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
30 #include "libavutil/common.h"
31 #include "libavutil/eval.h"
32 #include "libavutil/avstring.h"
33 #include "libavutil/opt.h"
34 #include "libavutil/pixdesc.h"
35 #include "libavutil/imgutils.h"
36 #include "libavutil/mathematics.h"
37 #include "libavutil/opt.h"
39 #include "dualinput.h"
40 #include "drawutils.h"
43 static const char *const var_names[] = {
44 "main_w", "W", ///< width of the main video
45 "main_h", "H", ///< height of the main video
46 "overlay_w", "w", ///< width of the overlay video
47 "overlay_h", "h", ///< height of the overlay video
52 "n", ///< number of frame
53 "pos", ///< position in the file
54 "t", ///< timestamp expressed in seconds
61 VAR_OVERLAY_W, VAR_OW,
62 VAR_OVERLAY_H, VAR_OH,
87 int x, y; ///< position of overlayed picture
90 uint8_t main_is_packed_rgb;
91 uint8_t main_rgba_map[4];
92 uint8_t main_has_alpha;
93 uint8_t overlay_is_packed_rgb;
94 uint8_t overlay_rgba_map[4];
95 uint8_t overlay_has_alpha;
96 enum OverlayFormat { OVERLAY_FORMAT_YUV420, OVERLAY_FORMAT_YUV444, OVERLAY_FORMAT_RGB, OVERLAY_FORMAT_NB} format;
97 enum EvalMode { EVAL_MODE_INIT, EVAL_MODE_FRAME, EVAL_MODE_NB } eval_mode;
99 FFDualInputContext dinput;
101 int main_pix_step[4]; ///< steps per pixel for each plane of the main output
102 int overlay_pix_step[4]; ///< steps per pixel for each plane of the overlay
103 int hsub, vsub; ///< chroma subsampling values
105 double var_values[VAR_VARS_NB];
106 char *x_expr, *y_expr;
107 AVExpr *x_pexpr, *y_pexpr;
110 static av_cold void uninit(AVFilterContext *ctx)
112 OverlayContext *s = ctx->priv;
114 ff_dualinput_uninit(&s->dinput);
115 av_expr_free(s->x_pexpr); s->x_pexpr = NULL;
116 av_expr_free(s->y_pexpr); s->y_pexpr = NULL;
119 static inline int normalize_xy(double d, int chroma_sub)
123 return (int)d & ~((1 << chroma_sub) - 1);
126 static void eval_expr(AVFilterContext *ctx)
128 OverlayContext *s = ctx->priv;
130 s->var_values[VAR_X] = av_expr_eval(s->x_pexpr, s->var_values, NULL);
131 s->var_values[VAR_Y] = av_expr_eval(s->y_pexpr, s->var_values, NULL);
132 s->var_values[VAR_X] = av_expr_eval(s->x_pexpr, s->var_values, NULL);
133 s->x = normalize_xy(s->var_values[VAR_X], s->hsub);
134 s->y = normalize_xy(s->var_values[VAR_Y], s->vsub);
137 static int set_expr(AVExpr **pexpr, const char *expr, const char *option, void *log_ctx)
144 ret = av_expr_parse(pexpr, expr, var_names,
145 NULL, NULL, NULL, NULL, 0, log_ctx);
147 av_log(log_ctx, AV_LOG_ERROR,
148 "Error when evaluating the expression '%s' for %s\n",
158 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
159 char *res, int res_len, int flags)
161 OverlayContext *s = ctx->priv;
164 if (!strcmp(cmd, "x"))
165 ret = set_expr(&s->x_pexpr, args, cmd, ctx);
166 else if (!strcmp(cmd, "y"))
167 ret = set_expr(&s->y_pexpr, args, cmd, ctx);
169 ret = AVERROR(ENOSYS);
174 if (s->eval_mode == EVAL_MODE_INIT) {
176 av_log(ctx, AV_LOG_VERBOSE, "x:%f xi:%d y:%f yi:%d\n",
177 s->var_values[VAR_X], s->x,
178 s->var_values[VAR_Y], s->y);
183 static int query_formats(AVFilterContext *ctx)
185 OverlayContext *s = ctx->priv;
187 /* overlay formats contains alpha, for avoiding conversion with alpha information loss */
188 static const enum AVPixelFormat main_pix_fmts_yuv420[] = {
189 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_NONE
191 static const enum AVPixelFormat overlay_pix_fmts_yuv420[] = {
192 AV_PIX_FMT_YUVA420P, AV_PIX_FMT_NONE
195 static const enum AVPixelFormat main_pix_fmts_yuv444[] = {
196 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUVA444P, AV_PIX_FMT_NONE
198 static const enum AVPixelFormat overlay_pix_fmts_yuv444[] = {
199 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_NONE
202 static const enum AVPixelFormat main_pix_fmts_rgb[] = {
203 AV_PIX_FMT_ARGB, AV_PIX_FMT_RGBA,
204 AV_PIX_FMT_ABGR, AV_PIX_FMT_BGRA,
205 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
208 static const enum AVPixelFormat overlay_pix_fmts_rgb[] = {
209 AV_PIX_FMT_ARGB, AV_PIX_FMT_RGBA,
210 AV_PIX_FMT_ABGR, AV_PIX_FMT_BGRA,
214 AVFilterFormats *main_formats;
215 AVFilterFormats *overlay_formats;
218 case OVERLAY_FORMAT_YUV420:
219 main_formats = ff_make_format_list(main_pix_fmts_yuv420);
220 overlay_formats = ff_make_format_list(overlay_pix_fmts_yuv420);
222 case OVERLAY_FORMAT_YUV444:
223 main_formats = ff_make_format_list(main_pix_fmts_yuv444);
224 overlay_formats = ff_make_format_list(overlay_pix_fmts_yuv444);
226 case OVERLAY_FORMAT_RGB:
227 main_formats = ff_make_format_list(main_pix_fmts_rgb);
228 overlay_formats = ff_make_format_list(overlay_pix_fmts_rgb);
234 ff_formats_ref(main_formats, &ctx->inputs [MAIN ]->out_formats);
235 ff_formats_ref(overlay_formats, &ctx->inputs [OVERLAY]->out_formats);
236 ff_formats_ref(main_formats, &ctx->outputs[MAIN ]->in_formats );
241 static const enum AVPixelFormat alpha_pix_fmts[] = {
242 AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVA444P,
243 AV_PIX_FMT_ARGB, AV_PIX_FMT_ABGR, AV_PIX_FMT_RGBA,
244 AV_PIX_FMT_BGRA, AV_PIX_FMT_NONE
247 static int config_input_main(AVFilterLink *inlink)
249 OverlayContext *s = inlink->dst->priv;
250 const AVPixFmtDescriptor *pix_desc = av_pix_fmt_desc_get(inlink->format);
252 av_image_fill_max_pixsteps(s->main_pix_step, NULL, pix_desc);
254 s->hsub = pix_desc->log2_chroma_w;
255 s->vsub = pix_desc->log2_chroma_h;
257 s->main_is_packed_rgb =
258 ff_fill_rgba_map(s->main_rgba_map, inlink->format) >= 0;
259 s->main_has_alpha = ff_fmt_is_in(inlink->format, alpha_pix_fmts);
263 static int config_input_overlay(AVFilterLink *inlink)
265 AVFilterContext *ctx = inlink->dst;
266 OverlayContext *s = inlink->dst->priv;
268 const AVPixFmtDescriptor *pix_desc = av_pix_fmt_desc_get(inlink->format);
270 av_image_fill_max_pixsteps(s->overlay_pix_step, NULL, pix_desc);
272 /* Finish the configuration by evaluating the expressions
273 now when both inputs are configured. */
274 s->var_values[VAR_MAIN_W ] = s->var_values[VAR_MW] = ctx->inputs[MAIN ]->w;
275 s->var_values[VAR_MAIN_H ] = s->var_values[VAR_MH] = ctx->inputs[MAIN ]->h;
276 s->var_values[VAR_OVERLAY_W] = s->var_values[VAR_OW] = ctx->inputs[OVERLAY]->w;
277 s->var_values[VAR_OVERLAY_H] = s->var_values[VAR_OH] = ctx->inputs[OVERLAY]->h;
278 s->var_values[VAR_HSUB] = 1<<pix_desc->log2_chroma_w;
279 s->var_values[VAR_VSUB] = 1<<pix_desc->log2_chroma_h;
280 s->var_values[VAR_X] = NAN;
281 s->var_values[VAR_Y] = NAN;
282 s->var_values[VAR_N] = 0;
283 s->var_values[VAR_T] = NAN;
284 s->var_values[VAR_POS] = NAN;
286 if ((ret = set_expr(&s->x_pexpr, s->x_expr, "x", ctx)) < 0 ||
287 (ret = set_expr(&s->y_pexpr, s->y_expr, "y", ctx)) < 0)
290 s->overlay_is_packed_rgb =
291 ff_fill_rgba_map(s->overlay_rgba_map, inlink->format) >= 0;
292 s->overlay_has_alpha = ff_fmt_is_in(inlink->format, alpha_pix_fmts);
294 if (s->eval_mode == EVAL_MODE_INIT) {
296 av_log(ctx, AV_LOG_VERBOSE, "x:%f xi:%d y:%f yi:%d\n",
297 s->var_values[VAR_X], s->x,
298 s->var_values[VAR_Y], s->y);
301 av_log(ctx, AV_LOG_VERBOSE,
302 "main w:%d h:%d fmt:%s overlay w:%d h:%d fmt:%s\n",
303 ctx->inputs[MAIN]->w, ctx->inputs[MAIN]->h,
304 av_get_pix_fmt_name(ctx->inputs[MAIN]->format),
305 ctx->inputs[OVERLAY]->w, ctx->inputs[OVERLAY]->h,
306 av_get_pix_fmt_name(ctx->inputs[OVERLAY]->format));
310 static int config_output(AVFilterLink *outlink)
312 AVFilterContext *ctx = outlink->src;
314 outlink->w = ctx->inputs[MAIN]->w;
315 outlink->h = ctx->inputs[MAIN]->h;
316 outlink->time_base = ctx->inputs[MAIN]->time_base;
321 // divide by 255 and round to nearest
322 // apply a fast variant: (X+127)/255 = ((X+127)*257+257)>>16 = ((X+128)*257)>>16
323 #define FAST_DIV255(x) ((((x) + 128) * 257) >> 16)
325 // calculate the unpremultiplied alpha, applying the general equation:
326 // alpha = alpha_overlay / ( (alpha_main + alpha_overlay) - (alpha_main * alpha_overlay) )
327 // (((x) << 16) - ((x) << 9) + (x)) is a faster version of: 255 * 255 * x
328 // ((((x) + (y)) << 8) - ((x) + (y)) - (y) * (x)) is a faster version of: 255 * (x + y)
329 #define UNPREMULTIPLY_ALPHA(x, y) ((((x) << 16) - ((x) << 9) + (x)) / ((((x) + (y)) << 8) - ((x) + (y)) - (y) * (x)))
332 * Blend image in src to destination buffer dst at position (x, y).
334 static void blend_image(AVFilterContext *ctx,
335 AVFrame *dst, const AVFrame *src,
338 OverlayContext *s = ctx->priv;
339 int i, imax, j, jmax, k, kmax;
340 const int src_w = src->width;
341 const int src_h = src->height;
342 const int dst_w = dst->width;
343 const int dst_h = dst->height;
345 if (x >= dst_w || x+dst_w < 0 ||
346 y >= dst_h || y+dst_h < 0)
347 return; /* no intersection */
349 if (s->main_is_packed_rgb) {
350 uint8_t alpha; ///< the amount of overlay to blend on to main
351 const int dr = s->main_rgba_map[R];
352 const int dg = s->main_rgba_map[G];
353 const int db = s->main_rgba_map[B];
354 const int da = s->main_rgba_map[A];
355 const int dstep = s->main_pix_step[0];
356 const int sr = s->overlay_rgba_map[R];
357 const int sg = s->overlay_rgba_map[G];
358 const int sb = s->overlay_rgba_map[B];
359 const int sa = s->overlay_rgba_map[A];
360 const int sstep = s->overlay_pix_step[0];
361 const int main_has_alpha = s->main_has_alpha;
362 uint8_t *s, *sp, *d, *dp;
365 sp = src->data[0] + i * src->linesize[0];
366 dp = dst->data[0] + (y+i) * dst->linesize[0];
368 for (imax = FFMIN(-y + dst_h, src_h); i < imax; i++) {
371 d = dp + (x+j) * dstep;
373 for (jmax = FFMIN(-x + dst_w, src_w); j < jmax; j++) {
376 // if the main channel has an alpha channel, alpha has to be calculated
377 // to create an un-premultiplied (straight) alpha value
378 if (main_has_alpha && alpha != 0 && alpha != 255) {
379 uint8_t alpha_d = d[da];
380 alpha = UNPREMULTIPLY_ALPHA(alpha, alpha_d);
392 // main_value = main_value * (1 - alpha) + overlay_value * alpha
393 // since alpha is in the range 0-255, the result must divided by 255
394 d[dr] = FAST_DIV255(d[dr] * (255 - alpha) + s[sr] * alpha);
395 d[dg] = FAST_DIV255(d[dg] * (255 - alpha) + s[sg] * alpha);
396 d[db] = FAST_DIV255(d[db] * (255 - alpha) + s[sb] * alpha);
398 if (main_has_alpha) {
406 // apply alpha compositing: main_alpha += (1-main_alpha) * overlay_alpha
407 d[da] += FAST_DIV255((255 - d[da]) * s[sa]);
413 dp += dst->linesize[0];
414 sp += src->linesize[0];
417 const int main_has_alpha = s->main_has_alpha;
418 if (main_has_alpha) {
419 uint8_t alpha; ///< the amount of overlay to blend on to main
420 uint8_t *s, *sa, *d, *da;
423 sa = src->data[3] + i * src->linesize[3];
424 da = dst->data[3] + (y+i) * dst->linesize[3];
426 for (imax = FFMIN(-y + dst_h, src_h); i < imax; i++) {
431 for (jmax = FFMIN(-x + dst_w, src_w); j < jmax; j++) {
433 if (alpha != 0 && alpha != 255) {
434 uint8_t alpha_d = *d;
435 alpha = UNPREMULTIPLY_ALPHA(alpha, alpha_d);
444 // apply alpha compositing: main_alpha += (1-main_alpha) * overlay_alpha
445 *d += FAST_DIV255((255 - *d) * *s);
450 da += dst->linesize[3];
451 sa += src->linesize[3];
454 for (i = 0; i < 3; i++) {
455 int hsub = i ? s->hsub : 0;
456 int vsub = i ? s->vsub : 0;
457 int src_wp = FF_CEIL_RSHIFT(src_w, hsub);
458 int src_hp = FF_CEIL_RSHIFT(src_h, vsub);
459 int dst_wp = FF_CEIL_RSHIFT(dst_w, hsub);
460 int dst_hp = FF_CEIL_RSHIFT(dst_h, vsub);
463 uint8_t *s, *sp, *d, *dp, *a, *ap;
466 sp = src->data[i] + j * src->linesize[i];
467 dp = dst->data[i] + (yp+j) * dst->linesize[i];
468 ap = src->data[3] + (j<<vsub) * src->linesize[3];
470 for (jmax = FFMIN(-yp + dst_hp, src_hp); j < jmax; j++) {
476 for (kmax = FFMIN(-xp + dst_wp, src_wp); k < kmax; k++) {
477 int alpha_v, alpha_h, alpha;
479 // average alpha for color components, improve quality
480 if (hsub && vsub && j+1 < src_hp && k+1 < src_wp) {
481 alpha = (a[0] + a[src->linesize[3]] +
482 a[1] + a[src->linesize[3]+1]) >> 2;
483 } else if (hsub || vsub) {
484 alpha_h = hsub && k+1 < src_wp ?
485 (a[0] + a[1]) >> 1 : a[0];
486 alpha_v = vsub && j+1 < src_hp ?
487 (a[0] + a[src->linesize[3]]) >> 1 : a[0];
488 alpha = (alpha_v + alpha_h) >> 1;
491 // if the main channel has an alpha channel, alpha has to be calculated
492 // to create an un-premultiplied (straight) alpha value
493 if (main_has_alpha && alpha != 0 && alpha != 255) {
494 // average alpha for color components, improve quality
496 if (hsub && vsub && j+1 < src_hp && k+1 < src_wp) {
497 alpha_d = (d[0] + d[src->linesize[3]] +
498 d[1] + d[src->linesize[3]+1]) >> 2;
499 } else if (hsub || vsub) {
500 alpha_h = hsub && k+1 < src_wp ?
501 (d[0] + d[1]) >> 1 : d[0];
502 alpha_v = vsub && j+1 < src_hp ?
503 (d[0] + d[src->linesize[3]]) >> 1 : d[0];
504 alpha_d = (alpha_v + alpha_h) >> 1;
507 alpha = UNPREMULTIPLY_ALPHA(alpha, alpha_d);
509 *d = FAST_DIV255(*d * (255 - alpha) + *s * alpha);
514 dp += dst->linesize[i];
515 sp += src->linesize[i];
516 ap += (1 << vsub) * src->linesize[3];
522 static AVFrame *do_blend(AVFilterContext *ctx, AVFrame *mainpic,
523 const AVFrame *second)
525 OverlayContext *s = ctx->priv;
526 AVFilterLink *inlink = ctx->inputs[0];
529 if (s->eval_mode == EVAL_MODE_FRAME) {
530 int64_t pos = av_frame_get_pkt_pos(mainpic);
532 s->var_values[VAR_N] = inlink->frame_count;
533 s->var_values[VAR_T] = mainpic->pts == AV_NOPTS_VALUE ?
534 NAN : mainpic->pts * av_q2d(inlink->time_base);
535 s->var_values[VAR_POS] = pos == -1 ? NAN : pos;
538 av_log(ctx, AV_LOG_DEBUG, "n:%f t:%f pos:%f x:%f xi:%d y:%f yi:%d\n",
539 s->var_values[VAR_N], s->var_values[VAR_T], s->var_values[VAR_POS],
540 s->var_values[VAR_X], s->x,
541 s->var_values[VAR_Y], s->y);
544 blend_image(ctx, mainpic, second, s->x, s->y);
548 static int filter_frame_main(AVFilterLink *inlink, AVFrame *inpicref)
550 OverlayContext *s = inlink->dst->priv;
551 return ff_dualinput_filter_frame_main(&s->dinput, inlink, inpicref);
554 static int filter_frame_over(AVFilterLink *inlink, AVFrame *inpicref)
556 OverlayContext *s = inlink->dst->priv;
557 return ff_dualinput_filter_frame_second(&s->dinput, inlink, inpicref);
560 static int request_frame(AVFilterLink *outlink)
562 OverlayContext *s = outlink->src->priv;
563 return ff_dualinput_request_frame(&s->dinput, outlink);
566 static av_cold int init(AVFilterContext *ctx)
568 OverlayContext *s = ctx->priv;
570 if (s->allow_packed_rgb) {
571 av_log(ctx, AV_LOG_WARNING,
572 "The rgb option is deprecated and is overriding the format option, use format instead\n");
573 s->format = OVERLAY_FORMAT_RGB;
575 s->dinput.process = do_blend;
579 #define OFFSET(x) offsetof(OverlayContext, x)
580 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
582 static const AVOption overlay_options[] = {
583 { "x", "set the x expression", OFFSET(x_expr), AV_OPT_TYPE_STRING, {.str = "0"}, CHAR_MIN, CHAR_MAX, FLAGS },
584 { "y", "set the y expression", OFFSET(y_expr), AV_OPT_TYPE_STRING, {.str = "0"}, CHAR_MIN, CHAR_MAX, FLAGS },
585 { "eval", "specify when to evaluate expressions", OFFSET(eval_mode), AV_OPT_TYPE_INT, {.i64 = EVAL_MODE_FRAME}, 0, EVAL_MODE_NB-1, FLAGS, "eval" },
586 { "init", "eval expressions once during initialization", 0, AV_OPT_TYPE_CONST, {.i64=EVAL_MODE_INIT}, .flags = FLAGS, .unit = "eval" },
587 { "frame", "eval expressions per-frame", 0, AV_OPT_TYPE_CONST, {.i64=EVAL_MODE_FRAME}, .flags = FLAGS, .unit = "eval" },
588 { "rgb", "force packed RGB in input and output (deprecated)", OFFSET(allow_packed_rgb), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, FLAGS },
589 { "shortest", "force termination when the shortest input terminates", OFFSET(dinput.shortest), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS },
590 { "format", "set output format", OFFSET(format), AV_OPT_TYPE_INT, {.i64=OVERLAY_FORMAT_YUV420}, 0, OVERLAY_FORMAT_NB-1, FLAGS, "format" },
591 { "yuv420", "", 0, AV_OPT_TYPE_CONST, {.i64=OVERLAY_FORMAT_YUV420}, .flags = FLAGS, .unit = "format" },
592 { "yuv444", "", 0, AV_OPT_TYPE_CONST, {.i64=OVERLAY_FORMAT_YUV444}, .flags = FLAGS, .unit = "format" },
593 { "rgb", "", 0, AV_OPT_TYPE_CONST, {.i64=OVERLAY_FORMAT_RGB}, .flags = FLAGS, .unit = "format" },
594 { "repeatlast", "repeat overlay of the last overlay frame", OFFSET(dinput.repeatlast), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, FLAGS },
598 AVFILTER_DEFINE_CLASS(overlay);
600 static const AVFilterPad avfilter_vf_overlay_inputs[] = {
603 .type = AVMEDIA_TYPE_VIDEO,
604 .get_video_buffer = ff_null_get_video_buffer,
605 .config_props = config_input_main,
606 .filter_frame = filter_frame_main,
611 .type = AVMEDIA_TYPE_VIDEO,
612 .config_props = config_input_overlay,
613 .filter_frame = filter_frame_over,
618 static const AVFilterPad avfilter_vf_overlay_outputs[] = {
621 .type = AVMEDIA_TYPE_VIDEO,
622 .config_props = config_output,
623 .request_frame = request_frame,
628 AVFilter avfilter_vf_overlay = {
630 .description = NULL_IF_CONFIG_SMALL("Overlay a video source on top of the input."),
635 .priv_size = sizeof(OverlayContext),
636 .priv_class = &overlay_class,
638 .query_formats = query_formats,
639 .process_command = process_command,
641 .inputs = avfilter_vf_overlay_inputs,
642 .outputs = avfilter_vf_overlay_outputs,
643 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,