1 /*
2 * Copyright (c) 2019 Paul B Mahol
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 "libavutil/avstring.h"
22 #include "libavutil/imgutils.h"
23 #include "libavutil/intreadwrite.h"
24 #include "libavutil/opt.h"
25 #include "libavutil/pixdesc.h"
26 #include "libavutil/qsort.h"
27
28 #include "avfilter.h"
29 #include "formats.h"
30 #include "internal.h"
31 #include "framesync.h"
32 #include "video.h"
33
34 typedef struct XMedianContext {
35 const AVClass *class;
36 const AVPixFmtDescriptor *desc;
37 int nb_inputs;
38 int nb_frames;
39 int planes;
40 float percentile;
41
42 int tmedian;
43 int radius;
44 int index;
45 int depth;
46 int max;
47 int nb_planes;
48 int linesize[4];
49 int width[4];
50 int height[4];
51
52 AVFrame **frames;
53 FFFrameSync fs;
54
55 int (*median_frames)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
56 } XMedianContext;
57
query_formats(AVFilterContext * ctx)58 static int query_formats(AVFilterContext *ctx)
59 {
60 static const enum AVPixelFormat pixel_fmts[] = {
61 AV_PIX_FMT_GRAY8,
62 AV_PIX_FMT_GRAY9,
63 AV_PIX_FMT_GRAY10,
64 AV_PIX_FMT_GRAY12,
65 AV_PIX_FMT_GRAY14,
66 AV_PIX_FMT_GRAY16,
67 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P,
68 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P,
69 AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV444P,
70 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
71 AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ444P,
72 AV_PIX_FMT_YUVJ411P,
73 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
74 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
75 AV_PIX_FMT_YUV440P10,
76 AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV420P12,
77 AV_PIX_FMT_YUV440P12,
78 AV_PIX_FMT_YUV444P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV420P14,
79 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
80 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
81 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
82 AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA444P,
83 AV_PIX_FMT_YUVA444P9, AV_PIX_FMT_YUVA444P10, AV_PIX_FMT_YUVA444P12, AV_PIX_FMT_YUVA444P16,
84 AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA422P12, AV_PIX_FMT_YUVA422P16,
85 AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA420P16,
86 AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
87 AV_PIX_FMT_NONE
88 };
89 AVFilterFormats *formats = ff_make_format_list(pixel_fmts);
90 if (!formats)
91 return AVERROR(ENOMEM);
92 return ff_set_common_formats(ctx, formats);
93 }
94
init(AVFilterContext * ctx)95 static av_cold int init(AVFilterContext *ctx)
96 {
97 XMedianContext *s = ctx->priv;
98 int ret;
99
100 s->tmedian = !strcmp(ctx->filter->name, "tmedian");
101
102 if (!s->tmedian) {
103 s->radius = s->nb_inputs / 2;
104 } else {
105 s->nb_inputs = s->radius * 2 + 1;
106 }
107
108 if (s->nb_inputs & 1)
109 s->index = s->radius * 2.f * s->percentile;
110 else
111 s->index = av_clip(s->radius * 2.f * s->percentile, 1, s->nb_inputs - 1);
112 s->frames = av_calloc(s->nb_inputs, sizeof(*s->frames));
113 if (!s->frames)
114 return AVERROR(ENOMEM);
115
116 for (int i = 0; i < s->nb_inputs && !s->tmedian; i++) {
117 AVFilterPad pad = { 0 };
118
119 pad.type = AVMEDIA_TYPE_VIDEO;
120 pad.name = av_asprintf("input%d", i);
121 if (!pad.name)
122 return AVERROR(ENOMEM);
123
124 if ((ret = ff_insert_inpad(ctx, i, &pad)) < 0) {
125 av_freep(&pad.name);
126 return ret;
127 }
128 }
129
130 return 0;
131 }
132
133 typedef struct ThreadData {
134 AVFrame **in, *out;
135 } ThreadData;
136
comparei(const void * p1,const void * p2)137 static int comparei(const void *p1, const void *p2)
138 {
139 int left = *(const int *)p1;
140 int right = *(const int *)p2;
141 return FFDIFFSIGN(left, right);
142 }
143
median_frames16(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)144 static int median_frames16(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
145 {
146 XMedianContext *s = ctx->priv;
147 ThreadData *td = arg;
148 AVFrame **in = td->in;
149 AVFrame *out = td->out;
150 const int nb_inputs = s->nb_inputs;
151 const int radius = s->radius;
152 const int index = s->index;
153 int values[256];
154
155 for (int p = 0; p < s->nb_planes; p++) {
156 const int slice_start = (s->height[p] * jobnr) / nb_jobs;
157 const int slice_end = (s->height[p] * (jobnr+1)) / nb_jobs;
158 uint16_t *dst = (uint16_t *)(out->data[p] + slice_start * out->linesize[p]);
159
160 if (!((1 << p) & s->planes)) {
161 av_image_copy_plane((uint8_t *)dst, out->linesize[p],
162 in[radius]->data[p] + slice_start * in[radius]->linesize[p],
163 in[radius]->linesize[p],
164 s->linesize[p], slice_end - slice_start);
165 continue;
166 }
167
168 for (int y = slice_start; y < slice_end; y++) {
169 for (int x = 0; x < s->width[p]; x++) {
170 for (int i = 0; i < nb_inputs; i++) {
171 const uint16_t *src = (const uint16_t *)(in[i]->data[p] + y * in[i]->linesize[p]);
172 values[i] = src[x];
173 }
174
175 AV_QSORT(values, nb_inputs, int, comparei);
176 if (nb_inputs & 1)
177 dst[x] = values[index];
178 else
179 dst[x] = (values[index] + values[index - 1]) >> 1;
180 }
181
182 dst += out->linesize[p] / 2;
183 }
184 }
185
186 return 0;
187 }
188
median_frames8(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)189 static int median_frames8(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
190 {
191 XMedianContext *s = ctx->priv;
192 ThreadData *td = arg;
193 AVFrame **in = td->in;
194 AVFrame *out = td->out;
195 const int nb_inputs = s->nb_inputs;
196 const int radius = s->radius;
197 const int index = s->index;
198 int values[256];
199
200 for (int p = 0; p < s->nb_planes; p++) {
201 const int slice_start = (s->height[p] * jobnr) / nb_jobs;
202 const int slice_end = (s->height[p] * (jobnr+1)) / nb_jobs;
203 uint8_t *dst = out->data[p] + slice_start * out->linesize[p];
204
205 if (!((1 << p) & s->planes)) {
206 av_image_copy_plane(dst, out->linesize[p],
207 in[radius]->data[p] + slice_start * in[radius]->linesize[p],
208 in[radius]->linesize[p],
209 s->linesize[p], slice_end - slice_start);
210 continue;
211 }
212
213 for (int y = slice_start; y < slice_end; y++) {
214 for (int x = 0; x < s->width[p]; x++) {
215 for (int i = 0; i < nb_inputs; i++)
216 values[i] = in[i]->data[p][y * in[i]->linesize[p] + x];
217
218 AV_QSORT(values, nb_inputs, int, comparei);
219 if (nb_inputs & 1)
220 dst[x] = values[index];
221 else
222 dst[x] = (values[index] + values[index - 1]) >> 1;
223 }
224
225 dst += out->linesize[p];
226 }
227 }
228
229 return 0;
230 }
231
process_frame(FFFrameSync * fs)232 static int process_frame(FFFrameSync *fs)
233 {
234 AVFilterContext *ctx = fs->parent;
235 AVFilterLink *outlink = ctx->outputs[0];
236 XMedianContext *s = fs->opaque;
237 AVFrame **in = s->frames;
238 AVFrame *out;
239 ThreadData td;
240 int i, ret;
241
242 for (i = 0; i < s->nb_inputs; i++) {
243 if ((ret = ff_framesync_get_frame(&s->fs, i, &in[i], 0)) < 0)
244 return ret;
245 }
246
247 if (ctx->is_disabled) {
248 out = av_frame_clone(in[0]);
249 } else {
250 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
251 }
252 if (!out)
253 return AVERROR(ENOMEM);
254 out->pts = av_rescale_q(s->fs.pts, s->fs.time_base, outlink->time_base);
255
256 if (!ctx->is_disabled) {
257 td.in = in;
258 td.out = out;
259 ctx->internal->execute(ctx, s->median_frames, &td, NULL, FFMIN(s->height[1], ff_filter_get_nb_threads(ctx)));
260 }
261
262 return ff_filter_frame(outlink, out);
263 }
264
config_output(AVFilterLink * outlink)265 static int config_output(AVFilterLink *outlink)
266 {
267 AVFilterContext *ctx = outlink->src;
268 XMedianContext *s = ctx->priv;
269 AVRational frame_rate = ctx->inputs[0]->frame_rate;
270 AVRational sar = ctx->inputs[0]->sample_aspect_ratio;
271 AVFilterLink *inlink = ctx->inputs[0];
272 int height = ctx->inputs[0]->h;
273 int width = ctx->inputs[0]->w;
274 FFFrameSyncIn *in;
275 int i, ret;
276
277 for (int i = 1; i < s->nb_inputs && !s->tmedian; i++) {
278 if (ctx->inputs[i]->h != height || ctx->inputs[i]->w != width) {
279 av_log(ctx, AV_LOG_ERROR, "Input %d size (%dx%d) does not match input %d size (%dx%d).\n", i, ctx->inputs[i]->w, ctx->inputs[i]->h, 0, width, height);
280 return AVERROR(EINVAL);
281 }
282 }
283
284 s->desc = av_pix_fmt_desc_get(outlink->format);
285 if (!s->desc)
286 return AVERROR_BUG;
287 s->nb_planes = av_pix_fmt_count_planes(outlink->format);
288 s->depth = s->desc->comp[0].depth;
289 s->max = (1 << s->depth) - 1;
290
291 if (s->depth <= 8)
292 s->median_frames = median_frames8;
293 else
294 s->median_frames = median_frames16;
295
296 if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
297 return ret;
298
299 s->width[1] = s->width[2] = AV_CEIL_RSHIFT(inlink->w, s->desc->log2_chroma_w);
300 s->width[0] = s->width[3] = inlink->w;
301 s->height[1] = s->height[2] = AV_CEIL_RSHIFT(inlink->h, s->desc->log2_chroma_h);
302 s->height[0] = s->height[3] = inlink->h;
303
304 if (s->tmedian)
305 return 0;
306
307 outlink->w = width;
308 outlink->h = height;
309 outlink->frame_rate = frame_rate;
310 outlink->sample_aspect_ratio = sar;
311
312 if ((ret = ff_framesync_init(&s->fs, ctx, s->nb_inputs)) < 0)
313 return ret;
314
315 in = s->fs.in;
316 s->fs.opaque = s;
317 s->fs.on_event = process_frame;
318
319 for (i = 0; i < s->nb_inputs; i++) {
320 AVFilterLink *inlink = ctx->inputs[i];
321
322 in[i].time_base = inlink->time_base;
323 in[i].sync = 1;
324 in[i].before = EXT_STOP;
325 in[i].after = EXT_INFINITY;
326 }
327
328 ret = ff_framesync_configure(&s->fs);
329 outlink->time_base = s->fs.time_base;
330
331 return ret;
332 }
333
uninit(AVFilterContext * ctx)334 static av_cold void uninit(AVFilterContext *ctx)
335 {
336 XMedianContext *s = ctx->priv;
337
338 ff_framesync_uninit(&s->fs);
339
340 for (int i = 0; i < ctx->nb_inputs && !s->tmedian; i++)
341 av_freep(&ctx->input_pads[i].name);
342 for (int i = 0; i < s->nb_frames && s->frames && s->tmedian; i++)
343 av_frame_free(&s->frames[i]);
344 av_freep(&s->frames);
345 }
346
activate(AVFilterContext * ctx)347 static int activate(AVFilterContext *ctx)
348 {
349 XMedianContext *s = ctx->priv;
350 return ff_framesync_activate(&s->fs);
351 }
352
process_command(AVFilterContext * ctx,const char * cmd,const char * args,char * res,int res_len,int flags)353 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
354 char *res, int res_len, int flags)
355 {
356 XMedianContext *s = ctx->priv;
357 int ret;
358
359 ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
360 if (ret < 0)
361 return ret;
362
363 if (s->nb_inputs & 1)
364 s->index = s->radius * 2.f * s->percentile;
365 else
366 s->index = av_clip(s->radius * 2.f * s->percentile, 1, s->nb_inputs - 1);
367
368 return 0;
369 }
370
371 #define OFFSET(x) offsetof(XMedianContext, x)
372 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_FILTERING_PARAM
373 #define TFLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_RUNTIME_PARAM
374
375 static const AVOption xmedian_options[] = {
376 { "inputs", "set number of inputs", OFFSET(nb_inputs), AV_OPT_TYPE_INT, {.i64=3}, 3, 255, .flags = FLAGS },
377 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=15}, 0, 15, .flags =TFLAGS },
378 { "percentile", "set percentile", OFFSET(percentile),AV_OPT_TYPE_FLOAT,{.dbl=0.5}, 0, 1, .flags =TFLAGS },
379 { NULL },
380 };
381
382 static const AVFilterPad outputs[] = {
383 {
384 .name = "default",
385 .type = AVMEDIA_TYPE_VIDEO,
386 .config_props = config_output,
387 },
388 { NULL }
389 };
390
391 #if CONFIG_XMEDIAN_FILTER
392 FRAMESYNC_DEFINE_CLASS(xmedian, XMedianContext, fs);
393
394 AVFilter ff_vf_xmedian = {
395 .name = "xmedian",
396 .description = NULL_IF_CONFIG_SMALL("Pick median pixels from several video inputs."),
397 .priv_size = sizeof(XMedianContext),
398 .priv_class = &xmedian_class,
399 .query_formats = query_formats,
400 .outputs = outputs,
401 .preinit = xmedian_framesync_preinit,
402 .init = init,
403 .uninit = uninit,
404 .activate = activate,
405 .flags = AVFILTER_FLAG_DYNAMIC_INPUTS | AVFILTER_FLAG_SLICE_THREADS |
406 AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
407 .process_command = process_command,
408 };
409
410 #endif /* CONFIG_XMEDIAN_FILTER */
411
412 #if CONFIG_TMEDIAN_FILTER
tmedian_filter_frame(AVFilterLink * inlink,AVFrame * in)413 static int tmedian_filter_frame(AVFilterLink *inlink, AVFrame *in)
414 {
415 AVFilterContext *ctx = inlink->dst;
416 AVFilterLink *outlink = ctx->outputs[0];
417 XMedianContext *s = ctx->priv;
418 ThreadData td;
419 AVFrame *out;
420
421 if (s->nb_frames < s->nb_inputs) {
422 s->frames[s->nb_frames] = in;
423 s->nb_frames++;
424 if (s->nb_frames < s->nb_inputs)
425 return 0;
426 } else {
427 av_frame_free(&s->frames[0]);
428 memmove(&s->frames[0], &s->frames[1], sizeof(*s->frames) * (s->nb_inputs - 1));
429 s->frames[s->nb_inputs - 1] = in;
430 }
431
432 if (ctx->is_disabled) {
433 out = av_frame_clone(s->frames[0]);
434 if (!out)
435 return AVERROR(ENOMEM);
436 return ff_filter_frame(outlink, out);
437 }
438
439 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
440 if (!out)
441 return AVERROR(ENOMEM);
442 out->pts = s->frames[0]->pts;
443
444 td.out = out;
445 td.in = s->frames;
446 ctx->internal->execute(ctx, s->median_frames, &td, NULL, FFMIN(s->height[0], ff_filter_get_nb_threads(ctx)));
447
448 return ff_filter_frame(outlink, out);
449 }
450
451 static const AVOption tmedian_options[] = {
452 { "radius", "set median filter radius", OFFSET(radius), AV_OPT_TYPE_INT, {.i64=1}, 1, 127, .flags = FLAGS },
453 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=15}, 0, 15, .flags =TFLAGS },
454 { "percentile", "set percentile", OFFSET(percentile), AV_OPT_TYPE_FLOAT, {.dbl=0.5}, 0, 1, .flags =TFLAGS },
455 { NULL },
456 };
457
458 static const AVFilterPad tmedian_inputs[] = {
459 {
460 .name = "default",
461 .type = AVMEDIA_TYPE_VIDEO,
462 .filter_frame = tmedian_filter_frame,
463 },
464 { NULL }
465 };
466
467 static const AVFilterPad tmedian_outputs[] = {
468 {
469 .name = "default",
470 .type = AVMEDIA_TYPE_VIDEO,
471 .config_props = config_output,
472 },
473 { NULL }
474 };
475
476 AVFILTER_DEFINE_CLASS(tmedian);
477
478 AVFilter ff_vf_tmedian = {
479 .name = "tmedian",
480 .description = NULL_IF_CONFIG_SMALL("Pick median pixels from successive frames."),
481 .priv_size = sizeof(XMedianContext),
482 .priv_class = &tmedian_class,
483 .query_formats = query_formats,
484 .inputs = tmedian_inputs,
485 .outputs = tmedian_outputs,
486 .init = init,
487 .uninit = uninit,
488 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
489 .process_command = process_command,
490 };
491
492 #endif /* CONFIG_TMEDIAN_FILTER */
493