_rem( title->list_audio, audio );
858 * This routine checks to see if we've ID'd all the audio streams associated
859 * with a title. It returns 0 if there are more to ID & 1 if all are done.
861 static int AllAudioOK( hb_title_t * title )
866 for( i = 0; i < hb_list_count( title->list_audio ); i++ )
868 audio = hb_list_item( title->list_audio, i );
869 if( audio->config.in.bitrate == 0 )