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 (hb_demux[title->demuxer])(buf_ps, list_es, 0 );
463 while( ( buf_es = hb_list_item( list_es, 0 ) ) )
465 hb_list_rem( list_es, buf_es );
466 if( buf_es->id == title->video_id && vid_buf == NULL )
468 vid_decoder->work( vid_decoder, &buf_es, &vid_buf );
470 else if( ! AllAudioOK( title ) )
472 LookForAudio( title, buf_es );
475 hb_buffer_close( &buf_es );
478 if( vid_buf && AllAudioOK( title ) )
484 hb_log( "scan: could not get a decoded picture" );
488 /* Get size and rate infos */
490 hb_work_info_t vid_info;
491 if( !vid_decoder->info( vid_decoder, &vid_info ) )
494 * Could not fill vid_info, don't continue and try to use vid_info
497 vid_decoder->close( vid_decoder );
501 vid_decoder->close( vid_decoder );
504 remember_info( info_list, &vid_info );
506 title->video_codec_name = strdup( vid_info.name );
507 title->width = vid_info.width;
508 title->height = vid_info.height;
509 title->rate = vid_info.rate;
510 title->rate_base = vid_info.rate_base;
511 title->video_bitrate = vid_info.bitrate;
513 if( title->rate_base == 1126125 )
515 /* Frame FPS is 23.976 (meaning it's progressive), so
516 start keeping track of how many are reporting at
517 that speed. When enough show up that way, we want
518 to make that the overall title FPS.
522 if( progressive_count < 6 )
524 /* Not enough frames are reporting as progressive,
525 which means we should be conservative and use
526 29.97 as the title's FPS for now.
528 title->rate_base = 900900;
532 /* A majority of the scan frames are progressive. Make that
533 the title's FPS, and announce it once to the log.
535 if( progressive_count == 6 )
537 hb_deep_log( 2, "Title's mostly NTSC Film, setting fps to 23.976");
539 title->rate_base = 1126125;
542 else if( title->rate_base == 900900 && progressive_count >= 6 )
545 * We've already deduced that the frame rate is 23.976, so set it
548 title->rate_base = 1126125;
551 while( ( buf_es = hb_list_item( list_es, 0 ) ) )
553 hb_list_rem( list_es, buf_es );
554 hb_buffer_close( &buf_es );
557 /* Check preview for interlacing artifacts */
558 if( hb_detect_comb( vid_buf, title->width, title->height, 10, 30, 9, 10, 30, 9 ) )
560 hb_deep_log( 2, "Interlacing detected in preview frame %i", i+1);
561 interlaced_preview_count++;
564 hb_get_tempory_filename( data->h, filename, "%x%d",
565 (intptr_t)title, i );
567 file_preview = fopen( filename, "w" );
570 fwrite( vid_buf->data, title->width * title->height * 3 / 2,
572 fclose( file_preview );
576 hb_log( "scan: fopen failed (%s)", filename );
579 /* Detect black borders */
581 #define Y vid_buf->data
582 int top, bottom, left, right;
583 int h4 = title->height / 4, w4 = title->width / 4;
585 // When widescreen content is matted to 16:9 or 4:3 there's sometimes
586 // a thin border on the outer edge of the matte. On TV content it can be
587 // "line 21" VBI data that's normally hidden in the overscan. For HD
588 // content it can just be a diagnostic added in post production so that
589 // the frame borders are visible. We try to ignore these borders so
590 // we can crop the matte. The border width depends on the resolution
591 // (12 pixels on 1080i looks visually the same as 4 pixels on 480i)
592 // so we allow the border to be up to 1% of the frame height.
593 const int border = title->height / 100;
595 for ( top = border; top < h4; ++top )
597 if ( ! row_all_dark( title, Y, top ) )
602 // we never made it past the border region - see if the rows we
603 // didn't check are dark or if we shouldn't crop at all.
604 for ( top = 0; top < border; ++top )
606 if ( ! row_all_dark( title, Y, top ) )
614 for ( bottom = border; bottom < h4; ++bottom )
616 if ( ! row_all_dark( title, Y, title->height - 1 - bottom ) )
619 if ( bottom <= border )
621 for ( bottom = 0; bottom < border; ++bottom )
623 if ( ! row_all_dark( title, Y, title->height - 1 - bottom ) )
626 if ( bottom >= border )
631 for ( left = 0; left < w4; ++left )
633 if ( ! column_all_dark( title, Y, top, bottom, left ) )
636 for ( right = 0; right < w4; ++right )
638 if ( ! column_all_dark( title, Y, top, bottom, title->width - 1 - right ) )
642 // only record the result if all the crops are less than a quarter of
643 // the frame otherwise we can get fooled by frames with a lot of black
644 // like titles, credits & fade-thru-black transitions.
645 if ( top < h4 && bottom < h4 && left < w4 && right < w4 )
647 record_crop( crops, top, bottom, left, right );
653 hb_buffer_close( &vid_buf );
658 // use the most common frame info for our final title dimensions
659 hb_work_info_t vid_info;
660 most_common_info( info_list, &vid_info );
662 title->width = vid_info.width;
663 title->height = vid_info.height;
664 title->pixel_aspect_width = vid_info.pixel_aspect_width;
665 title->pixel_aspect_height = vid_info.pixel_aspect_height;
667 // compute the aspect ratio based on the storage dimensions and the
668 // pixel aspect ratio (if supplied) or just storage dimensions if no PAR.
669 title->aspect = (double)title->width / (double)title->height;
670 if( title->pixel_aspect_width && title->pixel_aspect_height )
672 title->aspect *= (double)title->pixel_aspect_width /
673 (double)title->pixel_aspect_height;
675 // For unknown reasons some French PAL DVDs put the original
676 // content's aspect ratio into the mpeg PAR even though it's
677 // the wrong PAR for the DVD. Apparently they rely on the fact
678 // that DVD players ignore the content PAR and just use the
679 // aspect ratio from the DVD metadata. So, if the aspect computed
680 // from the PAR is different from the container's aspect we use
681 // the container's aspect & recompute the PAR from it.
682 if( title->container_aspect && title->aspect != title->container_aspect )
684 hb_log("scan: content PAR gives wrong aspect %.2f; "
685 "using container aspect %.2f", title->aspect,
686 title->container_aspect );
687 title->aspect = title->container_aspect;
688 hb_reduce( &title->pixel_aspect_width, &title->pixel_aspect_height,
689 (int)(title->aspect * title->height), title->width );
693 // don't try to crop unless we got at least 3 previews
697 // The next line selects median cropping - at least
698 // 50% of the frames will have their borders removed.
699 // Other possible choices are loose cropping (i = 0) where
700 // no non-black pixels will be cropped from any frame and a
701 // tight cropping (i = crops->n - (crops->n >> 2)) where at
702 // least 75% of the frames will have their borders removed.
704 title->crop[0] = EVEN( crops->t[i] );
705 title->crop[1] = EVEN( crops->b[i] );
706 title->crop[2] = EVEN( crops->l[i] );
707 title->crop[3] = EVEN( crops->r[i] );
711 hb_log( "scan: %d previews, %dx%d, %.3f fps, autocrop = %d/%d/%d/%d, "
712 "aspect %s, PAR %d:%d",
713 npreviews, title->width, title->height, (float) title->rate /
714 (float) title->rate_base,
715 title->crop[0], title->crop[1], title->crop[2], title->crop[3],
716 aspect_to_string( title->aspect ), title->pixel_aspect_width,
717 title->pixel_aspect_height );
719 if( interlaced_preview_count >= ( npreviews / 2 ) )
721 hb_log("Title is likely interlaced or telecined (%i out of %i previews). You should do something about that.",
722 interlaced_preview_count, npreviews);
723 title->detected_interlacing = 1;
727 title->detected_interlacing = 0;
731 hb_buffer_close( &buf_ps );
732 while( ( buf_es = hb_list_item( list_es, 0 ) ) )
734 hb_list_rem( list_es, buf_es );
735 hb_buffer_close( &buf_es );
737 hb_list_close( &list_es );
739 hb_dvd_stop( data->dvd );
745 * This routine is called for every frame from a non-video elementary stream.
746 * These are a mix of audio & subtitle streams, some of which we want & some
747 * we're ignoring. This routine checks the frame against all our audio streams
748 * to see if it's one we want and haven't identified yet. If yes, it passes the
749 * frame to a codec-specific id routine which is responsible for filling in
750 * the sample rate, bit rate, channels & other audio parameters.
752 * Since a sample rate is essential for further audio processing, any audio
753 * stream which isn't successfully id'd by is deleted at the end of the scan.
754 * This is necessary to avoid ambiguities where things that might be audio
755 * aren't (e.g., some European DVD Teletext streams use the same IDs as US ATSC
758 static void LookForAudio( hb_title_t * title, hb_buffer_t * b )
762 hb_audio_t * audio = NULL;
763 for( i = 0; i < hb_list_count( title->list_audio ); i++ )
765 audio = hb_list_item( title->list_audio, i );
766 /* check if this elementary stream is one we want */
767 if ( audio->id == b->id )
776 if( !audio || audio->config.in.bitrate != 0 )
778 /* not found or already done */
782 hb_work_object_t *w = hb_codec_decoder( audio->config.in.codec );
784 if ( w == NULL || w->bsinfo == NULL )
786 hb_log( "Internal error in scan: unhandled audio type %d for id 0x%x",
787 audio->config.in.codec, audio->id );
793 w->codec_param = audio->config.in.codec_param;
794 int ret = w->bsinfo( w, b, &info );
797 hb_log( "no info on audio type %d/0x%x for id 0x%x",
798 audio->config.in.codec, audio->config.in.codec_param,
804 /* didn't find any info */
807 audio->config.in.samplerate = info.rate;
808 audio->config.in.bitrate = info.bitrate;
809 audio->config.in.channel_layout = info.channel_layout;
810 audio->config.in.version = info.version;
811 audio->config.in.mode = info.mode;
812 audio->config.flags.ac3 = info.flags;
814 // update the audio description string based on the info we found
815 if ( audio->config.flags.ac3 & AUDIO_F_DOLBY )
817 strcat( audio->config.lang.description, " (Dolby Surround)" );
821 int layout = audio->config.in.channel_layout;
822 char *desc = audio->config.lang.description +
823 strlen( audio->config.lang.description );
824 sprintf( desc, " (%d.%d ch)",
825 HB_INPUT_CH_LAYOUT_GET_DISCRETE_FRONT_COUNT(layout) +
826 HB_INPUT_CH_LAYOUT_GET_DISCRETE_REAR_COUNT(layout),
827 HB_INPUT_CH_LAYOUT_GET_DISCRETE_LFE_COUNT(layout) );
830 hb_log( "scan: audio 0x%x: %s, rate=%dHz, bitrate=%d %s", audio->id,
831 info.name, audio->config.in.samplerate, audio->config.in.bitrate,
832 audio->config.lang.description );
837 // We get here if there's no hope of finding info on an audio bitstream,
838 // either because we don't have a decoder (or a decoder with a bitstream
839 // info proc) or because the decoder's info proc said that the stream
840 // wasn't something it could handle. Delete the item from the title's
841 // audio list so we won't keep reading packets while trying to get its
847 hb_list_rem( title->list_audio, audio );
851 * This routine checks to see if we've ID'd all the audio streams associated
852 * with a title. It returns 0 if there are more to ID & 1 if all are done.
854 static int AllAudioOK( hb_title_t * title )
859 for( i = 0; i < hb_list_count( title->list_audio ); i++ )
861 audio = hb_list_item( title->list_audio, i );
862 if( audio->config.in.bitrate == 0 )