avfilter/vf_idet: Fixing idet for single-frame inputs.
[ffmpeg.git] / libavfilter / vf_idet.c
1 /*
2  * Copyright (C) 2012 Michael Niedermayer <michaelni@gmx.at>
3  *
4  * This file is part of FFmpeg.
5  *
6  * FFmpeg is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * FFmpeg is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with FFmpeg; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20
21 #include <float.h> /* FLT_MAX */
22
23 #include "libavutil/cpu.h"
24 #include "libavutil/common.h"
25 #include "libavutil/opt.h"
26 #include "internal.h"
27 #include "vf_idet.h"
28
29 #define OFFSET(x) offsetof(IDETContext, x)
30 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
31
32 static const AVOption idet_options[] = {
33     { "intl_thres", "set interlacing threshold", OFFSET(interlace_threshold),   AV_OPT_TYPE_FLOAT, {.dbl = 1.04}, -1, FLT_MAX, FLAGS },
34     { "prog_thres", "set progressive threshold", OFFSET(progressive_threshold), AV_OPT_TYPE_FLOAT, {.dbl = 1.5},  -1, FLT_MAX, FLAGS },
35     { "rep_thres",  "set repeat threshold",      OFFSET(repeat_threshold),      AV_OPT_TYPE_FLOAT, {.dbl = 3.0},  -1, FLT_MAX, FLAGS },
36     { "half_life", "half life of cumulative statistics", OFFSET(half_life),     AV_OPT_TYPE_FLOAT, {.dbl = 0.0},  -1, INT_MAX, FLAGS },
37     { NULL }
38 };
39
40 AVFILTER_DEFINE_CLASS(idet);
41
42 static const char *type2str(Type type)
43 {
44     switch(type) {
45         case TFF          : return "tff";
46         case BFF          : return "bff";
47         case PROGRESSIVE  : return "progressive";
48         case UNDETERMINED : return "undetermined";
49     }
50     return NULL;
51 }
52
53 #define PRECISION 1048576
54
55 static uint64_t uintpow(uint64_t b,unsigned int e)
56 {
57     uint64_t r=1;
58     while(e--) r*=b;
59     return r;
60 }
61
62 static int av_dict_set_fxp(AVDictionary **pm, const char *key, uint64_t value, unsigned int digits,
63                 int flags)
64 {
65     char valuestr[44];
66     uint64_t print_precision = uintpow(10, digits);
67
68     value = av_rescale(value, print_precision, PRECISION);
69
70     snprintf(valuestr, sizeof(valuestr), "%"PRId64".%0*"PRId64,
71              value / print_precision, digits, value % print_precision);
72
73     return av_dict_set(pm, key, valuestr, flags);
74 }
75
76 static const char *rep2str(RepeatedField repeated_field)
77 {
78     switch(repeated_field) {
79         case REPEAT_NONE    : return "neither";
80         case REPEAT_TOP     : return "top";
81         case REPEAT_BOTTOM  : return "bottom";
82     }
83     return NULL;
84 }
85
86 int ff_idet_filter_line_c(const uint8_t *a, const uint8_t *b, const uint8_t *c, int w)
87 {
88     int x;
89     int ret=0;
90
91     for(x=0; x<w; x++){
92         int v = (*a++ + *c++) - 2 * *b++;
93         ret += FFABS(v);
94     }
95
96     return ret;
97 }
98
99 int ff_idet_filter_line_c_16bit(const uint16_t *a, const uint16_t *b, const uint16_t *c, int w)
100 {
101     int x;
102     int ret=0;
103
104     for(x=0; x<w; x++){
105         int v = (*a++ + *c++) - 2 * *b++;
106         ret += FFABS(v);
107     }
108
109     return ret;
110 }
111
112 static void filter(AVFilterContext *ctx)
113 {
114     IDETContext *idet = ctx->priv;
115     int y, i;
116     int64_t alpha[2]={0};
117     int64_t delta=0;
118     int64_t gamma[2]={0};
119     Type type, best_type;
120     RepeatedField repeat;
121     int match = 0;
122     AVDictionary **metadata = avpriv_frame_get_metadatap(idet->cur);
123
124     for (i = 0; i < idet->csp->nb_components; i++) {
125         int w = idet->cur->width;
126         int h = idet->cur->height;
127         int refs = idet->cur->linesize[i];
128
129         if (i && i<3) {
130             w = FF_CEIL_RSHIFT(w, idet->csp->log2_chroma_w);
131             h = FF_CEIL_RSHIFT(h, idet->csp->log2_chroma_h);
132         }
133
134         for (y = 2; y < h - 2; y++) {
135             uint8_t *prev = &idet->prev->data[i][y*refs];
136             uint8_t *cur  = &idet->cur ->data[i][y*refs];
137             uint8_t *next = &idet->next->data[i][y*refs];
138             alpha[ y   &1] += idet->filter_line(cur-refs, prev, cur+refs, w);
139             alpha[(y^1)&1] += idet->filter_line(cur-refs, next, cur+refs, w);
140             delta          += idet->filter_line(cur-refs,  cur, cur+refs, w);
141             gamma[(y^1)&1] += idet->filter_line(cur     , prev, cur     , w);
142         }
143     }
144
145     if      (alpha[0] > idet->interlace_threshold * alpha[1]){
146         type = TFF;
147     }else if(alpha[1] > idet->interlace_threshold * alpha[0]){
148         type = BFF;
149     }else if(alpha[1] > idet->progressive_threshold * delta){
150         type = PROGRESSIVE;
151     }else{
152         type = UNDETERMINED;
153     }
154
155     if ( gamma[0] > idet->repeat_threshold * gamma[1] ){
156         repeat = REPEAT_TOP;
157     } else if ( gamma[1] > idet->repeat_threshold * gamma[0] ){
158         repeat = REPEAT_BOTTOM;
159     } else {
160         repeat = REPEAT_NONE;
161     }
162
163     memmove(idet->history+1, idet->history, HIST_SIZE-1);
164     idet->history[0] = type;
165     best_type = UNDETERMINED;
166     for(i=0; i<HIST_SIZE; i++){
167         if(idet->history[i] != UNDETERMINED){
168             if(best_type == UNDETERMINED)
169                 best_type = idet->history[i];
170
171             if(idet->history[i] == best_type) {
172                 match++;
173             }else{
174                 match=0;
175                 break;
176             }
177         }
178     }
179     if(idet->last_type == UNDETERMINED){
180         if(match  ) idet->last_type = best_type;
181     }else{
182         if(match>2) idet->last_type = best_type;
183     }
184
185     if      (idet->last_type == TFF){
186         idet->cur->top_field_first = 1;
187         idet->cur->interlaced_frame = 1;
188     }else if(idet->last_type == BFF){
189         idet->cur->top_field_first = 0;
190         idet->cur->interlaced_frame = 1;
191     }else if(idet->last_type == PROGRESSIVE){
192         idet->cur->interlaced_frame = 0;
193     }
194
195     for(i=0; i<3; i++)
196         idet->repeats[i]  = av_rescale(idet->repeats [i], idet->decay_coefficient, PRECISION);
197
198     for(i=0; i<4; i++){
199         idet->prestat [i] = av_rescale(idet->prestat [i], idet->decay_coefficient, PRECISION);
200         idet->poststat[i] = av_rescale(idet->poststat[i], idet->decay_coefficient, PRECISION);
201     }
202
203     idet->total_repeats [         repeat] ++;
204     idet->repeats       [         repeat] += PRECISION;
205
206     idet->total_prestat [           type] ++;
207     idet->prestat       [           type] += PRECISION;
208
209     idet->total_poststat[idet->last_type] ++;
210     idet->poststat      [idet->last_type] += PRECISION;
211
212     av_log(ctx, AV_LOG_DEBUG, "Repeated Field:%12s, Single frame:%12s, Multi frame:%12s\n",
213            rep2str(repeat), type2str(type), type2str(idet->last_type));
214
215     av_dict_set    (metadata, "lavfi.idet.repeated.current_frame", rep2str(repeat), 0);
216     av_dict_set_fxp(metadata, "lavfi.idet.repeated.neither",       idet->repeats[REPEAT_NONE], 2, 0);
217     av_dict_set_fxp(metadata, "lavfi.idet.repeated.top",           idet->repeats[REPEAT_TOP], 2, 0);
218     av_dict_set_fxp(metadata, "lavfi.idet.repeated.bottom",        idet->repeats[REPEAT_BOTTOM], 2, 0);
219
220     av_dict_set    (metadata, "lavfi.idet.single.current_frame",   type2str(type), 0);
221     av_dict_set_fxp(metadata, "lavfi.idet.single.tff",             idet->prestat[TFF], 2 , 0);
222     av_dict_set_fxp(metadata, "lavfi.idet.single.bff",             idet->prestat[BFF], 2, 0);
223     av_dict_set_fxp(metadata, "lavfi.idet.single.progressive",     idet->prestat[PROGRESSIVE], 2, 0);
224     av_dict_set_fxp(metadata, "lavfi.idet.single.undetermined",    idet->prestat[UNDETERMINED], 2, 0);
225
226     av_dict_set    (metadata, "lavfi.idet.multiple.current_frame", type2str(idet->last_type), 0);
227     av_dict_set_fxp(metadata, "lavfi.idet.multiple.tff",           idet->poststat[TFF], 2, 0);
228     av_dict_set_fxp(metadata, "lavfi.idet.multiple.bff",           idet->poststat[BFF], 2, 0);
229     av_dict_set_fxp(metadata, "lavfi.idet.multiple.progressive",   idet->poststat[PROGRESSIVE], 2, 0);
230     av_dict_set_fxp(metadata, "lavfi.idet.multiple.undetermined",  idet->poststat[UNDETERMINED], 2, 0);
231 }
232
233 static int filter_frame(AVFilterLink *link, AVFrame *picref)
234 {
235     AVFilterContext *ctx = link->dst;
236     IDETContext *idet = ctx->priv;
237
238     if (idet->prev)
239         av_frame_free(&idet->prev);
240     idet->prev = idet->cur;
241     idet->cur  = idet->next;
242     idet->next = picref;
243
244     if (!idet->cur &&
245         !(idet->cur = av_frame_clone(idet->next)))
246         return AVERROR(ENOMEM);
247
248     if (!idet->prev)
249         return 0;
250
251     if (!idet->csp)
252         idet->csp = av_pix_fmt_desc_get(link->format);
253     if (idet->csp->comp[0].depth_minus1 / 8 == 1){
254         idet->filter_line = (ff_idet_filter_func)ff_idet_filter_line_c_16bit;
255         if (ARCH_X86)
256             ff_idet_init_x86(idet, 1);
257     }
258
259     filter(ctx);
260
261     return ff_filter_frame(ctx->outputs[0], av_frame_clone(idet->cur));
262 }
263
264 static int request_frame(AVFilterLink *link)
265 {
266     AVFilterContext *ctx = link->src;
267     IDETContext *idet = ctx->priv;
268
269     do {
270         int ret;
271
272         if (idet->eof)
273             return AVERROR_EOF;
274
275         ret = ff_request_frame(link->src->inputs[0]);
276
277         if (ret == AVERROR_EOF && idet->cur) {
278             AVFrame *next = av_frame_clone(idet->next);
279
280             if (!next)
281                 return AVERROR(ENOMEM);
282
283             filter_frame(link->src->inputs[0], next);
284             idet->eof = 1;
285         } else if (ret < 0) {
286             return ret;
287         }
288     } while (!idet->prev);
289
290     return 0;
291 }
292
293 static av_cold void uninit(AVFilterContext *ctx)
294 {
295     IDETContext *idet = ctx->priv;
296
297     av_log(ctx, AV_LOG_INFO, "Repeated Fields: Neither:%6"PRId64" Top:%6"PRId64" Bottom:%6"PRId64"\n",
298            idet->total_repeats[REPEAT_NONE],
299            idet->total_repeats[REPEAT_TOP],
300            idet->total_repeats[REPEAT_BOTTOM]
301         );
302     av_log(ctx, AV_LOG_INFO, "Single frame detection: TFF:%6"PRId64" BFF:%6"PRId64" Progressive:%6"PRId64" Undetermined:%6"PRId64"\n",
303            idet->total_prestat[TFF],
304            idet->total_prestat[BFF],
305            idet->total_prestat[PROGRESSIVE],
306            idet->total_prestat[UNDETERMINED]
307         );
308     av_log(ctx, AV_LOG_INFO, "Multi frame detection: TFF:%6"PRId64" BFF:%6"PRId64" Progressive:%6"PRId64" Undetermined:%6"PRId64"\n",
309            idet->total_poststat[TFF],
310            idet->total_poststat[BFF],
311            idet->total_poststat[PROGRESSIVE],
312            idet->total_poststat[UNDETERMINED]
313         );
314
315     av_frame_free(&idet->prev);
316     av_frame_free(&idet->cur );
317     av_frame_free(&idet->next);
318 }
319
320 static int query_formats(AVFilterContext *ctx)
321 {
322     static const enum AVPixelFormat pix_fmts[] = {
323         AV_PIX_FMT_YUV420P,
324         AV_PIX_FMT_YUV422P,
325         AV_PIX_FMT_YUV444P,
326         AV_PIX_FMT_YUV410P,
327         AV_PIX_FMT_YUV411P,
328         AV_PIX_FMT_GRAY8,
329         AV_PIX_FMT_YUVJ420P,
330         AV_PIX_FMT_YUVJ422P,
331         AV_PIX_FMT_YUVJ444P,
332         AV_PIX_FMT_GRAY16,
333         AV_PIX_FMT_YUV440P,
334         AV_PIX_FMT_YUVJ440P,
335         AV_PIX_FMT_YUV420P10,
336         AV_PIX_FMT_YUV422P10,
337         AV_PIX_FMT_YUV444P10,
338         AV_PIX_FMT_YUV420P16,
339         AV_PIX_FMT_YUV422P16,
340         AV_PIX_FMT_YUV444P16,
341         AV_PIX_FMT_YUVA420P,
342         AV_PIX_FMT_NONE
343     };
344
345     ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
346
347     return 0;
348 }
349
350 static int config_output(AVFilterLink *outlink)
351 {
352     outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
353     return 0;
354 }
355
356 static av_cold int init(AVFilterContext *ctx)
357 {
358     IDETContext *idet = ctx->priv;
359
360     idet->eof = 0;
361     idet->last_type = UNDETERMINED;
362     memset(idet->history, UNDETERMINED, HIST_SIZE);
363
364     if( idet->half_life > 0 )
365         idet->decay_coefficient = (uint64_t) round( PRECISION * exp2(-1.0 / idet->half_life) );
366     else
367         idet->decay_coefficient = PRECISION;
368
369     idet->filter_line = ff_idet_filter_line_c;
370
371     if (ARCH_X86)
372         ff_idet_init_x86(idet, 0);
373
374     return 0;
375 }
376
377 static const AVFilterPad idet_inputs[] = {
378     {
379         .name         = "default",
380         .type         = AVMEDIA_TYPE_VIDEO,
381         .filter_frame = filter_frame,
382     },
383     { NULL }
384 };
385
386 static const AVFilterPad idet_outputs[] = {
387     {
388         .name         = "default",
389         .type         = AVMEDIA_TYPE_VIDEO,
390         .config_props = config_output,
391         .request_frame = request_frame
392     },
393     { NULL }
394 };
395
396 AVFilter ff_vf_idet = {
397     .name          = "idet",
398     .description   = NULL_IF_CONFIG_SMALL("Interlace detect Filter."),
399     .priv_size     = sizeof(IDETContext),
400     .init          = init,
401     .uninit        = uninit,
402     .query_formats = query_formats,
403     .inputs        = idet_inputs,
404     .outputs       = idet_outputs,
405     .priv_class    = &idet_class,
406 };