2 * Copyright (c) 2010 Stefano Sabatini
3 * Copyright (c) 2010 Baptiste Coudurier
4 * Copyright (c) 2007 Bobby Bingham
6 * This file is part of FFmpeg.
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.
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.
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
25 * overlay one video on top of another
32 #include "libavutil/common.h"
33 #include "libavutil/eval.h"
34 #include "libavutil/avstring.h"
35 #include "libavutil/opt.h"
36 #include "libavutil/pixdesc.h"
37 #include "libavutil/imgutils.h"
38 #include "libavutil/mathematics.h"
39 #include "libavutil/timestamp.h"
41 #include "bufferqueue.h"
42 #include "drawutils.h"
45 static const char *const var_names[] = {
46 "main_w", "W", ///< width of the main video
47 "main_h", "H", ///< height of the main video
48 "overlay_w", "w", ///< width of the overlay video
49 "overlay_h", "h", ///< height of the overlay video
56 VAR_OVERLAY_W, VAR_OW,
57 VAR_OVERLAY_H, VAR_OH,
75 int x, y; ///< position of overlayed picture
78 uint8_t frame_requested;
80 uint8_t main_is_packed_rgb;
81 uint8_t main_rgba_map[4];
82 uint8_t main_has_alpha;
83 uint8_t overlay_is_packed_rgb;
84 uint8_t overlay_rgba_map[4];
85 uint8_t overlay_has_alpha;
87 AVFilterBufferRef *overpicref;
88 struct FFBufQueue queue_main;
89 struct FFBufQueue queue_over;
91 int main_pix_step[4]; ///< steps per pixel for each plane of the main output
92 int overlay_pix_step[4]; ///< steps per pixel for each plane of the overlay
93 int hsub, vsub; ///< chroma subsampling values
95 char *x_expr, *y_expr;
98 #define OFFSET(x) offsetof(OverlayContext, x)
99 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
101 static const AVOption overlay_options[] = {
102 { "x", "set the x expression", OFFSET(x_expr), AV_OPT_TYPE_STRING, {.str = "0"}, CHAR_MIN, CHAR_MAX, FLAGS },
103 { "y", "set the y expression", OFFSET(y_expr), AV_OPT_TYPE_STRING, {.str = "0"}, CHAR_MIN, CHAR_MAX, FLAGS },
104 {"rgb", "force packed RGB in input and output", OFFSET(allow_packed_rgb), AV_OPT_TYPE_INT, {.dbl=0}, 0, 1, FLAGS },
108 AVFILTER_DEFINE_CLASS(overlay);
110 static av_cold int init(AVFilterContext *ctx, const char *args)
112 OverlayContext *over = ctx->priv;
113 char *args1 = av_strdup(args);
114 char *expr, *bufptr = NULL;
117 over->class = &overlay_class;
118 av_opt_set_defaults(over);
120 if (expr = av_strtok(args1, ":", &bufptr)) {
121 av_free(over->x_expr);
122 if (!(over->x_expr = av_strdup(expr))) {
123 ret = AVERROR(ENOMEM);
127 if (expr = av_strtok(NULL, ":", &bufptr)) {
128 av_free(over->y_expr);
129 if (!(over->y_expr = av_strdup(expr))) {
130 ret = AVERROR(ENOMEM);
135 if (bufptr && (ret = av_set_options_string(over, bufptr, "=", ":")) < 0)
143 static av_cold void uninit(AVFilterContext *ctx)
145 OverlayContext *over = ctx->priv;
147 av_freep(&over->x_expr);
148 av_freep(&over->y_expr);
150 avfilter_unref_bufferp(&over->overpicref);
151 ff_bufqueue_discard_all(&over->queue_main);
152 ff_bufqueue_discard_all(&over->queue_over);
155 static int query_formats(AVFilterContext *ctx)
157 OverlayContext *over = ctx->priv;
159 /* overlay formats contains alpha, for avoiding conversion with alpha information loss */
160 const enum PixelFormat main_pix_fmts_yuv[] = { PIX_FMT_YUV420P, PIX_FMT_NONE };
161 const enum PixelFormat overlay_pix_fmts_yuv[] = { PIX_FMT_YUVA420P, PIX_FMT_NONE };
162 const enum PixelFormat main_pix_fmts_rgb[] = {
163 PIX_FMT_ARGB, PIX_FMT_RGBA,
164 PIX_FMT_ABGR, PIX_FMT_BGRA,
165 PIX_FMT_RGB24, PIX_FMT_BGR24,
168 const enum PixelFormat overlay_pix_fmts_rgb[] = {
169 PIX_FMT_ARGB, PIX_FMT_RGBA,
170 PIX_FMT_ABGR, PIX_FMT_BGRA,
174 AVFilterFormats *main_formats;
175 AVFilterFormats *overlay_formats;
177 if (over->allow_packed_rgb) {
178 main_formats = ff_make_format_list(main_pix_fmts_rgb);
179 overlay_formats = ff_make_format_list(overlay_pix_fmts_rgb);
181 main_formats = ff_make_format_list(main_pix_fmts_yuv);
182 overlay_formats = ff_make_format_list(overlay_pix_fmts_yuv);
185 ff_formats_ref(main_formats, &ctx->inputs [MAIN ]->out_formats);
186 ff_formats_ref(overlay_formats, &ctx->inputs [OVERLAY]->out_formats);
187 ff_formats_ref(main_formats, &ctx->outputs[MAIN ]->in_formats );
192 static const enum PixelFormat alpha_pix_fmts[] = {
193 PIX_FMT_YUVA420P, PIX_FMT_ARGB, PIX_FMT_ABGR, PIX_FMT_RGBA,
194 PIX_FMT_BGRA, PIX_FMT_NONE
197 static int config_input_main(AVFilterLink *inlink)
199 OverlayContext *over = inlink->dst->priv;
200 const AVPixFmtDescriptor *pix_desc = &av_pix_fmt_descriptors[inlink->format];
202 av_image_fill_max_pixsteps(over->main_pix_step, NULL, pix_desc);
204 over->hsub = pix_desc->log2_chroma_w;
205 over->vsub = pix_desc->log2_chroma_h;
207 over->main_is_packed_rgb =
208 ff_fill_rgba_map(over->main_rgba_map, inlink->format) >= 0;
209 over->main_has_alpha = ff_fmt_is_in(inlink->format, alpha_pix_fmts);
213 static int config_input_overlay(AVFilterLink *inlink)
215 AVFilterContext *ctx = inlink->dst;
216 OverlayContext *over = inlink->dst->priv;
218 double var_values[VAR_VARS_NB], res;
220 const AVPixFmtDescriptor *pix_desc = &av_pix_fmt_descriptors[inlink->format];
222 av_image_fill_max_pixsteps(over->overlay_pix_step, NULL, pix_desc);
224 /* Finish the configuration by evaluating the expressions
225 now when both inputs are configured. */
226 var_values[VAR_MAIN_W ] = var_values[VAR_MW] = ctx->inputs[MAIN ]->w;
227 var_values[VAR_MAIN_H ] = var_values[VAR_MH] = ctx->inputs[MAIN ]->h;
228 var_values[VAR_OVERLAY_W] = var_values[VAR_OW] = ctx->inputs[OVERLAY]->w;
229 var_values[VAR_OVERLAY_H] = var_values[VAR_OH] = ctx->inputs[OVERLAY]->h;
231 if ((ret = av_expr_parse_and_eval(&res, (expr = over->x_expr), var_names, var_values,
232 NULL, NULL, NULL, NULL, NULL, 0, ctx)) < 0)
235 if ((ret = av_expr_parse_and_eval(&res, (expr = over->y_expr), var_names, var_values,
236 NULL, NULL, NULL, NULL, NULL, 0, ctx)))
239 /* x may depend on y */
240 if ((ret = av_expr_parse_and_eval(&res, (expr = over->x_expr), var_names, var_values,
241 NULL, NULL, NULL, NULL, NULL, 0, ctx)) < 0)
245 over->overlay_is_packed_rgb =
246 ff_fill_rgba_map(over->overlay_rgba_map, inlink->format) >= 0;
247 over->overlay_has_alpha = ff_fmt_is_in(inlink->format, alpha_pix_fmts);
249 av_log(ctx, AV_LOG_VERBOSE,
250 "main w:%d h:%d fmt:%s overlay x:%d y:%d w:%d h:%d fmt:%s\n",
251 ctx->inputs[MAIN]->w, ctx->inputs[MAIN]->h,
252 av_pix_fmt_descriptors[ctx->inputs[MAIN]->format].name,
254 ctx->inputs[OVERLAY]->w, ctx->inputs[OVERLAY]->h,
255 av_pix_fmt_descriptors[ctx->inputs[OVERLAY]->format].name);
257 if (over->x < 0 || over->y < 0 ||
258 over->x + var_values[VAR_OVERLAY_W] > var_values[VAR_MAIN_W] ||
259 over->y + var_values[VAR_OVERLAY_H] > var_values[VAR_MAIN_H]) {
260 av_log(ctx, AV_LOG_ERROR,
261 "Overlay area (%d,%d)<->(%d,%d) not within the main area (0,0)<->(%d,%d) or zero-sized\n",
263 (int)(over->x + var_values[VAR_OVERLAY_W]),
264 (int)(over->y + var_values[VAR_OVERLAY_H]),
265 (int)var_values[VAR_MAIN_W], (int)var_values[VAR_MAIN_H]);
266 return AVERROR(EINVAL);
271 av_log(NULL, AV_LOG_ERROR,
272 "Error when evaluating the expression '%s'\n", expr);
276 static int config_output(AVFilterLink *outlink)
278 AVFilterContext *ctx = outlink->src;
280 // common timebase computation:
281 AVRational tb1 = ctx->inputs[MAIN ]->time_base;
282 AVRational tb2 = ctx->inputs[OVERLAY]->time_base;
283 AVRational *tb = &ctx->outputs[0]->time_base;
284 exact = av_reduce(&tb->num, &tb->den,
285 av_gcd((int64_t)tb1.num * tb2.den,
286 (int64_t)tb2.num * tb1.den),
287 (int64_t)tb1.den * tb2.den, INT_MAX);
288 av_log(ctx, AV_LOG_VERBOSE,
289 "main_tb:%d/%d overlay_tb:%d/%d -> tb:%d/%d exact:%d\n",
290 tb1.num, tb1.den, tb2.num, tb2.den, tb->num, tb->den, exact);
292 av_log(ctx, AV_LOG_WARNING,
293 "Timestamp conversion inexact, timestamp information loss may occurr\n");
295 outlink->w = ctx->inputs[MAIN]->w;
296 outlink->h = ctx->inputs[MAIN]->h;
301 static AVFilterBufferRef *get_video_buffer(AVFilterLink *link, int perms, int w, int h)
303 return ff_get_video_buffer(link->dst->outputs[0], perms, w, h);
306 // divide by 255 and round to nearest
307 // apply a fast variant: (X+127)/255 = ((X+127)*257+257)>>16 = ((X+128)*257)>>16
308 #define FAST_DIV255(x) ((((x) + 128) * 257) >> 16)
310 static void blend_slice(AVFilterContext *ctx,
311 AVFilterBufferRef *dst, AVFilterBufferRef *src,
312 int x, int y, int w, int h,
313 int slice_y, int slice_w, int slice_h)
315 OverlayContext *over = ctx->priv;
318 int overlay_end_y = y+h;
319 int slice_end_y = slice_y+slice_h;
322 width = FFMIN(slice_w - x, w);
323 end_y = FFMIN(slice_end_y, overlay_end_y);
324 start_y = FFMAX(y, slice_y);
325 height = end_y - start_y;
327 if (over->main_is_packed_rgb) {
328 uint8_t *dp = dst->data[0] + x * over->main_pix_step[0] +
329 start_y * dst->linesize[0];
330 uint8_t *sp = src->data[0];
331 uint8_t alpha; ///< the amount of overlay to blend on to main
332 const int dr = over->main_rgba_map[R];
333 const int dg = over->main_rgba_map[G];
334 const int db = over->main_rgba_map[B];
335 const int da = over->main_rgba_map[A];
336 const int dstep = over->main_pix_step[0];
337 const int sr = over->overlay_rgba_map[R];
338 const int sg = over->overlay_rgba_map[G];
339 const int sb = over->overlay_rgba_map[B];
340 const int sa = over->overlay_rgba_map[A];
341 const int sstep = over->overlay_pix_step[0];
342 const int main_has_alpha = over->main_has_alpha;
344 sp += (slice_y - y) * src->linesize[0];
345 for (i = 0; i < height; i++) {
346 uint8_t *d = dp, *s = sp;
347 for (j = 0; j < width; j++) {
350 // if the main channel has an alpha channel, alpha has to be calculated
351 // to create an un-premultiplied (straight) alpha value
352 if (main_has_alpha && alpha != 0 && alpha != 255) {
353 // apply the general equation:
354 // alpha = alpha_overlay / ( (alpha_main + alpha_overlay) - (alpha_main * alpha_overlay) )
356 // the next line is a faster version of: 255 * 255 * alpha
357 ( (alpha << 16) - (alpha << 9) + alpha )
359 // the next line is a faster version of: 255 * (alpha + d[da])
360 ( ((alpha + d[da]) << 8 ) - (alpha + d[da])
373 // main_value = main_value * (1 - alpha) + overlay_value * alpha
374 // since alpha is in the range 0-255, the result must divided by 255
375 d[dr] = FAST_DIV255(d[dr] * (255 - alpha) + s[sr] * alpha);
376 d[dg] = FAST_DIV255(d[dg] * (255 - alpha) + s[sg] * alpha);
377 d[db] = FAST_DIV255(d[db] * (255 - alpha) + s[sb] * alpha);
379 if (main_has_alpha) {
387 // apply alpha compositing: main_alpha += (1-main_alpha) * overlay_alpha
388 d[da] += FAST_DIV255((255 - d[da]) * s[sa]);
394 dp += dst->linesize[0];
395 sp += src->linesize[0];
398 for (i = 0; i < 3; i++) {
399 int hsub = i ? over->hsub : 0;
400 int vsub = i ? over->vsub : 0;
401 uint8_t *dp = dst->data[i] + (x >> hsub) +
402 (start_y >> vsub) * dst->linesize[i];
403 uint8_t *sp = src->data[i];
404 uint8_t *ap = src->data[3];
405 int wp = FFALIGN(width, 1<<hsub) >> hsub;
406 int hp = FFALIGN(height, 1<<vsub) >> vsub;
408 sp += ((slice_y - y) >> vsub) * src->linesize[i];
409 ap += (slice_y - y) * src->linesize[3];
411 for (j = 0; j < hp; j++) {
412 uint8_t *d = dp, *s = sp, *a = ap;
413 for (k = 0; k < wp; k++) {
414 // average alpha for color components, improve quality
415 int alpha_v, alpha_h, alpha;
416 if (hsub && vsub && j+1 < hp && k+1 < wp) {
417 alpha = (a[0] + a[src->linesize[3]] +
418 a[1] + a[src->linesize[3]+1]) >> 2;
419 } else if (hsub || vsub) {
420 alpha_h = hsub && k+1 < wp ?
421 (a[0] + a[1]) >> 1 : a[0];
422 alpha_v = vsub && j+1 < hp ?
423 (a[0] + a[src->linesize[3]]) >> 1 : a[0];
424 alpha = (alpha_v + alpha_h) >> 1;
427 *d = FAST_DIV255(*d * (255 - alpha) + *s * alpha);
432 dp += dst->linesize[i];
433 sp += src->linesize[i];
434 ap += (1 << vsub) * src->linesize[3];
440 static int try_start_frame(AVFilterContext *ctx, AVFilterBufferRef *mainpic)
442 OverlayContext *over = ctx->priv;
443 AVFilterLink *outlink = ctx->outputs[0];
444 AVFilterBufferRef *next_overpic, *outpicref;
446 /* Discard obsolete overlay frames: if there is a next frame with pts is
447 * before the main frame, we can drop the current overlay. */
449 next_overpic = ff_bufqueue_peek(&over->queue_over, 0);
450 if (!next_overpic || next_overpic->pts > mainpic->pts)
452 ff_bufqueue_get(&over->queue_over);
453 avfilter_unref_buffer(over->overpicref);
454 over->overpicref = next_overpic;
456 /* If there is no next frame and no EOF and the overlay frame is before
457 * the main frame, we can not know yet if it will be superseded. */
458 if (!over->queue_over.available && !over->overlay_eof &&
459 (!over->overpicref || over->overpicref->pts < mainpic->pts))
460 return AVERROR(EAGAIN);
461 /* At this point, we know that the current overlay frame extends to the
462 * time of the main frame. */
463 outlink->out_buf = outpicref = avfilter_ref_buffer(mainpic, ~0);
465 av_dlog(ctx, "main_pts:%s main_pts_time:%s",
466 av_ts2str(outpicref->pts), av_ts2timestr(outpicref->pts, &outlink->time_base));
467 if (over->overpicref)
468 av_dlog(ctx, " over_pts:%s over_pts_time:%s",
469 av_ts2str(over->overpicref->pts), av_ts2timestr(over->overpicref->pts, &outlink->time_base));
472 ff_start_frame(ctx->outputs[0], avfilter_ref_buffer(outpicref, ~0));
473 over->frame_requested = 0;
477 static int try_start_next_frame(AVFilterContext *ctx)
479 OverlayContext *over = ctx->priv;
480 AVFilterBufferRef *next_mainpic = ff_bufqueue_peek(&over->queue_main, 0);
481 if (!next_mainpic || try_start_frame(ctx, next_mainpic) < 0)
482 return AVERROR(EAGAIN);
483 avfilter_unref_buffer(ff_bufqueue_get(&over->queue_main));
487 static int try_push_frame(AVFilterContext *ctx)
489 OverlayContext *over = ctx->priv;
490 AVFilterLink *outlink = ctx->outputs[0];
491 AVFilterBufferRef *outpicref;
493 if (try_start_next_frame(ctx) < 0)
494 return AVERROR(EAGAIN);
495 outpicref = outlink->out_buf;
496 if (over->overpicref)
497 blend_slice(ctx, outpicref, over->overpicref, over->x, over->y,
498 over->overpicref->video->w, over->overpicref->video->h,
499 0, outpicref->video->w, outpicref->video->h);
500 ff_draw_slice(outlink, 0, outpicref->video->h, +1);
501 ff_end_frame(outlink);
505 static void flush_frames(AVFilterContext *ctx)
507 while (!try_push_frame(ctx));
510 static int start_frame_main(AVFilterLink *inlink, AVFilterBufferRef *inpicref)
512 AVFilterContext *ctx = inlink->dst;
513 OverlayContext *over = ctx->priv;
516 inpicref->pts = av_rescale_q(inpicref->pts, ctx->inputs[MAIN]->time_base,
517 ctx->outputs[0]->time_base);
518 if (try_start_frame(ctx, inpicref) < 0) {
519 ff_bufqueue_add(ctx, &over->queue_main, inpicref);
520 av_assert1(inpicref == inlink->cur_buf);
521 inlink->cur_buf = NULL;
526 static int draw_slice_main(AVFilterLink *inlink, int y, int h, int slice_dir)
528 AVFilterContext *ctx = inlink->dst;
529 OverlayContext *over = ctx->priv;
530 AVFilterLink *outlink = ctx->outputs[0];
531 AVFilterBufferRef *outpicref = outlink->out_buf;
535 if (over->overpicref &&
536 y + h > over->y && y < over->y + over->overpicref->video->h) {
537 blend_slice(ctx, outpicref, over->overpicref, over->x, over->y,
538 over->overpicref->video->w, over->overpicref->video->h,
539 y, outpicref->video->w, h);
541 return ff_draw_slice(outlink, y, h, slice_dir);
544 static int end_frame_main(AVFilterLink *inlink)
546 AVFilterContext *ctx = inlink->dst;
547 AVFilterLink *outlink = ctx->outputs[0];
548 AVFilterBufferRef *outpicref = outlink->out_buf;
553 return ff_end_frame(ctx->outputs[0]);
556 static int start_frame_over(AVFilterLink *inlink, AVFilterBufferRef *inpicref)
561 static int end_frame_over(AVFilterLink *inlink)
563 AVFilterContext *ctx = inlink->dst;
564 OverlayContext *over = ctx->priv;
565 AVFilterBufferRef *inpicref = inlink->cur_buf;
566 inlink->cur_buf = NULL;
569 inpicref->pts = av_rescale_q(inpicref->pts, ctx->inputs[OVERLAY]->time_base,
570 ctx->outputs[0]->time_base);
571 ff_bufqueue_add(ctx, &over->queue_over, inpicref);
572 return try_push_frame(ctx);
575 static int request_frame(AVFilterLink *outlink)
577 AVFilterContext *ctx = outlink->src;
578 OverlayContext *over = ctx->priv;
581 if (!try_push_frame(ctx))
583 over->frame_requested = 1;
584 while (over->frame_requested) {
585 /* TODO if we had a frame duration, we could guess more accurately */
586 input = !over->overlay_eof && (over->queue_main.available ||
587 over->queue_over.available < 2) ?
589 ret = ff_request_frame(ctx->inputs[input]);
590 /* EOF on main is reported immediately */
591 if (ret == AVERROR_EOF && input == OVERLAY) {
592 over->overlay_eof = 1;
593 if (!try_start_next_frame(ctx))
595 ret = 0; /* continue requesting frames on main */
603 static int null_draw_slice(AVFilterLink *inlink, int y, int h, int slice_dir)
608 AVFilter avfilter_vf_overlay = {
610 .description = NULL_IF_CONFIG_SMALL("Overlay a video source on top of the input."),
615 .priv_size = sizeof(OverlayContext),
617 .query_formats = query_formats,
619 .inputs = (const AVFilterPad[]) {{ .name = "main",
620 .type = AVMEDIA_TYPE_VIDEO,
621 .get_video_buffer= get_video_buffer,
622 .config_props = config_input_main,
623 .start_frame = start_frame_main,
624 .draw_slice = draw_slice_main,
625 .end_frame = end_frame_main,
626 .min_perms = AV_PERM_READ | AV_PERM_WRITE | AV_PERM_PRESERVE },
628 .type = AVMEDIA_TYPE_VIDEO,
629 .config_props = config_input_overlay,
630 .start_frame = start_frame_over,
631 .draw_slice = null_draw_slice,
632 .end_frame = end_frame_over,
633 .min_perms = AV_PERM_READ | AV_PERM_PRESERVE },
635 .outputs = (const AVFilterPad[]) {{ .name = "default",
636 .type = AVMEDIA_TYPE_VIDEO,
637 .rej_perms = AV_PERM_WRITE,
638 .config_props = config_output,
639 .request_frame = request_frame, },
641 .priv_class = &overlay_class,