FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
vf_signalstats.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2010 Mark Heath mjpeg0 @ silicontrip dot org
3  * Copyright (c) 2014 Clément Bœsch
4  * Copyright (c) 2014 Dave Rice @dericed
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22 
23 #include "libavutil/opt.h"
24 #include "libavutil/pixdesc.h"
25 #include "internal.h"
26 
27 enum FilterMode {
33 };
34 
35 typedef struct {
36  const AVClass *class;
37  int chromah; // height of chroma plane
38  int chromaw; // width of chroma plane
39  int hsub; // horizontal subsampling
40  int vsub; // vertical subsampling
41  int fs; // pixel count per frame
42  int cfs; // pixel count per frame of chroma planes
44  int filters;
46  uint8_t rgba_color[4];
47  int yuv_color[3];
48  int nb_jobs;
49  int *jobs_rets;
50 
54 
55 typedef struct ThreadData {
56  const AVFrame *in;
57  AVFrame *out;
58 } ThreadData;
59 
60 typedef struct ThreadDataHueSatMetrics {
61  const AVFrame *src;
64 
65 #define OFFSET(x) offsetof(SignalstatsContext, x)
66 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
67 
68 static const AVOption signalstats_options[] = {
69  {"stat", "set statistics filters", OFFSET(filters), AV_OPT_TYPE_FLAGS, {.i64=0}, 0, INT_MAX, FLAGS, "filters"},
70  {"tout", "analyze pixels for temporal outliers", 0, AV_OPT_TYPE_CONST, {.i64=1<<FILTER_TOUT}, 0, 0, FLAGS, "filters"},
71  {"vrep", "analyze video lines for vertical line repetition", 0, AV_OPT_TYPE_CONST, {.i64=1<<FILTER_VREP}, 0, 0, FLAGS, "filters"},
72  {"brng", "analyze for pixels outside of broadcast range", 0, AV_OPT_TYPE_CONST, {.i64=1<<FILTER_BRNG}, 0, 0, FLAGS, "filters"},
73  {"out", "set video filter", OFFSET(outfilter), AV_OPT_TYPE_INT, {.i64=FILTER_NONE}, -1, FILT_NUMB-1, FLAGS, "out"},
74  {"tout", "highlight pixels that depict temporal outliers", 0, AV_OPT_TYPE_CONST, {.i64=FILTER_TOUT}, 0, 0, FLAGS, "out"},
75  {"vrep", "highlight video lines that depict vertical line repetition", 0, AV_OPT_TYPE_CONST, {.i64=FILTER_VREP}, 0, 0, FLAGS, "out"},
76  {"brng", "highlight pixels that are outside of broadcast range", 0, AV_OPT_TYPE_CONST, {.i64=FILTER_BRNG}, 0, 0, FLAGS, "out"},
77  {"c", "set highlight color", OFFSET(rgba_color), AV_OPT_TYPE_COLOR, {.str="yellow"}, .flags=FLAGS},
78  {"color", "set highlight color", OFFSET(rgba_color), AV_OPT_TYPE_COLOR, {.str="yellow"}, .flags=FLAGS},
79  {NULL}
80 };
81 
82 AVFILTER_DEFINE_CLASS(signalstats);
83 
84 static av_cold int init(AVFilterContext *ctx)
85 {
86  uint8_t r, g, b;
87  SignalstatsContext *s = ctx->priv;
88 
89  if (s->outfilter != FILTER_NONE)
90  s->filters |= 1 << s->outfilter;
91 
92  r = s->rgba_color[0];
93  g = s->rgba_color[1];
94  b = s->rgba_color[2];
95  s->yuv_color[0] = (( 66*r + 129*g + 25*b + (1<<7)) >> 8) + 16;
96  s->yuv_color[1] = ((-38*r + -74*g + 112*b + (1<<7)) >> 8) + 128;
97  s->yuv_color[2] = ((112*r + -94*g + -18*b + (1<<7)) >> 8) + 128;
98  return 0;
99 }
100 
101 static av_cold void uninit(AVFilterContext *ctx)
102 {
103  SignalstatsContext *s = ctx->priv;
107  av_freep(&s->jobs_rets);
108 }
109 
111 {
112  // TODO: add more
113  static const enum AVPixelFormat pix_fmts[] = {
119  };
120 
122  return 0;
123 }
124 
125 static AVFrame *alloc_frame(enum AVPixelFormat pixfmt, int w, int h)
126 {
128  if (!frame)
129  return NULL;
130 
131  frame->format = pixfmt;
132  frame->width = w;
133  frame->height = h;
134 
135  if (av_frame_get_buffer(frame, 32) < 0)
136  return NULL;
137 
138  return frame;
139 }
140 
141 static int config_props(AVFilterLink *outlink)
142 {
143  AVFilterContext *ctx = outlink->src;
144  SignalstatsContext *s = ctx->priv;
145  AVFilterLink *inlink = outlink->src->inputs[0];
146  const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(outlink->format);
147  s->hsub = desc->log2_chroma_w;
148  s->vsub = desc->log2_chroma_h;
149 
150  outlink->w = inlink->w;
151  outlink->h = inlink->h;
152 
153  s->chromaw = FF_CEIL_RSHIFT(inlink->w, s->hsub);
154  s->chromah = FF_CEIL_RSHIFT(inlink->h, s->vsub);
155 
156  s->fs = inlink->w * inlink->h;
157  s->cfs = s->chromaw * s->chromah;
158 
159  s->nb_jobs = FFMAX(1, FFMIN(inlink->h, ctx->graph->nb_threads));
160  s->jobs_rets = av_malloc_array(s->nb_jobs, sizeof(*s->jobs_rets));
161  if (!s->jobs_rets)
162  return AVERROR(ENOMEM);
163 
164  s->frame_sat = alloc_frame(AV_PIX_FMT_GRAY8, inlink->w, inlink->h);
165  s->frame_hue = alloc_frame(AV_PIX_FMT_GRAY16, inlink->w, inlink->h);
166  if (!s->frame_sat || !s->frame_hue)
167  return AVERROR(ENOMEM);
168 
169  return 0;
170 }
171 
172 static void burn_frame(const SignalstatsContext *s, AVFrame *f, int x, int y)
173 {
174  const int chromax = x >> s->hsub;
175  const int chromay = y >> s->vsub;
176  f->data[0][y * f->linesize[0] + x] = s->yuv_color[0];
177  f->data[1][chromay * f->linesize[1] + chromax] = s->yuv_color[1];
178  f->data[2][chromay * f->linesize[2] + chromax] = s->yuv_color[2];
179 }
180 
181 static int filter_brng(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
182 {
183  ThreadData *td = arg;
184  const SignalstatsContext *s = ctx->priv;
185  const AVFrame *in = td->in;
186  AVFrame *out = td->out;
187  const int w = in->width;
188  const int h = in->height;
189  const int slice_start = (h * jobnr ) / nb_jobs;
190  const int slice_end = (h * (jobnr+1)) / nb_jobs;
191  int x, y, score = 0;
192 
193  for (y = slice_start; y < slice_end; y++) {
194  const int yc = y >> s->vsub;
195  const uint8_t *pluma = &in->data[0][y * in->linesize[0]];
196  const uint8_t *pchromau = &in->data[1][yc * in->linesize[1]];
197  const uint8_t *pchromav = &in->data[2][yc * in->linesize[2]];
198 
199  for (x = 0; x < w; x++) {
200  const int xc = x >> s->hsub;
201  const int luma = pluma[x];
202  const int chromau = pchromau[xc];
203  const int chromav = pchromav[xc];
204  const int filt = luma < 16 || luma > 235 ||
205  chromau < 16 || chromau > 240 ||
206  chromav < 16 || chromav > 240;
207  score += filt;
208  if (out && filt)
209  burn_frame(s, out, x, y);
210  }
211  }
212  return score;
213 }
214 
216 {
217  return ((abs(x - y) + abs (z - y)) / 2) - abs(z - x) > 4; // make 4 configurable?
218 }
219 
220 static int filter_tout(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
221 {
222  ThreadData *td = arg;
223  const SignalstatsContext *s = ctx->priv;
224  const AVFrame *in = td->in;
225  AVFrame *out = td->out;
226  const int w = in->width;
227  const int h = in->height;
228  const int slice_start = (h * jobnr ) / nb_jobs;
229  const int slice_end = (h * (jobnr+1)) / nb_jobs;
230  const uint8_t *p = in->data[0];
231  int lw = in->linesize[0];
232  int x, y, score = 0, filt;
233 
234  for (y = slice_start; y < slice_end; y++) {
235 
236  if (y - 1 < 0 || y + 1 >= h)
237  continue;
238 
239  // detect two pixels above and below (to eliminate interlace artefacts)
240  // should check that video format is infact interlaced.
241 
242 #define FILTER(i, j) \
243  filter_tout_outlier(p[(y-j) * lw + x + i], \
244  p[ y * lw + x + i], \
245  p[(y+j) * lw + x + i])
246 
247 #define FILTER3(j) (FILTER(-1, j) && FILTER(0, j) && FILTER(1, j))
248 
249  if (y - 2 >= 0 && y + 2 < h) {
250  for (x = 1; x < w - 1; x++) {
251  filt = FILTER3(2) && FILTER3(1);
252  score += filt;
253  if (filt && out)
254  burn_frame(s, out, x, y);
255  }
256  } else {
257  for (x = 1; x < w - 1; x++) {
258  filt = FILTER3(1);
259  score += filt;
260  if (filt && out)
261  burn_frame(s, out, x, y);
262  }
263  }
264  }
265  return score;
266 }
267 
268 #define VREP_START 4
269 
270 static int filter_vrep(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
271 {
272  ThreadData *td = arg;
273  const SignalstatsContext *s = ctx->priv;
274  const AVFrame *in = td->in;
275  AVFrame *out = td->out;
276  const int w = in->width;
277  const int h = in->height;
278  const int slice_start = (h * jobnr ) / nb_jobs;
279  const int slice_end = (h * (jobnr+1)) / nb_jobs;
280  const uint8_t *p = in->data[0];
281  const int lw = in->linesize[0];
282  int x, y, score = 0;
283 
284  for (y = slice_start; y < slice_end; y++) {
285  const int y2lw = (y - VREP_START) * lw;
286  const int ylw = y * lw;
287  int filt, totdiff = 0;
288 
289  if (y < VREP_START)
290  continue;
291 
292  for (x = 0; x < w; x++)
293  totdiff += abs(p[y2lw + x] - p[ylw + x]);
294  filt = totdiff < w;
295 
296  score += filt;
297  if (filt && out)
298  for (x = 0; x < w; x++)
299  burn_frame(s, out, x, y);
300  }
301  return score * w;
302 }
303 
304 static const struct {
305  const char *name;
306  int (*process)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
307 } filters_def[] = {
308  {"TOUT", filter_tout},
309  {"VREP", filter_vrep},
310  {"BRNG", filter_brng},
311  {NULL}
312 };
313 
314 #define DEPTH 256
315 
316 static int compute_sat_hue_metrics(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
317 {
318  int i, j;
320  const SignalstatsContext *s = ctx->priv;
321  const AVFrame *src = td->src;
322  AVFrame *dst_sat = td->dst_sat;
323  AVFrame *dst_hue = td->dst_hue;
324 
325  const int slice_start = (s->chromah * jobnr ) / nb_jobs;
326  const int slice_end = (s->chromah * (jobnr+1)) / nb_jobs;
327 
328  const int lsz_u = src->linesize[1];
329  const int lsz_v = src->linesize[2];
330  const uint8_t *p_u = src->data[1] + slice_start * lsz_u;
331  const uint8_t *p_v = src->data[2] + slice_start * lsz_v;
332 
333  const int lsz_sat = dst_sat->linesize[0];
334  const int lsz_hue = dst_hue->linesize[0];
335  uint8_t *p_sat = dst_sat->data[0] + slice_start * lsz_sat;
336  uint8_t *p_hue = dst_hue->data[0] + slice_start * lsz_hue;
337 
338  for (j = slice_start; j < slice_end; j++) {
339  for (i = 0; i < s->chromaw; i++) {
340  const int yuvu = p_u[i];
341  const int yuvv = p_v[i];
342  p_sat[i] = hypot(yuvu - 128, yuvv - 128); // int or round?
343  ((int16_t*)p_hue)[i] = floor((180 / M_PI) * atan2f(yuvu-128, yuvv-128) + 180);
344  }
345  p_u += lsz_u;
346  p_v += lsz_v;
347  p_sat += lsz_sat;
348  p_hue += lsz_hue;
349  }
350 
351  return 0;
352 }
353 
354 static int filter_frame(AVFilterLink *link, AVFrame *in)
355 {
356  AVFilterContext *ctx = link->dst;
357  SignalstatsContext *s = ctx->priv;
358  AVFilterLink *outlink = ctx->outputs[0];
359  AVFrame *out = in;
360  int i, j;
361  int w = 0, cw = 0, // in
362  pw = 0, cpw = 0; // prev
363  int fil;
364  char metabuf[128];
365  unsigned int histy[DEPTH] = {0},
366  histu[DEPTH] = {0},
367  histv[DEPTH] = {0},
368  histhue[360] = {0},
369  histsat[DEPTH] = {0}; // limited to 8 bit data.
370  int miny = -1, minu = -1, minv = -1;
371  int maxy = -1, maxu = -1, maxv = -1;
372  int lowy = -1, lowu = -1, lowv = -1;
373  int highy = -1, highu = -1, highv = -1;
374  int minsat = -1, maxsat = -1, lowsat = -1, highsat = -1;
375  int lowp, highp, clowp, chighp;
376  int accy, accu, accv;
377  int accsat, acchue = 0;
378  int medhue, maxhue;
379  int toty = 0, totu = 0, totv = 0, totsat=0;
380  int tothue = 0;
381  int dify = 0, difu = 0, difv = 0;
382 
383  int filtot[FILT_NUMB] = {0};
384  AVFrame *prev;
385 
386  AVFrame *sat = s->frame_sat;
387  AVFrame *hue = s->frame_hue;
388  const uint8_t *p_sat = sat->data[0];
389  const uint8_t *p_hue = hue->data[0];
390  const int lsz_sat = sat->linesize[0];
391  const int lsz_hue = hue->linesize[0];
392  ThreadDataHueSatMetrics td_huesat = {
393  .src = in,
394  .dst_sat = sat,
395  .dst_hue = hue,
396  };
397 
398  if (!s->frame_prev)
399  s->frame_prev = av_frame_clone(in);
400 
401  prev = s->frame_prev;
402 
403  if (s->outfilter != FILTER_NONE) {
404  out = av_frame_clone(in);
406  }
407 
408  ctx->internal->execute(ctx, compute_sat_hue_metrics, &td_huesat,
409  NULL, FFMIN(s->chromah, ctx->graph->nb_threads));
410 
411  // Calculate luma histogram and difference with previous frame or field.
412  for (j = 0; j < link->h; j++) {
413  for (i = 0; i < link->w; i++) {
414  const int yuv = in->data[0][w + i];
415  histy[yuv]++;
416  dify += abs(yuv - prev->data[0][pw + i]);
417  }
418  w += in->linesize[0];
419  pw += prev->linesize[0];
420  }
421 
422  // Calculate chroma histogram and difference with previous frame or field.
423  for (j = 0; j < s->chromah; j++) {
424  for (i = 0; i < s->chromaw; i++) {
425  const int yuvu = in->data[1][cw+i];
426  const int yuvv = in->data[2][cw+i];
427  histu[yuvu]++;
428  difu += abs(yuvu - prev->data[1][cpw+i]);
429  histv[yuvv]++;
430  difv += abs(yuvv - prev->data[2][cpw+i]);
431 
432  histsat[p_sat[i]]++;
433  histhue[((int16_t*)p_hue)[i]]++;
434  }
435  cw += in->linesize[1];
436  cpw += prev->linesize[1];
437  p_sat += lsz_sat;
438  p_hue += lsz_hue;
439  }
440 
441  for (fil = 0; fil < FILT_NUMB; fil ++) {
442  if (s->filters & 1<<fil) {
443  ThreadData td = {
444  .in = in,
445  .out = out != in && s->outfilter == fil ? out : NULL,
446  };
447  memset(s->jobs_rets, 0, s->nb_jobs * sizeof(*s->jobs_rets));
448  ctx->internal->execute(ctx, filters_def[fil].process,
449  &td, s->jobs_rets, s->nb_jobs);
450  for (i = 0; i < s->nb_jobs; i++)
451  filtot[fil] += s->jobs_rets[i];
452  }
453  }
454 
455  // find low / high based on histogram percentile
456  // these only need to be calculated once.
457 
458  lowp = lrint(s->fs * 10 / 100.);
459  highp = lrint(s->fs * 90 / 100.);
460  clowp = lrint(s->cfs * 10 / 100.);
461  chighp = lrint(s->cfs * 90 / 100.);
462 
463  accy = accu = accv = accsat = 0;
464  for (fil = 0; fil < DEPTH; fil++) {
465  if (miny < 0 && histy[fil]) miny = fil;
466  if (minu < 0 && histu[fil]) minu = fil;
467  if (minv < 0 && histv[fil]) minv = fil;
468  if (minsat < 0 && histsat[fil]) minsat = fil;
469 
470  if (histy[fil]) maxy = fil;
471  if (histu[fil]) maxu = fil;
472  if (histv[fil]) maxv = fil;
473  if (histsat[fil]) maxsat = fil;
474 
475  toty += histy[fil] * fil;
476  totu += histu[fil] * fil;
477  totv += histv[fil] * fil;
478  totsat += histsat[fil] * fil;
479 
480  accy += histy[fil];
481  accu += histu[fil];
482  accv += histv[fil];
483  accsat += histsat[fil];
484 
485  if (lowy == -1 && accy >= lowp) lowy = fil;
486  if (lowu == -1 && accu >= clowp) lowu = fil;
487  if (lowv == -1 && accv >= clowp) lowv = fil;
488  if (lowsat == -1 && accsat >= clowp) lowsat = fil;
489 
490  if (highy == -1 && accy >= highp) highy = fil;
491  if (highu == -1 && accu >= chighp) highu = fil;
492  if (highv == -1 && accv >= chighp) highv = fil;
493  if (highsat == -1 && accsat >= chighp) highsat = fil;
494  }
495 
496  maxhue = histhue[0];
497  medhue = -1;
498  for (fil = 0; fil < 360; fil++) {
499  tothue += histhue[fil] * fil;
500  acchue += histhue[fil];
501 
502  if (medhue == -1 && acchue > s->cfs / 2)
503  medhue = fil;
504  if (histhue[fil] > maxhue) {
505  maxhue = histhue[fil];
506  }
507  }
508 
510  s->frame_prev = av_frame_clone(in);
511 
512 #define SET_META(key, fmt, val) do { \
513  snprintf(metabuf, sizeof(metabuf), fmt, val); \
514  av_dict_set(&out->metadata, "lavfi.signalstats." key, metabuf, 0); \
515 } while (0)
516 
517  SET_META("YMIN", "%d", miny);
518  SET_META("YLOW", "%d", lowy);
519  SET_META("YAVG", "%g", 1.0 * toty / s->fs);
520  SET_META("YHIGH", "%d", highy);
521  SET_META("YMAX", "%d", maxy);
522 
523  SET_META("UMIN", "%d", minu);
524  SET_META("ULOW", "%d", lowu);
525  SET_META("UAVG", "%g", 1.0 * totu / s->cfs);
526  SET_META("UHIGH", "%d", highu);
527  SET_META("UMAX", "%d", maxu);
528 
529  SET_META("VMIN", "%d", minv);
530  SET_META("VLOW", "%d", lowv);
531  SET_META("VAVG", "%g", 1.0 * totv / s->cfs);
532  SET_META("VHIGH", "%d", highv);
533  SET_META("VMAX", "%d", maxv);
534 
535  SET_META("SATMIN", "%d", minsat);
536  SET_META("SATLOW", "%d", lowsat);
537  SET_META("SATAVG", "%g", 1.0 * totsat / s->cfs);
538  SET_META("SATHIGH", "%d", highsat);
539  SET_META("SATMAX", "%d", maxsat);
540 
541  SET_META("HUEMED", "%d", medhue);
542  SET_META("HUEAVG", "%g", 1.0 * tothue / s->cfs);
543 
544  SET_META("YDIF", "%g", 1.0 * dify / s->fs);
545  SET_META("UDIF", "%g", 1.0 * difu / s->cfs);
546  SET_META("VDIF", "%g", 1.0 * difv / s->cfs);
547 
548  for (fil = 0; fil < FILT_NUMB; fil ++) {
549  if (s->filters & 1<<fil) {
550  char metaname[128];
551  snprintf(metabuf, sizeof(metabuf), "%g", 1.0 * filtot[fil] / s->fs);
552  snprintf(metaname, sizeof(metaname), "lavfi.signalstats.%s", filters_def[fil].name);
553  av_dict_set(&out->metadata, metaname, metabuf, 0);
554  }
555  }
556 
557  if (in != out)
558  av_frame_free(&in);
559  return ff_filter_frame(outlink, out);
560 }
561 
562 static const AVFilterPad signalstats_inputs[] = {
563  {
564  .name = "default",
565  .type = AVMEDIA_TYPE_VIDEO,
566  .filter_frame = filter_frame,
567  },
568  { NULL }
569 };
570 
572  {
573  .name = "default",
574  .config_props = config_props,
575  .type = AVMEDIA_TYPE_VIDEO,
576  },
577  { NULL }
578 };
579 
581  .name = "signalstats",
582  .description = "Generate statistics from video analysis.",
583  .init = init,
584  .uninit = uninit,
585  .query_formats = query_formats,
586  .priv_size = sizeof(SignalstatsContext),
587  .inputs = signalstats_inputs,
588  .outputs = signalstats_outputs,
589  .priv_class = &signalstats_class,
591 };