Commit | Line | Data |
---|---|---|
2ba45a60 DM |
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 { | |
28 | FILTER_NONE = -1, | |
29 | FILTER_TOUT, | |
30 | FILTER_VREP, | |
31 | FILTER_BRNG, | |
32 | FILT_NUMB | |
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 | |
43 | enum FilterMode outfilter; | |
44 | int filters; | |
45 | AVFrame *frame_prev; | |
46 | char *vrep_line; | |
47 | uint8_t rgba_color[4]; | |
48 | int yuv_color[3]; | |
49 | } SignalstatsContext; | |
50 | ||
51 | #define OFFSET(x) offsetof(SignalstatsContext, x) | |
52 | #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM | |
53 | ||
54 | static const AVOption signalstats_options[] = { | |
55 | {"stat", "set statistics filters", OFFSET(filters), AV_OPT_TYPE_FLAGS, {.i64=0}, 0, INT_MAX, FLAGS, "filters"}, | |
56 | {"tout", "analyze pixels for temporal outliers", 0, AV_OPT_TYPE_CONST, {.i64=1<<FILTER_TOUT}, 0, 0, FLAGS, "filters"}, | |
57 | {"vrep", "analyze video lines for vertical line repitition", 0, AV_OPT_TYPE_CONST, {.i64=1<<FILTER_VREP}, 0, 0, FLAGS, "filters"}, | |
58 | {"brng", "analyze for pixels outside of broadcast range", 0, AV_OPT_TYPE_CONST, {.i64=1<<FILTER_BRNG}, 0, 0, FLAGS, "filters"}, | |
59 | {"out", "set video filter", OFFSET(outfilter), AV_OPT_TYPE_INT, {.i64=FILTER_NONE}, -1, FILT_NUMB-1, FLAGS, "out"}, | |
60 | {"tout", "highlight pixels that depict temporal outliers", 0, AV_OPT_TYPE_CONST, {.i64=FILTER_TOUT}, 0, 0, FLAGS, "out"}, | |
61 | {"vrep", "highlight video lines that depict vertical line repitition", 0, AV_OPT_TYPE_CONST, {.i64=FILTER_VREP}, 0, 0, FLAGS, "out"}, | |
62 | {"brng", "highlight pixels that are outside of broadcast range", 0, AV_OPT_TYPE_CONST, {.i64=FILTER_BRNG}, 0, 0, FLAGS, "out"}, | |
63 | {"c", "set highlight color", OFFSET(rgba_color), AV_OPT_TYPE_COLOR, {.str="yellow"}, .flags=FLAGS}, | |
64 | {"color", "set highlight color", OFFSET(rgba_color), AV_OPT_TYPE_COLOR, {.str="yellow"}, .flags=FLAGS}, | |
65 | {NULL} | |
66 | }; | |
67 | ||
68 | AVFILTER_DEFINE_CLASS(signalstats); | |
69 | ||
70 | static av_cold int init(AVFilterContext *ctx) | |
71 | { | |
72 | uint8_t r, g, b; | |
73 | SignalstatsContext *s = ctx->priv; | |
74 | ||
75 | if (s->outfilter != FILTER_NONE) | |
76 | s->filters |= 1 << s->outfilter; | |
77 | ||
78 | r = s->rgba_color[0]; | |
79 | g = s->rgba_color[1]; | |
80 | b = s->rgba_color[2]; | |
81 | s->yuv_color[0] = (( 66*r + 129*g + 25*b + (1<<7)) >> 8) + 16; | |
82 | s->yuv_color[1] = ((-38*r + -74*g + 112*b + (1<<7)) >> 8) + 128; | |
83 | s->yuv_color[2] = ((112*r + -94*g + -18*b + (1<<7)) >> 8) + 128; | |
84 | return 0; | |
85 | } | |
86 | ||
87 | static av_cold void uninit(AVFilterContext *ctx) | |
88 | { | |
89 | SignalstatsContext *s = ctx->priv; | |
90 | av_frame_free(&s->frame_prev); | |
91 | av_freep(&s->vrep_line); | |
92 | } | |
93 | ||
94 | static int query_formats(AVFilterContext *ctx) | |
95 | { | |
96 | // TODO: add more | |
97 | static const enum AVPixelFormat pix_fmts[] = { | |
98 | AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P, | |
99 | AV_PIX_FMT_NONE | |
100 | }; | |
101 | ||
102 | ff_set_common_formats(ctx, ff_make_format_list(pix_fmts)); | |
103 | return 0; | |
104 | } | |
105 | ||
106 | static int config_props(AVFilterLink *outlink) | |
107 | { | |
108 | AVFilterContext *ctx = outlink->src; | |
109 | SignalstatsContext *s = ctx->priv; | |
110 | AVFilterLink *inlink = outlink->src->inputs[0]; | |
111 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(outlink->format); | |
112 | s->hsub = desc->log2_chroma_w; | |
113 | s->vsub = desc->log2_chroma_h; | |
114 | ||
115 | outlink->w = inlink->w; | |
116 | outlink->h = inlink->h; | |
117 | ||
118 | s->chromaw = FF_CEIL_RSHIFT(inlink->w, s->hsub); | |
119 | s->chromah = FF_CEIL_RSHIFT(inlink->h, s->vsub); | |
120 | ||
121 | s->fs = inlink->w * inlink->h; | |
122 | s->cfs = s->chromaw * s->chromah; | |
123 | ||
124 | if (s->filters & 1<<FILTER_VREP) { | |
125 | s->vrep_line = av_malloc(inlink->h * sizeof(*s->vrep_line)); | |
126 | if (!s->vrep_line) | |
127 | return AVERROR(ENOMEM); | |
128 | } | |
129 | ||
130 | return 0; | |
131 | } | |
132 | ||
133 | static void burn_frame(SignalstatsContext *s, AVFrame *f, int x, int y) | |
134 | { | |
135 | const int chromax = x >> s->hsub; | |
136 | const int chromay = y >> s->vsub; | |
137 | f->data[0][y * f->linesize[0] + x] = s->yuv_color[0]; | |
138 | f->data[1][chromay * f->linesize[1] + chromax] = s->yuv_color[1]; | |
139 | f->data[2][chromay * f->linesize[2] + chromax] = s->yuv_color[2]; | |
140 | } | |
141 | ||
142 | static int filter_brng(SignalstatsContext *s, const AVFrame *in, AVFrame *out, int y, int w, int h) | |
143 | { | |
144 | int x, score = 0; | |
145 | const int yc = y >> s->vsub; | |
146 | const uint8_t *pluma = &in->data[0][y * in->linesize[0]]; | |
147 | const uint8_t *pchromau = &in->data[1][yc * in->linesize[1]]; | |
148 | const uint8_t *pchromav = &in->data[2][yc * in->linesize[2]]; | |
149 | ||
150 | for (x = 0; x < w; x++) { | |
151 | const int xc = x >> s->hsub; | |
152 | const int luma = pluma[x]; | |
153 | const int chromau = pchromau[xc]; | |
154 | const int chromav = pchromav[xc]; | |
155 | const int filt = luma < 16 || luma > 235 || | |
156 | chromau < 16 || chromau > 240 || | |
157 | chromav < 16 || chromav > 240; | |
158 | score += filt; | |
159 | if (out && filt) | |
160 | burn_frame(s, out, x, y); | |
161 | } | |
162 | return score; | |
163 | } | |
164 | ||
165 | static int filter_tout_outlier(uint8_t x, uint8_t y, uint8_t z) | |
166 | { | |
167 | return ((abs(x - y) + abs (z - y)) / 2) - abs(z - x) > 4; // make 4 configurable? | |
168 | } | |
169 | ||
170 | static int filter_tout(SignalstatsContext *s, const AVFrame *in, AVFrame *out, int y, int w, int h) | |
171 | { | |
172 | const uint8_t *p = in->data[0]; | |
173 | int lw = in->linesize[0]; | |
174 | int x, score = 0, filt; | |
175 | ||
176 | if (y - 1 < 0 || y + 1 >= h) | |
177 | return 0; | |
178 | ||
179 | // detect two pixels above and below (to eliminate interlace artefacts) | |
180 | // should check that video format is infact interlaced. | |
181 | ||
182 | #define FILTER(i, j) \ | |
183 | filter_tout_outlier(p[(y-j) * lw + x + i], \ | |
184 | p[ y * lw + x + i], \ | |
185 | p[(y+j) * lw + x + i]) | |
186 | ||
187 | #define FILTER3(j) (FILTER(-1, j) && FILTER(0, j) && FILTER(1, j)) | |
188 | ||
189 | if (y - 2 >= 0 && y + 2 < h) { | |
190 | for (x = 1; x < w - 1; x++) { | |
191 | filt = FILTER3(2) && FILTER3(1); | |
192 | score += filt; | |
193 | if (filt && out) | |
194 | burn_frame(s, out, x, y); | |
195 | } | |
196 | } else { | |
197 | for (x = 1; x < w - 1; x++) { | |
198 | filt = FILTER3(1); | |
199 | score += filt; | |
200 | if (filt && out) | |
201 | burn_frame(s, out, x, y); | |
202 | } | |
203 | } | |
204 | return score; | |
205 | } | |
206 | ||
207 | #define VREP_START 4 | |
208 | ||
209 | static void filter_init_vrep(SignalstatsContext *s, const AVFrame *p, int w, int h) | |
210 | { | |
211 | int i, y; | |
212 | int lw = p->linesize[0]; | |
213 | ||
214 | for (y = VREP_START; y < h; y++) { | |
215 | int totdiff = 0; | |
216 | int y2lw = (y - VREP_START) * lw; | |
217 | int ylw = y * lw; | |
218 | ||
219 | for (i = 0; i < w; i++) | |
220 | totdiff += abs(p->data[0][y2lw + i] - p->data[0][ylw + i]); | |
221 | ||
222 | /* this value should be definable */ | |
223 | s->vrep_line[y] = totdiff < w; | |
224 | } | |
225 | } | |
226 | ||
227 | static int filter_vrep(SignalstatsContext *s, const AVFrame *in, AVFrame *out, int y, int w, int h) | |
228 | { | |
229 | int x, score = 0; | |
230 | ||
231 | if (y < VREP_START) | |
232 | return 0; | |
233 | ||
234 | for (x = 0; x < w; x++) { | |
235 | if (s->vrep_line[y]) { | |
236 | score++; | |
237 | if (out) | |
238 | burn_frame(s, out, x, y); | |
239 | } | |
240 | } | |
241 | return score; | |
242 | } | |
243 | ||
244 | static const struct { | |
245 | const char *name; | |
246 | void (*init)(SignalstatsContext *s, const AVFrame *p, int w, int h); | |
247 | int (*process)(SignalstatsContext *s, const AVFrame *in, AVFrame *out, int y, int w, int h); | |
248 | } filters_def[] = { | |
249 | {"TOUT", NULL, filter_tout}, | |
250 | {"VREP", filter_init_vrep, filter_vrep}, | |
251 | {"BRNG", NULL, filter_brng}, | |
252 | {NULL} | |
253 | }; | |
254 | ||
255 | #define DEPTH 256 | |
256 | ||
257 | static int filter_frame(AVFilterLink *link, AVFrame *in) | |
258 | { | |
259 | SignalstatsContext *s = link->dst->priv; | |
260 | AVFilterLink *outlink = link->dst->outputs[0]; | |
261 | AVFrame *out = in; | |
262 | int i, j; | |
263 | int w = 0, cw = 0, // in | |
264 | pw = 0, cpw = 0; // prev | |
265 | int yuv, yuvu, yuvv; | |
266 | int fil; | |
267 | char metabuf[128]; | |
268 | unsigned int histy[DEPTH] = {0}, | |
269 | histu[DEPTH] = {0}, | |
270 | histv[DEPTH] = {0}, | |
271 | histhue[360] = {0}, | |
272 | histsat[DEPTH] = {0}; // limited to 8 bit data. | |
273 | int miny = -1, minu = -1, minv = -1; | |
274 | int maxy = -1, maxu = -1, maxv = -1; | |
275 | int lowy = -1, lowu = -1, lowv = -1; | |
276 | int highy = -1, highu = -1, highv = -1; | |
277 | int minsat = -1, maxsat = -1, lowsat = -1, highsat = -1; | |
278 | int lowp, highp, clowp, chighp; | |
279 | int accy, accu, accv; | |
280 | int accsat, acchue = 0; | |
281 | int medhue, maxhue; | |
282 | int toty = 0, totu = 0, totv = 0, totsat=0; | |
283 | int tothue = 0; | |
284 | int dify = 0, difu = 0, difv = 0; | |
285 | ||
286 | int filtot[FILT_NUMB] = {0}; | |
287 | AVFrame *prev; | |
288 | ||
289 | if (!s->frame_prev) | |
290 | s->frame_prev = av_frame_clone(in); | |
291 | ||
292 | prev = s->frame_prev; | |
293 | ||
294 | if (s->outfilter != FILTER_NONE) | |
295 | out = av_frame_clone(in); | |
296 | ||
297 | for (fil = 0; fil < FILT_NUMB; fil ++) | |
298 | if ((s->filters & 1<<fil) && filters_def[fil].init) | |
299 | filters_def[fil].init(s, in, link->w, link->h); | |
300 | ||
301 | // Calculate luma histogram and difference with previous frame or field. | |
302 | for (j = 0; j < link->h; j++) { | |
303 | for (i = 0; i < link->w; i++) { | |
304 | yuv = in->data[0][w + i]; | |
305 | histy[yuv]++; | |
306 | dify += abs(in->data[0][w + i] - prev->data[0][pw + i]); | |
307 | } | |
308 | w += in->linesize[0]; | |
309 | pw += prev->linesize[0]; | |
310 | } | |
311 | ||
312 | // Calculate chroma histogram and difference with previous frame or field. | |
313 | for (j = 0; j < s->chromah; j++) { | |
314 | for (i = 0; i < s->chromaw; i++) { | |
315 | int sat, hue; | |
316 | ||
317 | yuvu = in->data[1][cw+i]; | |
318 | yuvv = in->data[2][cw+i]; | |
319 | histu[yuvu]++; | |
320 | difu += abs(in->data[1][cw+i] - prev->data[1][cpw+i]); | |
321 | histv[yuvv]++; | |
322 | difv += abs(in->data[2][cw+i] - prev->data[2][cpw+i]); | |
323 | ||
324 | // int or round? | |
325 | sat = hypot(yuvu - 128, yuvv - 128); | |
326 | histsat[sat]++; | |
327 | hue = floor((180 / M_PI) * atan2f(yuvu-128, yuvv-128) + 180); | |
328 | histhue[hue]++; | |
329 | } | |
330 | cw += in->linesize[1]; | |
331 | cpw += prev->linesize[1]; | |
332 | } | |
333 | ||
334 | for (j = 0; j < link->h; j++) { | |
335 | for (fil = 0; fil < FILT_NUMB; fil ++) { | |
336 | if (s->filters & 1<<fil) { | |
337 | AVFrame *dbg = out != in && s->outfilter == fil ? out : NULL; | |
338 | filtot[fil] += filters_def[fil].process(s, in, dbg, j, link->w, link->h); | |
339 | } | |
340 | } | |
341 | } | |
342 | ||
343 | // find low / high based on histogram percentile | |
344 | // these only need to be calculated once. | |
345 | ||
346 | lowp = lrint(s->fs * 10 / 100.); | |
347 | highp = lrint(s->fs * 90 / 100.); | |
348 | clowp = lrint(s->cfs * 10 / 100.); | |
349 | chighp = lrint(s->cfs * 90 / 100.); | |
350 | ||
351 | accy = accu = accv = accsat = 0; | |
352 | for (fil = 0; fil < DEPTH; fil++) { | |
353 | if (miny < 0 && histy[fil]) miny = fil; | |
354 | if (minu < 0 && histu[fil]) minu = fil; | |
355 | if (minv < 0 && histv[fil]) minv = fil; | |
356 | if (minsat < 0 && histsat[fil]) minsat = fil; | |
357 | ||
358 | if (histy[fil]) maxy = fil; | |
359 | if (histu[fil]) maxu = fil; | |
360 | if (histv[fil]) maxv = fil; | |
361 | if (histsat[fil]) maxsat = fil; | |
362 | ||
363 | toty += histy[fil] * fil; | |
364 | totu += histu[fil] * fil; | |
365 | totv += histv[fil] * fil; | |
366 | totsat += histsat[fil] * fil; | |
367 | ||
368 | accy += histy[fil]; | |
369 | accu += histu[fil]; | |
370 | accv += histv[fil]; | |
371 | accsat += histsat[fil]; | |
372 | ||
373 | if (lowy == -1 && accy >= lowp) lowy = fil; | |
374 | if (lowu == -1 && accu >= clowp) lowu = fil; | |
375 | if (lowv == -1 && accv >= clowp) lowv = fil; | |
376 | if (lowsat == -1 && accsat >= clowp) lowsat = fil; | |
377 | ||
378 | if (highy == -1 && accy >= highp) highy = fil; | |
379 | if (highu == -1 && accu >= chighp) highu = fil; | |
380 | if (highv == -1 && accv >= chighp) highv = fil; | |
381 | if (highsat == -1 && accsat >= chighp) highsat = fil; | |
382 | } | |
383 | ||
384 | maxhue = histhue[0]; | |
385 | medhue = -1; | |
386 | for (fil = 0; fil < 360; fil++) { | |
387 | tothue += histhue[fil] * fil; | |
388 | acchue += histhue[fil]; | |
389 | ||
390 | if (medhue == -1 && acchue > s->cfs / 2) | |
391 | medhue = fil; | |
392 | if (histhue[fil] > maxhue) { | |
393 | maxhue = histhue[fil]; | |
394 | } | |
395 | } | |
396 | ||
397 | av_frame_free(&s->frame_prev); | |
398 | s->frame_prev = av_frame_clone(in); | |
399 | ||
400 | #define SET_META(key, fmt, val) do { \ | |
401 | snprintf(metabuf, sizeof(metabuf), fmt, val); \ | |
402 | av_dict_set(&out->metadata, "lavfi.signalstats." key, metabuf, 0); \ | |
403 | } while (0) | |
404 | ||
405 | SET_META("YMIN", "%d", miny); | |
406 | SET_META("YLOW", "%d", lowy); | |
407 | SET_META("YAVG", "%g", 1.0 * toty / s->fs); | |
408 | SET_META("YHIGH", "%d", highy); | |
409 | SET_META("YMAX", "%d", maxy); | |
410 | ||
411 | SET_META("UMIN", "%d", minu); | |
412 | SET_META("ULOW", "%d", lowu); | |
413 | SET_META("UAVG", "%g", 1.0 * totu / s->cfs); | |
414 | SET_META("UHIGH", "%d", highu); | |
415 | SET_META("UMAX", "%d", maxu); | |
416 | ||
417 | SET_META("VMIN", "%d", minv); | |
418 | SET_META("VLOW", "%d", lowv); | |
419 | SET_META("VAVG", "%g", 1.0 * totv / s->cfs); | |
420 | SET_META("VHIGH", "%d", highv); | |
421 | SET_META("VMAX", "%d", maxv); | |
422 | ||
423 | SET_META("SATMIN", "%d", minsat); | |
424 | SET_META("SATLOW", "%d", lowsat); | |
425 | SET_META("SATAVG", "%g", 1.0 * totsat / s->cfs); | |
426 | SET_META("SATHIGH", "%d", highsat); | |
427 | SET_META("SATMAX", "%d", maxsat); | |
428 | ||
429 | SET_META("HUEMED", "%d", medhue); | |
430 | SET_META("HUEAVG", "%g", 1.0 * tothue / s->cfs); | |
431 | ||
432 | SET_META("YDIF", "%g", 1.0 * dify / s->fs); | |
433 | SET_META("UDIF", "%g", 1.0 * difu / s->cfs); | |
434 | SET_META("VDIF", "%g", 1.0 * difv / s->cfs); | |
435 | ||
436 | for (fil = 0; fil < FILT_NUMB; fil ++) { | |
437 | if (s->filters & 1<<fil) { | |
438 | char metaname[128]; | |
439 | snprintf(metabuf, sizeof(metabuf), "%g", 1.0 * filtot[fil] / s->fs); | |
440 | snprintf(metaname, sizeof(metaname), "lavfi.signalstats.%s", filters_def[fil].name); | |
441 | av_dict_set(&out->metadata, metaname, metabuf, 0); | |
442 | } | |
443 | } | |
444 | ||
445 | if (in != out) | |
446 | av_frame_free(&in); | |
447 | return ff_filter_frame(outlink, out); | |
448 | } | |
449 | ||
450 | static const AVFilterPad signalstats_inputs[] = { | |
451 | { | |
452 | .name = "default", | |
453 | .type = AVMEDIA_TYPE_VIDEO, | |
454 | .filter_frame = filter_frame, | |
455 | }, | |
456 | { NULL } | |
457 | }; | |
458 | ||
459 | static const AVFilterPad signalstats_outputs[] = { | |
460 | { | |
461 | .name = "default", | |
462 | .config_props = config_props, | |
463 | .type = AVMEDIA_TYPE_VIDEO, | |
464 | }, | |
465 | { NULL } | |
466 | }; | |
467 | ||
468 | AVFilter ff_vf_signalstats = { | |
469 | .name = "signalstats", | |
470 | .description = "Generate statistics from video analysis.", | |
471 | .init = init, | |
472 | .uninit = uninit, | |
473 | .query_formats = query_formats, | |
474 | .priv_size = sizeof(SignalstatsContext), | |
475 | .inputs = signalstats_inputs, | |
476 | .outputs = signalstats_outputs, | |
477 | .priv_class = &signalstats_class, | |
478 | }; |