X-Git-Url: http://git.osdn.net/view?a=blobdiff_plain;f=libhb%2Fdecomb.c;fp=libhb%2Fdecomb.c;h=584a3408bf73ca94b65f43571d2a3490dac7929d;hb=57c0057e6e20056df7fbd68694a34e603f9f11bd;hp=2194f56335ebcf0668d6834ce7a4bb1d267a76b0;hpb=19bdd6d024eae6d20bee454c5f5320ba4583460f;p=handbrake-jp%2Fhandbrake-jp.git diff --git a/libhb/decomb.c b/libhb/decomb.c index 2194f56..584a340 100644 --- a/libhb/decomb.c +++ b/libhb/decomb.c @@ -4,14 +4,74 @@ Homepage: . It may be used under the terms of the GNU General Public License. - The yadif algorithm was created by Michael Niedermayer. */ + The yadif algorithm was created by Michael Niedermayer. + Tritical's work inspired much of the comb detection code: + http://web.missouri.edu/~kes25c/ +*/ + +/***** +Parameters: + Mode : Spatial metric : Motion thresh : Spatial thresh : Block thresh : + Block width : Block height + +Appended for EEDI2: + Magnitude thresh : Variance thresh : Laplacian thresh : Dilation thresh : + Erosion thresh : Noise thresh : Max search distance : Post-processing + +Plus: + Parity + +Defaults: + 7:2:6:9:80:16:16:10:20:20:4:2:50:24:1:-1 +*****/ + +#define MODE_YADIF 1 // Use yadif +#define MODE_BLEND 2 // Use blending interpolation +#define MODE_CUBIC 4 // Use cubic interpolation +#define MODE_EEDI2 8 // Use EEDI2 interpolation +#define MODE_MCDEINT 16 // Post-process with mcdeint +#define MODE_MASK 32 // Output combing masks instead of pictures + +/***** +These modes can be layered. For example, Yadif (1) + EEDI2 (8) = 9, +which will feed EEDI2 interpolations to yadif. + +** Working combos: + 1: Just yadif + 2: Just blend + 3: Switch between yadif and blend + 4: Just cubic interpolate + 5: Cubic->yadif + 6: Switch between cubic and blend + 7: Switch between cubic->yadif and blend + 8: Just EEDI2 interpolate + 9: EEDI2->yadif +10: Switch between EEDI2 and blend +11: Switch between EEDI2->yadif and blend +17: Yadif->mcdeint +18: Blend->mcdeint +19: Switch between blending and yadif -> mcdeint +20: Cubic->mdeint +21: Cubic->yadif->mcdeint +22: Cubic or blend -> mcdeint +23: Cubic->yadif or blend -> mcdeint +24: EEDI2->mcdeint +25: EEDI2->yadif->mcdeint +...okay I'm getting bored now listing all these different modes +32: Passes through the combing mask for every combed frame (white for combed pixels, otherwise black) +33+: Overlay the combing mask for every combed frame on top of the filtered output (white for combed pixels) + +12-15: EEDI2 will override cubic interpolation +16: DOES NOT WORK BY ITSELF-- mcdeint needs to be fed by another deinterlacer +*****/ + #include "hb.h" -#include "libavcodec/avcodec.h" +#include "hbffmpeg.h" #include "mpeg2dec/mpeg2.h" +#include "eedi2.h" #define SUPPRESS_AV_LOG -#define MODE_DEFAULT 1 #define PARITY_DEFAULT -1 #define MCDEINT_MODE_DEFAULT -1 @@ -21,17 +81,52 @@ #define MIN3(a,b,c) MIN(MIN(a,b),c) #define MAX3(a,b,c) MAX(MAX(a,b),c) -typedef struct yadif_arguments_s { +// Some names to correspond to the pv->eedi_half array's contents +#define SRCPF 0 +#define MSKPF 1 +#define TMPPF 2 +#define DSTPF 3 +// Some names to correspond to the pv->eedi_full array's contents +#define DST2PF 0 +#define TMP2PF2 1 +#define MSK2PF 2 +#define TMP2PF 3 +#define DST2MPF 4 + +struct yadif_arguments_s { uint8_t **dst; int parity; int tff; int stop; int is_combed; -} yadif_arguments_t; +}; + +struct decomb_arguments_s { + int stop; +}; -typedef struct decomb_arguments_s { +struct eedi2_arguments_s { int stop; -} decomb_arguments_t; +}; + +typedef struct yadif_arguments_s yadif_arguments_t; +typedef struct decomb_arguments_s decomb_arguments_t; +typedef struct eedi2_arguments_s eedi2_arguments_t; + +typedef struct eedi2_thread_arg_s { + hb_filter_private_t *pv; + int plane; +} eedi2_thread_arg_t; + +typedef struct decomb_thread_arg_s { + hb_filter_private_t *pv; + int segment; +} decomb_thread_arg_t; + +typedef struct yadif_thread_arg_s { + hb_filter_private_t *pv; + int segment; +} yadif_thread_arg_t; struct hb_filter_private_s { @@ -39,6 +134,7 @@ struct hb_filter_private_s int width[3]; int height[3]; + // Decomb parameters int mode; int spatial_metric; int motion_threshold; @@ -46,8 +142,19 @@ struct hb_filter_private_s int block_threshold; int block_width; int block_height; + + // EEDI2 parameters + int magnitude_threshold; + int variance_threshold; + int laplacian_threshold; + int dilation_threshold; + int erosion_threshold; + int noise_threshold; + int maximum_search_distance; + int post_processing; int parity; + int tff; int yadif_ready; @@ -60,8 +167,8 @@ struct hb_filter_private_s AVFrame * mcdeint_frame; AVFrame * mcdeint_frame_dec; - int yadif_deinterlaced_frames; - int blend_deinterlaced_frames; + int deinterlaced_frames; + int blended_frames; int unfiltered_frames; uint8_t * ref[4][3]; @@ -70,6 +177,13 @@ struct hb_filter_private_s /* Make a buffer to store a comb mask. */ uint8_t * mask[3]; + uint8_t * eedi_half[4][3]; + uint8_t * eedi_full[5][3]; + int * cx2; + int * cy2; + int * cxy; + int * tmpc; + AVPicture pic_in; AVPicture pic_out; hb_buffer_t * buf_out[2]; @@ -86,7 +200,13 @@ struct hb_filter_private_s hb_lock_t ** decomb_begin_lock; // Thread has work hb_lock_t ** decomb_complete_lock; // Thread has completed work decomb_arguments_t *decomb_arguments; // Arguments to thread for work - + + hb_thread_t ** eedi2_threads; // Threads for eedi2 - one per plane + hb_lock_t ** eedi2_begin_lock; // Thread has work + hb_lock_t ** eedi2_complete_lock; // Thread has completed work + eedi2_arguments_t *eedi2_arguments; // Arguments to thread for work + +// int alternator; // for bobbing parity when framedoubling }; hb_filter_private_t * hb_decomb_init( int pix_fmt, @@ -106,14 +226,14 @@ void hb_decomb_close( hb_filter_private_t * pv ); hb_filter_object_t hb_filter_decomb = { FILTER_DECOMB, - "Deinterlaces selectively with yadif/mcdeint and lowpass5 blending", + "Decomb", NULL, hb_decomb_init, hb_decomb_work, hb_decomb_close, }; -int cubic_interpolate( int y0, int y1, int y2, int y3 ) +int cubic_interpolate_pixel( int y0, int y1, int y2, int y3 ) { /* From http://www.neuron2.net/library/cubicinterp.html */ int result = ( y0 * -3 ) + ( y1 * 23 ) + ( y2 * 23 ) + ( y3 * -3 ); @@ -131,6 +251,111 @@ int cubic_interpolate( int y0, int y1, int y2, int y3 ) return result; } +static void cubic_interpolate_line( uint8_t *dst, + uint8_t *cur, + int plane, + int y, + hb_filter_private_t * pv ) +{ + int w = pv->width[plane]; + int refs = pv->ref_stride[plane]; + int x; + + for( x = 0; x < w; x++) + { + int a, b, c, d; + a = b = c = d = 0; + + if( y >= 3 ) + { + /* Normal top*/ + a = cur[-3*refs]; + b = cur[-refs]; + } + else if( y == 2 || y == 1 ) + { + /* There's only one sample above this pixel, use it twice. */ + a = cur[-refs]; + b = cur[-refs]; + } + else if( y == 0 ) + { + /* No samples above, triple up on the one below. */ + a = cur[+refs]; + b = cur[+refs]; + } + + if( y <= ( pv->height[plane] - 4 ) ) + { + /* Normal bottom*/ + c = cur[+refs]; + d = cur[3*refs]; + } + else if( y == ( pv->height[plane] - 3 ) || y == ( pv->height[plane] - 2 ) ) + { + /* There's only one sample below, use it twice. */ + c = cur[+refs]; + d = cur[+refs]; + } + else if( y == pv->height[plane] - 1) + { + /* No samples below, triple up on the one above. */ + c = cur[-refs]; + d = cur[-refs]; + } + + dst[0] = cubic_interpolate_pixel( a, b, c, d ); + + dst++; + cur++; + } +} + +void apply_mask_line( uint8_t * srcp, + uint8_t * mskp, + int width ) +{ + int x; + + for( x = 0; x < width; x++ ) + { + if( mskp[x] == 255 ) + { + srcp[x] = 255; + } + } +} + +void apply_mask( hb_filter_private_t * pv ) +{ + int plane, height; + + for( plane = 0; plane < 3; plane++ ) + { + uint8_t * srcp = ( pv->mode & MODE_MCDEINT ) ? pv->pic_in.data[plane] : pv->pic_out.data[plane]; + uint8_t * mskp = pv->mask[plane]; + + for( height = 0; height < pv->height[plane]; height++ ) + { + if( pv->mode == MODE_MASK && plane == 0 ) + { + memcpy( srcp, mskp, pv->width[plane] ); + } + else if( pv->mode == MODE_MASK ) + { + memset( srcp, 128, pv->width[plane] ); + } + else if( plane == 0 ) + { + apply_mask_line( srcp, mskp, pv->width[plane] ); + } + + srcp += pv->pic_out.linesize[plane]; + mskp += pv->ref_stride[plane]; + } + } +} + static void store_ref( const uint8_t ** pic, hb_filter_private_t * pv ) { @@ -153,7 +378,7 @@ static void store_ref( const uint8_t ** pic, int ref_stride = pv->ref_stride[i]; int y; - for( y = 0; y < pv->height[i]; y++ ) + for( y = 0; y < h; y++ ) { memcpy(ref, src, w); src = (uint8_t*)src + w; @@ -162,6 +387,8 @@ static void store_ref( const uint8_t ** pic, } } +/* This function may be useful in the future, if we want to output + a reference to an AVPicture, since they have different strides. static void get_ref( uint8_t ** pic, hb_filter_private_t * pv, int frm ) { int i; @@ -181,6 +408,7 @@ static void get_ref( uint8_t ** pic, hb_filter_private_t * pv, int frm ) } } } +*/ int blend_filter_pixel( int up2, int up1, int current, int down1, int down2 ) { @@ -316,7 +544,7 @@ int check_combing_mask( hb_filter_private_t * pv ) if( block_score >= ( threshold / 2 ) ) { #if 0 - hb_log("decomb: frame %i | score %i | type %s", pv->yadif_deinterlaced_frames + pv->blend_deinterlaced_frames + pv->unfiltered_frames + 1, block_score, pv->buf_settings->flags & 16 ? "Film" : "Video"); + hb_log("decomb: frame %i | score %i | type %s", pv->deinterlaced_frames + pv->blended_frames + pv->unfiltered_frames + 1, block_score, pv->buf_settings->flags & 16 ? "Film" : "Video"); #endif if ( block_score <= threshold && !( pv->buf_settings->flags & 16) ) { @@ -353,7 +581,7 @@ int check_combing_mask( hb_filter_private_t * pv ) } } -int detect_combed_segment( hb_filter_private_t * pv, int segment_start, int segment_stop ) +void detect_combed_segment( hb_filter_private_t * pv, int segment_start, int segment_stop ) { /* A mish-mash of various comb detection tricks picked up from neuron2's Decomb plugin for @@ -449,7 +677,7 @@ int detect_combed_segment( hb_filter_private_t * pv, int segment_start, int segm motion = 1; } - if( motion || ( pv->yadif_deinterlaced_frames==0 && pv->blend_deinterlaced_frames==0 && pv->unfiltered_frames==0) ) + if( motion || ( pv->deinterlaced_frames==0 && pv->blended_frames==0 && pv->unfiltered_frames==0) ) { /* That means it's time for the spatial check. We've got several options here. */ @@ -510,10 +738,171 @@ int detect_combed_segment( hb_filter_private_t * pv, int segment_start, int segm } } -typedef struct decomb_thread_arg_s { - hb_filter_private_t *pv; - int segment; -} decomb_thread_arg_t; +// This function calls all the eedi2 filters in sequence for a given plane. +// It outputs the final interpolated image to pv->eedi_full[DST2PF]. +void eedi2_interpolate_plane( hb_filter_private_t * pv, int k ) +{ + /* We need all these pointers. No, seriously. + I swear. It's not a joke. They're used. + All nine of them. */ + uint8_t * mskp = pv->eedi_half[MSKPF][k]; + uint8_t * srcp = pv->eedi_half[SRCPF][k]; + uint8_t * tmpp = pv->eedi_half[TMPPF][k]; + uint8_t * dstp = pv->eedi_half[DSTPF][k]; + uint8_t * dst2p = pv->eedi_full[DST2PF][k]; + uint8_t * tmp2p2 = pv->eedi_full[TMP2PF2][k]; + uint8_t * msk2p = pv->eedi_full[MSK2PF][k]; + uint8_t * tmp2p = pv->eedi_full[TMP2PF][k]; + uint8_t * dst2mp = pv->eedi_full[DST2MPF][k]; + int * cx2 = pv->cx2; + int * cy2 = pv->cy2; + int * cxy = pv->cxy; + int * tmpc = pv->tmpc; + + int pitch = pv->ref_stride[k]; + int height = pv->height[k]; int width = pv->width[k]; + int half_height = height / 2; + + // edge mask + eedi2_build_edge_mask( mskp, pitch, srcp, pitch, + pv->magnitude_threshold, pv->variance_threshold, pv->laplacian_threshold, + half_height, width ); + eedi2_erode_edge_mask( mskp, pitch, tmpp, pitch, pv->erosion_threshold, half_height, width ); + eedi2_dilate_edge_mask( tmpp, pitch, mskp, pitch, pv->dilation_threshold, half_height, width ); + eedi2_erode_edge_mask( mskp, pitch, tmpp, pitch, pv->erosion_threshold, half_height, width ); + eedi2_remove_small_gaps( tmpp, pitch, mskp, pitch, half_height, width ); + + // direction mask + eedi2_calc_directions( k, mskp, pitch, srcp, pitch, tmpp, pitch, + pv->maximum_search_distance, pv->noise_threshold, + half_height, width ); + eedi2_filter_dir_map( mskp, pitch, tmpp, pitch, dstp, pitch, half_height, width ); + eedi2_expand_dir_map( mskp, pitch, dstp, pitch, tmpp, pitch, half_height, width ); + eedi2_filter_map( mskp, pitch, tmpp, pitch, dstp, pitch, half_height, width ); + + // upscale 2x vertically + eedi2_upscale_by_2( srcp, dst2p, half_height, pitch ); + eedi2_upscale_by_2( dstp, tmp2p2, half_height, pitch ); + eedi2_upscale_by_2( mskp, msk2p, half_height, pitch ); + + // upscale the direction mask + eedi2_mark_directions_2x( msk2p, pitch, tmp2p2, pitch, tmp2p, pitch, pv->tff, height, width ); + eedi2_filter_dir_map_2x( msk2p, pitch, tmp2p, pitch, dst2mp, pitch, pv->tff, height, width ); + eedi2_expand_dir_map_2x( msk2p, pitch, dst2mp, pitch, tmp2p, pitch, pv->tff, height, width ); + eedi2_fill_gaps_2x( msk2p, pitch, tmp2p, pitch, dst2mp, pitch, pv->tff, height, width ); + eedi2_fill_gaps_2x( msk2p, pitch, dst2mp, pitch, tmp2p, pitch, pv->tff, height, width ); + + // interpolate a full-size plane + eedi2_interpolate_lattice( k, tmp2p, pitch, dst2p, pitch, tmp2p2, pitch, pv->tff, + pv->noise_threshold, height, width ); + + if( pv->post_processing == 1 || pv->post_processing == 3 ) + { + // make sure the edge directions are consistent + eedi2_bit_blit( tmp2p2, pitch, tmp2p, pitch, pv->width[k], pv->height[k] ); + eedi2_filter_dir_map_2x( msk2p, pitch, tmp2p, pitch, dst2mp, pitch, pv->tff, height, width ); + eedi2_expand_dir_map_2x( msk2p, pitch, dst2mp, pitch, tmp2p, pitch, pv->tff, height, width ); + eedi2_post_process( tmp2p, pitch, tmp2p2, pitch, dst2p, pitch, pv->tff, height, width ); + } + if( pv->post_processing == 2 || pv->post_processing == 3 ) + { + // filter junctions and corners + eedi2_gaussian_blur1( srcp, pitch, tmpp, pitch, srcp, pitch, half_height, width ); + eedi2_calc_derivatives( srcp, pitch, half_height, width, cx2, cy2, cxy ); + eedi2_gaussian_blur_sqrt2( cx2, tmpc, cx2, pitch, half_height, width); + eedi2_gaussian_blur_sqrt2( cy2, tmpc, cy2, pitch, half_height, width); + eedi2_gaussian_blur_sqrt2( cxy, tmpc, cxy, pitch, half_height, width); + eedi2_post_process_corner( cx2, cy2, cxy, pitch, tmp2p2, pitch, dst2p, pitch, height, width, pv->tff ); + } +} + +/* + * eedi2 interpolate this plane in a single thread. + */ +void eedi2_filter_thread( void *thread_args_v ) +{ + eedi2_arguments_t *eedi2_work = NULL; + hb_filter_private_t * pv; + int run = 1; + int plane; + eedi2_thread_arg_t *thread_args = thread_args_v; + + pv = thread_args->pv; + plane = thread_args->plane; + + hb_log("eedi2 thread started for plane %d", plane); + + while( run ) + { + /* + * Wait here until there is work to do. hb_lock() blocks until + * render releases it to say that there is more work to do. + */ + hb_lock( pv->eedi2_begin_lock[plane] ); + + eedi2_work = &pv->eedi2_arguments[plane]; + + if( eedi2_work->stop ) + { + /* + * No more work to do, exit this thread. + */ + run = 0; + continue; + } + + /* + * Process plane + */ + eedi2_interpolate_plane( pv, plane ); + + /* + * Finished this segment, let everyone know. + */ + hb_unlock( pv->eedi2_complete_lock[plane] ); + } + free( thread_args_v ); +} + +// Sets up the input field planes for EEDI2 in pv->eedi_half[SRCPF] +// and then runs eedi2_filter_thread for each plane. +void eedi2_planer( hb_filter_private_t * pv ) +{ + /* Copy the first field from the source to a half-height frame. */ + int i; + for( i = 0; i < 3; i++ ) + { + int pitch = pv->ref_stride[i]; + int start_line = !pv->tff; + eedi2_fill_half_height_buffer_plane( &pv->ref[1][i][pitch*start_line], pv->eedi_half[SRCPF][i], pitch, pv->height[i] ); + } + + int plane; + for( plane = 0; plane < 3; plane++ ) + { + /* + * Let the thread for this plane know that we've setup work + * for it by releasing the begin lock (ensuring that the + * complete lock is already locked so that we block when + * we try to lock it again below). + */ + hb_lock( pv->eedi2_complete_lock[plane] ); + hb_unlock( pv->eedi2_begin_lock[plane] ); + } + + /* + * Wait until all three threads have completed by trying to get + * the complete lock that we locked earlier for each thread, which + * will block until that thread has completed the work on that + * plane. + */ + for( plane = 0; plane < 3; plane++ ) + { + hb_lock( pv->eedi2_complete_lock[plane] ); + hb_unlock( pv->eedi2_complete_lock[plane] ); + } +} + /* * comb detect this segment of all three planes in a single thread. @@ -556,9 +945,7 @@ void decomb_filter_thread( void *thread_args_v ) for( plane = 0; plane < 1; plane++) { - int w = pv->width[plane]; int h = pv->height[plane]; - int ref_stride = pv->ref_stride[plane]; segment_start = ( h / pv->cpu_count ) * segment; if( segment == pv->cpu_count - 1 ) { @@ -626,16 +1013,21 @@ static void yadif_filter_line( uint8_t *dst, to the other field in the current frame--the one not being filtered. */ uint8_t *prev2 = parity ? prev : cur ; uint8_t *next2 = parity ? cur : next; + int w = pv->width[plane]; int refs = pv->ref_stride[plane]; int x; + int eedi2_mode = ( pv->mode & MODE_EEDI2 ); + /* We can replace spatial_pred with this interpolation*/ + uint8_t * eedi2_guess = &pv->eedi_full[DST2PF][plane][y*refs]; + /* Decomb's cubic interpolation can only function when there are three samples above and below, so regress to yadif's traditional two-tap interpolation when filtering at the top and bottom edges. */ - int edge = 0; + int vertical_edge = 0; if( ( y < 3 ) || ( y > ( pv->height[plane] - 4 ) ) ) - edge = 1; + vertical_edge = 1; for( x = 0; x < w; x++) { @@ -654,60 +1046,83 @@ static void yadif_filter_line( uint8_t *dst, int temporal_diff2 = ( ABS(next[-refs] - cur[-refs]) + ABS(next[+refs] - cur[+refs]) ) >> 1; /* For the actual difference, use the largest of the previous average diffs. */ int diff = MAX3(temporal_diff0>>1, temporal_diff1, temporal_diff2); - - /* SAD of how the pixel-1, the pixel, and the pixel+1 change from the line above to below. */ - int spatial_score = ABS(cur[-refs-1] - cur[+refs-1]) + ABS(cur[-refs]-cur[+refs]) + - ABS(cur[-refs+1] - cur[+refs+1]) - 1; + int spatial_pred; - - /* Spatial pred is either a bilinear or cubic vertical interpolation. */ - if( pv->mode > 0 && !edge) + + if( eedi2_mode ) { - spatial_pred = cubic_interpolate( cur[-3*refs], cur[-refs], cur[+refs], cur[3*refs] ); + /* Who needs yadif's spatial predictions when we can have EEDI2's? */ + spatial_pred = eedi2_guess[0]; + eedi2_guess++; } - else + else // Yadif spatial interpolation { - spatial_pred = (c+e)>>1; + /* SAD of how the pixel-1, the pixel, and the pixel+1 change from the line above to below. */ + int spatial_score = ABS(cur[-refs-1] - cur[+refs-1]) + ABS(cur[-refs]-cur[+refs]) + + ABS(cur[-refs+1] - cur[+refs+1]) - 1; + + /* Spatial pred is either a bilinear or cubic vertical interpolation. */ + if( ( pv->mode & MODE_CUBIC ) && !vertical_edge) + { + spatial_pred = cubic_interpolate_pixel( cur[-3*refs], cur[-refs], cur[+refs], cur[3*refs] ); + } + else + { + spatial_pred = (c+e)>>1; + } + + /* EDDI: Edge Directed Deinterlacing Interpolation + Checks 4 different slopes to see if there is more similarity along a diagonal + than there was vertically. If a diagonal is more similar, then it indicates + an edge, so interpolate along that instead of a vertical line, using either + linear or cubic interpolation depending on mode. */ + #define YADIF_CHECK(j)\ + { int score = ABS(cur[-refs-1+j] - cur[+refs-1-j])\ + + ABS(cur[-refs +j] - cur[+refs -j])\ + + ABS(cur[-refs+1+j] - cur[+refs+1-j]);\ + if( score < spatial_score ){\ + spatial_score = score;\ + if( ( pv->mode & MODE_CUBIC ) && !vertical_edge )\ + {\ + switch(j)\ + {\ + case -1:\ + spatial_pred = cubic_interpolate_pixel(cur[-3 * refs - 3], cur[-refs -1], cur[+refs + 1], cur[3* refs + 3] );\ + break;\ + case -2:\ + spatial_pred = cubic_interpolate_pixel( ( ( cur[-3*refs - 4] + cur[-refs - 4] ) / 2 ) , cur[-refs -2], cur[+refs + 2], ( ( cur[3*refs + 4] + cur[refs + 4] ) / 2 ) );\ + break;\ + case 1:\ + spatial_pred = cubic_interpolate_pixel(cur[-3 * refs +3], cur[-refs +1], cur[+refs - 1], cur[3* refs -3] );\ + break;\ + case 2:\ + spatial_pred = cubic_interpolate_pixel(( ( cur[-3*refs + 4] + cur[-refs + 4] ) / 2 ), cur[-refs +2], cur[+refs - 2], ( ( cur[3*refs - 4] + cur[refs - 4] ) / 2 ) );\ + break;\ + }\ + }\ + else\ + {\ + spatial_pred = ( cur[-refs +j] + cur[+refs -j] ) >>1;\ + }\ + + if( x >= 2 && x <= w - 3 ) + { + YADIF_CHECK(-1) + if( x >= 3 && x <= w - 4 ) + { + YADIF_CHECK(-2) }} }} + } + } + if( x >= 2 && x <= w - 3 ) + { + YADIF_CHECK(1) + if( x >= 3 && x <= w - 4 ) + { + YADIF_CHECK(2) }} }} + } + } } -/* EDDI: Edge Directed Deinterlacing Interpolation - Uses the Martinez-Lim Line Shift Parametric Modeling algorithm...I think. - Checks 4 different slopes to see if there is more similarity along a diagonal - than there was vertically. If a diagonal is more similar, then it indicates - an edge, so interpolate along that instead of a vertical line, using either - linear or cubic interpolation depending on mode. */ -#define YADIF_CHECK(j)\ - { int score = ABS(cur[-refs-1+j] - cur[+refs-1-j])\ - + ABS(cur[-refs +j] - cur[+refs -j])\ - + ABS(cur[-refs+1+j] - cur[+refs+1-j]);\ - if( score < spatial_score ){\ - spatial_score = score;\ - if( pv->mode > 0 && !edge )\ - {\ - switch(j)\ - {\ - case -1:\ - spatial_pred = cubic_interpolate(cur[-3 * refs - 3], cur[-refs -1], cur[+refs + 1], cur[3* refs + 3] );\ - break;\ - case -2:\ - spatial_pred = cubic_interpolate( ( ( cur[-3*refs - 4] + cur[-refs - 4] ) / 2 ) , cur[-refs -2], cur[+refs + 2], ( ( cur[3*refs + 4] + cur[refs + 4] ) / 2 ) );\ - break;\ - case 1:\ - spatial_pred = cubic_interpolate(cur[-3 * refs +3], cur[-refs +1], cur[+refs - 1], cur[3* refs -3] );\ - break;\ - case 2:\ - spatial_pred = cubic_interpolate(( ( cur[-3*refs + 4] + cur[-refs + 4] ) / 2 ), cur[-refs +2], cur[+refs - 2], ( ( cur[3*refs - 4] + cur[refs - 4] ) / 2 ) );\ - break;\ - }\ - }\ - else\ - {\ - spatial_pred = ( cur[-refs +j] + cur[+refs -j] ) >>1;\ - }\ - - YADIF_CHECK(-1) YADIF_CHECK(-2) }} }} - YADIF_CHECK( 1) YADIF_CHECK( 2) }} }} - /* Temporally adjust the spatial prediction by comparing against lines in the adjacent fields. */ int b = (prev2[-2*refs] + next2[-2*refs])>>1; @@ -738,11 +1153,6 @@ static void yadif_filter_line( uint8_t *dst, } } -typedef struct yadif_thread_arg_s { - hb_filter_private_t *pv; - int segment; -} yadif_thread_arg_t; - /* * deinterlace this segment of all three planes in a single thread. */ @@ -817,17 +1227,26 @@ void yadif_decomb_filter_thread( void *thread_args_v ) for( y = segment_start; y < segment_stop; y++ ) { - if( ( pv->mode == 4 && is_combed ) || is_combed == 2 ) + if( is_combed == 2 ) { /* This line gets blend filtered, not yadif filtered. */ - uint8_t *prev = &pv->ref[0][plane][y*ref_stride]; uint8_t *cur = &pv->ref[1][plane][y*ref_stride]; - uint8_t *next = &pv->ref[2][plane][y*ref_stride]; uint8_t *dst2 = &dst[plane][y*w]; + /* These will be useful if we ever do temporal blending. */ + // uint8_t *prev = &pv->ref[0][plane][y*ref_stride]; + // uint8_t *next = &pv->ref[2][plane][y*ref_stride]; blend_filter_line( dst2, cur, plane, y, pv ); } - else if( ( ( y ^ parity ) & 1 ) && ( is_combed == 1 ) ) + else if( pv->mode == MODE_CUBIC && is_combed && ( ( y ^ parity ) & 1 ) ) + { + /* Just apply vertical cubic interpolation */ + uint8_t *cur = &pv->ref[1][plane][y*ref_stride]; + uint8_t *dst2 = &dst[plane][y*w]; + + cubic_interpolate_line( dst2, cur, plane, y, pv ); + } + else if( pv->mode & MODE_YADIF && ( ( y ^ parity ) & 1 ) && ( is_combed == 1 ) ) { /* This line gets yadif filtered. It is the bottom field when TFF and vice-versa. It's the field that gets @@ -902,65 +1321,129 @@ static void yadif_filter( uint8_t ** dst, int tff, hb_filter_private_t * pv ) { + /* If we're running comb detection, do it now, otherwise default to true. */ + int is_combed = pv->spatial_metric >= 0 ? comb_segmenter( pv ) : 1; - int is_combed = comb_segmenter( pv ); + /* The comb detector suggests three different values: + 0: Don't comb this frame. + 1: Deinterlace this frame. + 2: Blend this frame. + Since that might conflict with the filter's mode, + it may be necesary to adjust this value. */ + if( is_combed == 1 && (pv->mode == MODE_BLEND) ) + { + /* All combed frames are getting blended */ + is_combed = 2; + } + else if( is_combed == 2 && !( pv->mode & MODE_BLEND ) ) + { + /* Blending is disabled, so force interpolation of these frames. */ + is_combed = 1; + } + if( is_combed == 1 && + ( pv->mode & MODE_BLEND ) && + !( pv->mode & ( MODE_YADIF | MODE_EEDI2 | MODE_CUBIC ) ) ) + { + /* Deinterlacers are disabled, blending isn't, so blend these frames. */ + is_combed = 2; + } + else if( is_combed && + !( pv->mode & ( MODE_BLEND | MODE_YADIF | MODE_EEDI2 | MODE_CUBIC | MODE_MASK ) ) ) + { + /* No deinterlacer or mask chosen, pass the frame through. */ + is_combed = 0; + } if( is_combed == 1 ) { - pv->yadif_deinterlaced_frames++; + pv->deinterlaced_frames++; } else if( is_combed == 2 ) { - pv->blend_deinterlaced_frames++; + pv->blended_frames++; } else { pv->unfiltered_frames++; } + if( is_combed == 1 && ( pv->mode & MODE_EEDI2 ) ) + { + /* Generate an EEDI2 interpolation */ + eedi2_planer( pv ); + } + if( is_combed ) { - int segment; + if( ( pv->mode & MODE_EEDI2 ) && !( pv->mode & MODE_YADIF ) && is_combed == 1 ) + { + // Just pass through the EEDI2 interpolation + int i; + for( i = 0; i < 3; i++ ) + { + uint8_t * ref = pv->eedi_full[DST2PF][i]; + uint8_t * dest = dst[i]; + + int w = pv->width[i]; + int ref_stride = pv->ref_stride[i]; + + int y; + for( y = 0; y < pv->height[i]; y++ ) + { + memcpy(dest, ref, w); + dest += w; + ref += ref_stride; + } + } + } + else + { + int segment; + + for( segment = 0; segment < pv->cpu_count; segment++ ) + { + /* + * Setup the work for this plane. + */ + pv->yadif_arguments[segment].parity = parity; + pv->yadif_arguments[segment].tff = tff; + pv->yadif_arguments[segment].dst = dst; + pv->yadif_arguments[segment].is_combed = is_combed; + + /* + * Let the thread for this plane know that we've setup work + * for it by releasing the begin lock (ensuring that the + * complete lock is already locked so that we block when + * we try to lock it again below). + */ + hb_lock( pv->yadif_complete_lock[segment] ); + hb_unlock( pv->yadif_begin_lock[segment] ); + } - for( segment = 0; segment < pv->cpu_count; segment++ ) - { /* - * Setup the work for this plane. + * Wait until all three threads have completed by trying to get + * the complete lock that we locked earlier for each thread, which + * will block until that thread has completed the work on that + * plane. */ - pv->yadif_arguments[segment].parity = parity; - pv->yadif_arguments[segment].tff = tff; - pv->yadif_arguments[segment].dst = dst; - pv->yadif_arguments[segment].is_combed = is_combed; + for( segment = 0; segment < pv->cpu_count; segment++ ) + { + hb_lock( pv->yadif_complete_lock[segment] ); + hb_unlock( pv->yadif_complete_lock[segment] ); + } /* - * Let the thread for this plane know that we've setup work - * for it by releasing the begin lock (ensuring that the - * complete lock is already locked so that we block when - * we try to lock it again below). + * Entire frame is now deinterlaced. */ - hb_lock( pv->yadif_complete_lock[segment] ); - hb_unlock( pv->yadif_begin_lock[segment] ); - } - - /* - * Wait until all three threads have completed by trying to get - * the complete lock that we locked earlier for each thread, which - * will block until that thread has completed the work on that - * plane. - */ - for( segment = 0; segment < pv->cpu_count; segment++ ) - { - hb_lock( pv->yadif_complete_lock[segment] ); - hb_unlock( pv->yadif_complete_lock[segment] ); } - - /* - * Entire frame is now deinterlaced. - */ } else { /* Just passing through... */ + + /* For mcdeint's benefit... */ + pv->yadif_arguments[0].is_combed = is_combed; // 0 + int i; for( i = 0; i < 3; i++ ) { @@ -979,6 +1462,12 @@ static void yadif_filter( uint8_t ** dst, } } } + + if( pv->mode & MODE_MASK && pv->spatial_metric >= 0 ) + { + if( pv->mode == MODE_MASK || is_combed ) + apply_mask( pv ); + } } static void mcdeint_filter( uint8_t ** dst, @@ -1024,7 +1513,7 @@ static void mcdeint_filter( uint8_t ** dst, { for( x = 0; x < w; x++ ) { - if( (x-2)+(y-1)*w >= 0 && (x+2)+(y+1)*w < w*h ) + if( (x-1)+(y-1)*w >= 0 && (x+1)+(y+1)*w < w*h ) { uint8_t * filp = &pv->mcdeint_frame_dec->data[i][x + y*fils]; @@ -1032,11 +1521,12 @@ static void mcdeint_filter( uint8_t ** dst, int diff0 = filp[-fils] - srcp[-srcs]; int diff1 = filp[+fils] - srcp[+srcs]; - - int spatial_score = - ABS(srcp[-srcs-1] - srcp[+srcs-1]) - + ABS(srcp[-srcs ] - srcp[+srcs ]) - + ABS(srcp[-srcs+1] - srcp[+srcs+1]) - 1; + int spatial_score; + + spatial_score = + ABS(srcp[-srcs-1] - srcp[+srcs-1]) + + ABS(srcp[-srcs ] - srcp[+srcs ]) + + ABS(srcp[-srcs+1] - srcp[+srcs+1]) - 1; int temp = filp[0]; @@ -1049,8 +1539,22 @@ static void mcdeint_filter( uint8_t ** dst, diff0 = filp[-fils+j] - srcp[-srcs+j];\ diff1 = filp[+fils-j] - srcp[+srcs-j]; - MCDEINT_CHECK(-1) MCDEINT_CHECK(-2) }} }} - MCDEINT_CHECK( 1) MCDEINT_CHECK( 2) }} }} + if( x >= 2 && x <= w - 3 ) + { + MCDEINT_CHECK(-1) + if( x >= 3 && x <= w - 4 ) + { + MCDEINT_CHECK(-2) }} }} + } + } + if( x >= 2 && x <= w - 3 ) + { + MCDEINT_CHECK(1) + if( x >= 3 && x <= w - 4 ) + { + MCDEINT_CHECK(2) }} }} + } + } if(diff0 + diff1 > 0) { @@ -1073,11 +1577,7 @@ static void mcdeint_filter( uint8_t ** dst, } } } - } - - for( y = 0; y < h; y++ ) - { - if( !((y ^ parity) & 1) ) + else { for( x = 0; x < w; x++ ) { @@ -1117,13 +1617,13 @@ hb_filter_private_t * hb_decomb_init( int pix_fmt, pv->buf_out[1] = hb_video_buffer_init( width, height ); pv->buf_settings = hb_buffer_init( 0 ); - pv->yadif_deinterlaced_frames = 0; - pv->blend_deinterlaced_frames = 0; + pv->deinterlaced_frames = 0; + pv->blended_frames = 0; pv->unfiltered_frames = 0; pv->yadif_ready = 0; - pv->mode = MODE_DEFAULT; + pv->mode = MODE_YADIF | MODE_BLEND | MODE_CUBIC; pv->spatial_metric = 2; pv->motion_threshold = 6; pv->spatial_threshold = 9; @@ -1131,6 +1631,15 @@ hb_filter_private_t * hb_decomb_init( int pix_fmt, pv->block_width = 16; pv->block_height = 16; + pv->magnitude_threshold = 10; + pv->variance_threshold = 20; + pv->laplacian_threshold = 20; + pv->dilation_threshold = 4; + pv->erosion_threshold = 2; + pv->noise_threshold = 50; + pv->maximum_search_distance = 24; + pv->post_processing = 1; + pv->parity = PARITY_DEFAULT; pv->mcdeint_mode = MCDEINT_MODE_DEFAULT; @@ -1138,22 +1647,31 @@ hb_filter_private_t * hb_decomb_init( int pix_fmt, if( settings ) { - sscanf( settings, "%d:%d:%d:%d:%d:%d:%d", + sscanf( settings, "%d:%d:%d:%d:%d:%d:%d:%d:%d:%d:%d:%d:%d:%d:%d:%d", &pv->mode, &pv->spatial_metric, &pv->motion_threshold, &pv->spatial_threshold, &pv->block_threshold, &pv->block_width, - &pv->block_height ); + &pv->block_height, + &pv->magnitude_threshold, + &pv->variance_threshold, + &pv->laplacian_threshold, + &pv->dilation_threshold, + &pv->erosion_threshold, + &pv->noise_threshold, + &pv->maximum_search_distance, + &pv->post_processing, + &pv->parity ); } pv->cpu_count = hb_get_cpu_count(); - if( pv->mode == 2 || pv->mode == 3 ) + if( pv->mode & MODE_MCDEINT ) { - pv->mcdeint_mode = 0; + pv->mcdeint_mode = 2; } /* Allocate yadif specific buffers */ @@ -1168,7 +1686,7 @@ hb_filter_private_t * hb_decomb_init( int pix_fmt, for( j = 0; j < 3; j++ ) { - pv->ref[j][i] = malloc( w*h*sizeof(uint8_t) ) + 3*w; + pv->ref[j][i] = calloc( 1, w*h*sizeof(uint8_t) ) + 3*w; } } @@ -1181,7 +1699,38 @@ hb_filter_private_t * hb_decomb_init( int pix_fmt, pv->mask[i] = calloc( 1, w*h*sizeof(uint8_t) ) + 3*w; } + + if( pv->mode & MODE_EEDI2 ) + { + /* Allocate half-height eedi2 buffers */ + height = pv->height[0] / 2; + for( i = 0; i < 3; i++ ) + { + int is_chroma = !!i; + int w = ((width + 31) & (~31))>>is_chroma; + int h = ((height+6+ 31) & (~31))>>is_chroma; + for( j = 0; j < 4; j++ ) + { + pv->eedi_half[j][i] = calloc( 1, w*h*sizeof(uint8_t) ) + 3*w; + } + } + + /* Allocate full-height eedi2 buffers */ + height = pv->height[0]; + for( i = 0; i < 3; i++ ) + { + int is_chroma = !!i; + int w = ((width + 31) & (~31))>>is_chroma; + int h = ((height+6+ 31) & (~31))>>is_chroma; + + for( j = 0; j < 5; j++ ) + { + pv->eedi_full[j][i] = calloc( 1, w*h*sizeof(uint8_t) ) + 3*w; + } + } + } + /* * Create yadif threads and locks. */ @@ -1264,7 +1813,62 @@ hb_filter_private_t * hb_decomb_init( int pix_fmt, hb_error( "decomb could not create threads" ); } } + + if( pv->mode & MODE_EEDI2 ) + { + /* + * Create eedi2 threads and locks. + */ + pv->eedi2_threads = malloc( sizeof( hb_thread_t* ) * 3 ); + pv->eedi2_begin_lock = malloc( sizeof( hb_lock_t * ) * 3 ); + pv->eedi2_complete_lock = malloc( sizeof( hb_lock_t * ) * 3 ); + pv->eedi2_arguments = malloc( sizeof( eedi2_arguments_t ) * 3 ); + if( pv->post_processing > 1 ) + { + pv->cx2 = (int*)eedi2_aligned_malloc(pv->height[0]*pv->ref_stride[0]*sizeof(int), 16); + pv->cy2 = (int*)eedi2_aligned_malloc(pv->height[0]*pv->ref_stride[0]*sizeof(int), 16); + pv->cxy = (int*)eedi2_aligned_malloc(pv->height[0]*pv->ref_stride[0]*sizeof(int), 16); + pv->tmpc = (int*)eedi2_aligned_malloc(pv->height[0]*pv->ref_stride[0]*sizeof(int), 16); + if( !pv->cx2 || !pv->cy2 || !pv->cxy || !pv->tmpc ) + hb_log("EEDI2: failed to malloc derivative arrays"); + else + hb_log("EEDI2: successfully mallloced derivative arrays"); + } + + for( i = 0; i < 3; i++ ) + { + eedi2_thread_arg_t *eedi2_thread_args; + + eedi2_thread_args = malloc( sizeof( eedi2_thread_arg_t ) ); + + if( eedi2_thread_args ) + { + eedi2_thread_args->pv = pv; + eedi2_thread_args->plane = i; + + pv->eedi2_begin_lock[i] = hb_lock_init(); + pv->eedi2_complete_lock[i] = hb_lock_init(); + + /* + * Important to start off with the threads locked waiting + * on input. + */ + hb_lock( pv->eedi2_begin_lock[i] ); + + pv->eedi2_arguments[i].stop = 0; + + pv->eedi2_threads[i] = hb_thread_init( "eedi2_filter_segment", + eedi2_filter_thread, + eedi2_thread_args, + HB_NORMAL_PRIORITY ); + } + else + { + hb_error( "eedi2 could not create threads" ); + } + } + } /* Allocate mcdeint specific buffers */ @@ -1272,9 +1876,7 @@ hb_filter_private_t * hb_decomb_init( int pix_fmt, { avcodec_init(); avcodec_register_all(); - AVCodec * enc = avcodec_find_encoder( CODEC_ID_SNOW ); - int i; for (i = 0; i < 3; i++ ) { @@ -1301,7 +1903,7 @@ hb_filter_private_t * hb_decomb_init( int pix_fmt, case 3: avctx_enc->refs = 3; case 2: - avctx_enc->me_method = ME_UMH; + avctx_enc->me_method = ME_ITER; case 1: avctx_enc->flags |= CODEC_FLAG_4MV; avctx_enc->dia_size =2; @@ -1309,7 +1911,7 @@ hb_filter_private_t * hb_decomb_init( int pix_fmt, avctx_enc->flags |= CODEC_FLAG_QPEL; } - avcodec_open(avctx_enc, enc); + hb_avcodec_open(avctx_enc, enc); } pv->mcdeint_frame = avcodec_alloc_frame(); @@ -1327,7 +1929,7 @@ void hb_decomb_close( hb_filter_private_t * pv ) return; } - hb_log("decomb: yadif deinterlaced %i | blend deinterlaced %i | unfiltered %i | total %i", pv->yadif_deinterlaced_frames, pv->blend_deinterlaced_frames, pv->unfiltered_frames, pv->yadif_deinterlaced_frames + pv->blend_deinterlaced_frames + pv->unfiltered_frames); + hb_log("decomb: deinterlaced %i | blended %i | unfiltered %i | total %i", pv->deinterlaced_frames, pv->blended_frames, pv->unfiltered_frames, pv->deinterlaced_frames + pv->blended_frames + pv->unfiltered_frames); /* Cleanup frame buffers */ if( pv->buf_out[0] ) @@ -1366,6 +1968,46 @@ void hb_decomb_close( hb_filter_private_t * pv ) } } + if( pv->mode & MODE_EEDI2 ) + { + /* Cleanup eedi-half buffers */ + int j; + for( i = 0; i<3; i++ ) + { + for( j = 0; j < 4; j++ ) + { + uint8_t **p = &pv->eedi_half[j][i]; + if (*p) + { + free( *p - 3*pv->ref_stride[i] ); + *p = NULL; + } + } + } + + /* Cleanup eedi-full buffers */ + for( i = 0; i<3; i++ ) + { + for( j = 0; j < 5; j++ ) + { + uint8_t **p = &pv->eedi_full[j][i]; + if (*p) + { + free( *p - 3*pv->ref_stride[i] ); + *p = NULL; + } + } + } + } + + if( pv->post_processing > 1 && ( pv->mode & MODE_EEDI2 ) ) + { + if (pv->cx2) eedi2_aligned_free(pv->cx2); + if (pv->cy2) eedi2_aligned_free(pv->cy2); + if (pv->cxy) eedi2_aligned_free(pv->cxy); + if (pv->tmpc) eedi2_aligned_free(pv->tmpc); + } + for( i = 0; i < pv->cpu_count; i++) { /* @@ -1408,12 +2050,36 @@ void hb_decomb_close( hb_filter_private_t * pv ) free( pv->decomb_complete_lock ); free( pv->decomb_arguments ); + if( pv->mode & MODE_EEDI2 ) + { + for( i = 0; i < 3; i++) + { + /* + * Tell each eedi2 thread to stop, and then cleanup. + */ + pv->eedi2_arguments[i].stop = 1; + hb_unlock( pv->eedi2_begin_lock[i] ); + + hb_thread_close( &pv->eedi2_threads[i] ); + hb_lock_close( &pv->eedi2_begin_lock[i] ); + hb_lock_close( &pv->eedi2_complete_lock[i] ); + } + + /* + * free memory for eedi2 structs + */ + free( pv->eedi2_threads ); + free( pv->eedi2_begin_lock ); + free( pv->eedi2_complete_lock ); + free( pv->eedi2_arguments ); + } + /* Cleanup mcdeint specific buffers */ if( pv->mcdeint_mode >= 0 ) { if( pv->mcdeint_avctx_enc ) { - avcodec_close( pv->mcdeint_avctx_enc ); + hb_avcodec_close( pv->mcdeint_avctx_enc ); av_freep( &pv->mcdeint_avctx_enc ); } if( pv->mcdeint_outbuf ) @@ -1476,16 +2142,48 @@ int hb_decomb_work( const hb_buffer_t * cbuf_in, /* Perform yadif filtering */ int frame; - for( frame = 0; frame <= ( ( pv->mode == 2 || pv->mode == 3 )? 1 : 0 ) ; frame++ ) + for( frame = 0; frame <= ( ( pv->mode & MODE_MCDEINT ) ? 1 : 0 ) ; frame++ ) +// This would be what to use for bobbing: for( frame = 0; frame <= 0 ; frame++ ) { + +#if 0 + /* Perhaps skip the second run if the frame is uncombed? */ + if( frame && !pv->yadif_arguments[0].is_combed ) + { + break; + } +#endif int parity = frame ^ tff ^ 1; +// This will be for bobbing +#if 0 + if( pv->alternator ) + { + parity = !parity; + pv->alternator = 0; + } + else + { + pv->alternator = 1; + } +#endif + pv->tff = !parity; + avpicture_fill( &pv->pic_out, pv->buf_out[!(frame^1)]->data, pix_fmt, width, height ); + /* XXX + Should check here and only bother filtering field 2 when + field 1 was detected as combed. + And when it's not, it's a progressive frame, + so mcdeint should be skipped... + */ yadif_filter( pv->pic_out.data, parity, tff, pv ); - if( pv->mcdeint_mode >= 0 ) + /* Commented out code in the line below would skip mcdeint + on uncombed frames. Possibly a bad idea, since mcdeint + maintains the same snow context for the entire video... */ + if( pv->mcdeint_mode >= 0 /* && pv->yadif_arguments[0].is_combed */) { /* Perform mcdeint filtering */ avpicture_fill( &pv->pic_in, pv->buf_out[(frame^1)]->data,