lavfi: add alphaextract and alphamerge filters
authorSteven Robertson <steven@strobe.cc>
Wed, 11 Jul 2012 05:14:57 +0000 (22:14 -0700)
committerStefano Sabatini <stefasab@gmail.com>
Sun, 22 Jul 2012 10:17:34 +0000 (12:17 +0200)
These filters are designed for storing and transmitting video sequences
with alpha using higher-efficiency codecs such as x264 which don't
natively support an alpha channel. 'alphaextract' takes an input stream
with an alpha channel and returns a video containing just the alpha
component as a grayscale value; 'alphamerge' takes an RGB or YUV stream
and adds an alpha channel recovered from a second grayscale stream.

Signed-off-by: Steven Robertson <steven@strobe.cc>
Signed-off-by: Stefano Sabatini <stefasab@gmail.com>
12 files changed:
Changelog
doc/filters.texi
libavfilter/Makefile
libavfilter/allfilters.c
libavfilter/version.h
libavfilter/vf_alphaextract.c [new file with mode: 0644]
libavfilter/vf_alphamerge.c [new file with mode: 0644]
tests/lavfi-regression.sh
tests/ref/lavfi/alphaextract_rgb [new file with mode: 0644]
tests/ref/lavfi/alphaextract_yuv [new file with mode: 0644]
tests/ref/lavfi/alphamerge_rgb [new file with mode: 0644]
tests/ref/lavfi/alphamerge_yuv [new file with mode: 0644]

index ab8e704..44e647f 100644 (file)
--- a/Changelog
+++ b/Changelog
@@ -34,6 +34,7 @@ version next:
 - ffmpeg -(no)stdin option
 - Opus decoder using libopus
 - caca output device using libcaca
+- alphaextract and alphamerge filters
 
 
 version 0.11:
index 4a6c092..2b5cc87 100644 (file)
@@ -1101,6 +1101,31 @@ build.
 
 Below is a description of the currently available video filters.
 
+@section alphaextract
+
+Extract the alpha component from the input as a grayscale video. This
+is especially useful with the @var{alphamerge} filter.
+
+@section alphamerge
+
+Add or replace the alpha component of the primary input with the
+grayscale value of a second input. This is intended for use with
+@var{alphaextract} to allow the transmission or storage of frame
+sequences that have alpha in a format that doesn't support an alpha
+channel.
+
+For example, to reconstruct full frames from a normal YUV-encoded video
+and a separate video created with @var{alphaextract}, you might use:
+@example
+movie=in_alpha.mkv [alpha]; [in][alpha] alphamerge [out]
+@end example
+
+Since this filter is designed for reconstruction, it operates on frame
+sequences without considering timestamps, and terminates when either
+input reaches end of stream. This will cause problems if your encoding
+pipeline drops frames. If you're trying to apply an image as an
+overlay to a video stream, consider the @var{overlay} filter instead.
+
 @section ass
 
 Draw ASS (Advanced Substation Alpha) subtitles on top of input video
index b094f59..a177752 100644 (file)
@@ -76,6 +76,8 @@ OBJS-$(CONFIG_ABUFFERSINK_FILTER)            += sink_buffer.o
 OBJS-$(CONFIG_ANULLSINK_FILTER)              += asink_anullsink.o
 
 OBJS-$(CONFIG_ASS_FILTER)                    += vf_ass.o
+OBJS-$(CONFIG_ALPHAEXTRACT_FILTER)           += vf_alphaextract.o
+OBJS-$(CONFIG_ALPHAMERGE_FILTER)             += vf_alphamerge.o
 OBJS-$(CONFIG_BBOX_FILTER)                   += bbox.o vf_bbox.o
 OBJS-$(CONFIG_BLACKDETECT_FILTER)            += vf_blackdetect.o
 OBJS-$(CONFIG_BLACKFRAME_FILTER)             += vf_blackframe.o
index 706405e..aad4534 100644 (file)
@@ -64,6 +64,8 @@ void avfilter_register_all(void)
     REGISTER_FILTER (ABUFFERSINK, abuffersink, asink);
     REGISTER_FILTER (ANULLSINK,   anullsink,   asink);
 
+    REGISTER_FILTER (ALPHAEXTRACT, alphaextract, vf);
+    REGISTER_FILTER (ALPHAMERGE,  alphamerge,  vf);
     REGISTER_FILTER (ASS,         ass,         vf);
     REGISTER_FILTER (BBOX,        bbox,        vf);
     REGISTER_FILTER (BLACKDETECT, blackdetect, vf);
index ea95cf3..96af8d6 100644 (file)
@@ -29,7 +29,7 @@
 #include "libavutil/avutil.h"
 
 #define LIBAVFILTER_VERSION_MAJOR  3
-#define LIBAVFILTER_VERSION_MINOR  2
+#define LIBAVFILTER_VERSION_MINOR  3
 #define LIBAVFILTER_VERSION_MICRO 100
 
 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \
diff --git a/libavfilter/vf_alphaextract.c b/libavfilter/vf_alphaextract.c
new file mode 100644 (file)
index 0000000..06cbb71
--- /dev/null
@@ -0,0 +1,117 @@
+/*
+ * Copyright (c) 2012 Steven Robertson
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * simple channel-swapping filter to get at the alpha component
+ */
+
+#include <string.h>
+
+#include "libavutil/pixfmt.h"
+#include "avfilter.h"
+#include "drawutils.h"
+#include "formats.h"
+#include "video.h"
+
+enum { Y, U, V, A };
+
+typedef struct {
+    int is_packed_rgb;
+    uint8_t rgba_map[4];
+} AlphaExtractContext;
+
+static int query_formats(AVFilterContext *ctx)
+{
+    enum PixelFormat in_fmts[] = {
+        PIX_FMT_YUVA444P, PIX_FMT_YUVA422P, PIX_FMT_YUVA420P,
+        PIX_FMT_RGBA, PIX_FMT_BGRA, PIX_FMT_ARGB, PIX_FMT_ABGR,
+        PIX_FMT_NONE
+    };
+    enum PixelFormat out_fmts[] = { PIX_FMT_GRAY8, PIX_FMT_NONE };
+    ff_formats_ref(ff_make_format_list(in_fmts), &ctx->inputs[0]->out_formats);
+    ff_formats_ref(ff_make_format_list(out_fmts), &ctx->outputs[0]->in_formats);
+    return 0;
+}
+
+static int config_input(AVFilterLink *inlink)
+{
+    AlphaExtractContext *extract = inlink->dst->priv;
+    extract->is_packed_rgb =
+        ff_fill_rgba_map(extract->rgba_map, inlink->format) >= 0;
+    return 0;
+}
+
+static void draw_slice(AVFilterLink *inlink, int y0, int h, int slice_dir)
+{
+    AlphaExtractContext *extract = inlink->dst->priv;
+    AVFilterBufferRef *cur_buf = inlink->cur_buf;
+    AVFilterBufferRef *out_buf = inlink->dst->outputs[0]->out_buf;
+
+    if (extract->is_packed_rgb) {
+        int x, y;
+        uint8_t *pin, *pout;
+        for (y = y0; y < (y0 + h); y++) {
+            pin = cur_buf->data[0] + y * cur_buf->linesize[0] + extract->rgba_map[A];
+            pout = out_buf->data[0] + y * out_buf->linesize[0];
+            for (x = 0; x < out_buf->video->w; x++) {
+                *pout = *pin;
+                pout += 1;
+                pin += 4;
+            }
+        }
+    } else if (cur_buf->linesize[A] == out_buf->linesize[Y]) {
+        const int linesize = cur_buf->linesize[A];
+        memcpy(out_buf->data[Y] + y0 * linesize,
+               cur_buf->data[A] + y0 * linesize,
+               linesize * h);
+    } else {
+        const int linesize = FFMIN(out_buf->linesize[Y], cur_buf->linesize[A]);
+        int y;
+        for (y = y0; y < (y0 + h); y++) {
+            memcpy(out_buf->data[Y] + y * out_buf->linesize[Y],
+                   cur_buf->data[A] + y * cur_buf->linesize[A],
+                   linesize);
+        }
+    }
+    ff_draw_slice(inlink->dst->outputs[0], y0, h, slice_dir);
+}
+
+AVFilter avfilter_vf_alphaextract = {
+    .name           = "alphaextract",
+    .description    = NULL_IF_CONFIG_SMALL("Extract an alpha channel as a "
+                      "grayscale image component."),
+    .priv_size      = sizeof(AlphaExtractContext),
+    .query_formats  = query_formats,
+
+    .inputs    = (const AVFilterPad[]) {
+        { .name             = "default",
+          .type             = AVMEDIA_TYPE_VIDEO,
+          .config_props     = config_input,
+          .draw_slice       = draw_slice,
+          .min_perms        = AV_PERM_READ },
+        { .name = NULL }
+    },
+    .outputs   = (const AVFilterPad[]) {
+      { .name               = "default",
+        .type               = AVMEDIA_TYPE_VIDEO, },
+      { .name = NULL }
+    },
+};
diff --git a/libavfilter/vf_alphamerge.c b/libavfilter/vf_alphamerge.c
new file mode 100644 (file)
index 0000000..67422a1
--- /dev/null
@@ -0,0 +1,212 @@
+/*
+ * Copyright (c) 2012 Steven Robertson
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * copy an alpha component from another video's luma
+ */
+
+#include <string.h>
+
+#include "libavutil/pixfmt.h"
+#include "avfilter.h"
+#include "bufferqueue.h"
+#include "drawutils.h"
+#include "formats.h"
+#include "internal.h"
+#include "video.h"
+
+enum { Y, U, V, A };
+
+typedef struct {
+    int frame_requested;
+    int is_packed_rgb;
+    uint8_t rgba_map[4];
+    struct FFBufQueue queue_main;
+    struct FFBufQueue queue_alpha;
+} AlphaMergeContext;
+
+static av_cold void uninit(AVFilterContext *ctx)
+{
+    AlphaMergeContext *merge = ctx->priv;
+    ff_bufqueue_discard_all(&merge->queue_main);
+    ff_bufqueue_discard_all(&merge->queue_alpha);
+}
+
+static int query_formats(AVFilterContext *ctx)
+{
+    enum PixelFormat main_fmts[] = {
+        PIX_FMT_YUVA444P, PIX_FMT_YUVA422P, PIX_FMT_YUVA420P,
+        PIX_FMT_RGBA, PIX_FMT_BGRA, PIX_FMT_ARGB, PIX_FMT_ABGR,
+        PIX_FMT_NONE
+    };
+    enum PixelFormat alpha_fmts[] = { PIX_FMT_GRAY8, PIX_FMT_NONE };
+    AVFilterFormats *main_formats = ff_make_format_list(main_fmts);
+    AVFilterFormats *alpha_formats = ff_make_format_list(alpha_fmts);
+    ff_formats_ref(main_formats, &ctx->inputs[0]->out_formats);
+    ff_formats_ref(alpha_formats, &ctx->inputs[1]->out_formats);
+    ff_formats_ref(main_formats, &ctx->outputs[0]->in_formats);
+    return 0;
+}
+
+static int config_input_main(AVFilterLink *inlink)
+{
+    AlphaMergeContext *merge = inlink->dst->priv;
+    merge->is_packed_rgb =
+        ff_fill_rgba_map(merge->rgba_map, inlink->format) >= 0;
+    return 0;
+}
+
+static int config_output(AVFilterLink *outlink)
+{
+    AVFilterContext *ctx = outlink->src;
+    AVFilterLink *mainlink = ctx->inputs[0];
+    AVFilterLink *alphalink = ctx->inputs[1];
+    if (mainlink->w != alphalink->w || mainlink->h != alphalink->h) {
+        av_log(ctx, AV_LOG_ERROR,
+               "Input frame sizes do not match (%dx%d vs %dx%d).\n",
+               mainlink->w, mainlink->h,
+               alphalink->w, alphalink->h);
+        return AVERROR(EINVAL);
+    }
+
+    outlink->w = mainlink->w;
+    outlink->h = mainlink->h;
+    outlink->time_base = mainlink->time_base;
+    outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
+    outlink->frame_rate = mainlink->frame_rate;
+    return 0;
+}
+
+static void start_frame(AVFilterLink *inlink, AVFilterBufferRef *picref) {}
+static void draw_slice(AVFilterLink *inlink, int y, int h, int slice_dir) {}
+
+static void draw_frame(AVFilterContext *ctx,
+                       AVFilterBufferRef *main_buf,
+                       AVFilterBufferRef *alpha_buf)
+{
+    AlphaMergeContext *merge = ctx->priv;
+    int h = main_buf->video->h;
+
+    if (merge->is_packed_rgb) {
+        int x, y;
+        uint8_t *pin, *pout;
+        for (y = 0; y < h; y++) {
+            pin = alpha_buf->data[0] + y * alpha_buf->linesize[0];
+            pout = main_buf->data[0] + y * main_buf->linesize[0] + merge->rgba_map[A];
+            for (x = 0; x < main_buf->video->w; x++) {
+                *pout = *pin;
+                pin += 1;
+                pout += 4;
+            }
+        }
+    } else {
+        int y;
+        const int main_linesize = main_buf->linesize[A];
+        const int alpha_linesize = alpha_buf->linesize[Y];
+        for (y = 0; y < h && y < alpha_buf->video->h; y++) {
+            memcpy(main_buf->data[A] + y * main_linesize,
+                   alpha_buf->data[Y] + y * alpha_linesize,
+                   FFMIN(main_linesize, alpha_linesize));
+        }
+    }
+    ff_draw_slice(ctx->outputs[0], 0, h, 1);
+}
+
+static void end_frame(AVFilterLink *inlink)
+{
+    AVFilterContext *ctx = inlink->dst;
+    AlphaMergeContext *merge = ctx->priv;
+
+    int is_alpha = (inlink == ctx->inputs[1]);
+    struct FFBufQueue *queue =
+        (is_alpha ? &merge->queue_alpha : &merge->queue_main);
+    ff_bufqueue_add(ctx, queue, inlink->cur_buf);
+    inlink->cur_buf = NULL;
+
+    while (1) {
+        AVFilterBufferRef *main_buf, *alpha_buf;
+
+        if (!ff_bufqueue_peek(&merge->queue_main, 0) ||
+            !ff_bufqueue_peek(&merge->queue_alpha, 0)) break;
+
+        main_buf = ff_bufqueue_get(&merge->queue_main);
+        alpha_buf = ff_bufqueue_get(&merge->queue_alpha);
+
+        ctx->outputs[0]->out_buf = main_buf;
+        ff_start_frame(ctx->outputs[0], avfilter_ref_buffer(main_buf, ~0));
+        merge->frame_requested = 0;
+        draw_frame(ctx, main_buf, alpha_buf);
+        ff_end_frame(ctx->outputs[0]);
+        avfilter_unref_buffer(alpha_buf);
+    }
+}
+
+static int request_frame(AVFilterLink *outlink)
+{
+    AVFilterContext *ctx = outlink->src;
+    AlphaMergeContext *merge = ctx->priv;
+    int in, ret;
+
+    merge->frame_requested = 1;
+    while (merge->frame_requested) {
+        in = ff_bufqueue_peek(&merge->queue_main, 0) ? 0 : 1;
+        ret = ff_request_frame(ctx->inputs[in]);
+        if (ret < 0)
+            return ret;
+    }
+    return 0;
+}
+
+AVFilter avfilter_vf_alphamerge = {
+    .name           = "alphamerge",
+    .description    = NULL_IF_CONFIG_SMALL("Copy the luma value of the second "
+                      "input into the alpha channel of the first input."),
+    .uninit         = uninit,
+    .priv_size      = sizeof(AlphaMergeContext),
+    .query_formats  = query_formats,
+
+    .inputs    = (const AVFilterPad[]) {
+        { .name             = "main",
+          .type             = AVMEDIA_TYPE_VIDEO,
+          .config_props     = config_input_main,
+          .get_video_buffer = ff_null_get_video_buffer,
+          .start_frame      = start_frame,
+          .draw_slice       = draw_slice,
+          .end_frame        = end_frame,
+          .min_perms        = AV_PERM_READ | AV_PERM_WRITE,
+          .rej_perms        = AV_PERM_REUSE2 | AV_PERM_PRESERVE },
+        { .name             = "alpha",
+          .type             = AVMEDIA_TYPE_VIDEO,
+          .start_frame      = start_frame,
+          .draw_slice       = draw_slice,
+          .end_frame        = end_frame,
+          .min_perms        = AV_PERM_READ,
+          .rej_perms        = AV_PERM_REUSE2 },
+        { .name = NULL }
+    },
+    .outputs   = (const AVFilterPad[]) {
+      { .name               = "default",
+        .type               = AVMEDIA_TYPE_VIDEO,
+        .config_props       = config_output,
+        .request_frame      = request_frame },
+      { .name = NULL }
+    },
+};
index dd5e2da..6104044 100755 (executable)
@@ -13,21 +13,25 @@ eval do_$test=y
 
 do_video_filter() {
     label=$1
-    filters=$2
+    filters="$2"
     shift 2
     printf '%-20s' $label
     run_avconv $DEC_OPTS -f image2 -vcodec pgmyuv -i $raw_src    \
         $ENC_OPTS -vf "$filters" -vcodec rawvideo $* -f nut md5:
 }
 
-do_lavfi() {
-    vfilters="slicify=random,$2"
+do_lavfi_plain() {
+    vfilters="$2"
 
     if [ $test = $1 ] ; then
         do_video_filter $test "$vfilters"
     fi
 }
 
+do_lavfi() {
+    do_lavfi_plain $1 "slicify=random,$2"
+}
+
 do_lavfi_colormatrix() {
     do_lavfi "${1}1" "$1=$4:$5,$1=$5:$3,$1=$3:$4,$1=$4:$3,$1=$3:$5,$1=$5:$2"
     do_lavfi "${1}2" "$1=$2:$3,$1=$3:$2,$1=$2:$4,$1=$4:$2,$1=$2:$5,$1=$5:$4"
@@ -60,6 +64,11 @@ do_lavfi "vflip"              "vflip"
 do_lavfi "vflip_crop"         "vflip,crop=iw-100:ih-100:100:100"
 do_lavfi "vflip_vflip"        "vflip,vflip"
 
+do_lavfi_plain "alphamerge_rgb"     "[in]slicify=random,format=bgra,split,alphamerge[out]"
+do_lavfi_plain "alphamerge_yuv"     "[in]slicify=random,format=yuv420p,split,alphamerge[out]"
+do_lavfi_plain "alphaextract_rgb"   "[in]slicify=random,format=bgra,split,alphamerge,slicify=random,split[o3][o4];[o4]alphaextract[alpha];[o3][alpha]alphamerge[out]"
+do_lavfi_plain "alphaextract_yuv"   "[in]slicify=random,format=yuv420p,split,alphamerge,slicify=random,split[o3][o4];[o4]alphaextract[alpha];[o3][alpha]alphamerge[out]"
+
 do_lavfi_colormatrix "colormatrix" bt709 fcc bt601 smpte240m
 
 do_lavfi_pixfmts(){
diff --git a/tests/ref/lavfi/alphaextract_rgb b/tests/ref/lavfi/alphaextract_rgb
new file mode 100644 (file)
index 0000000..3b86986
--- /dev/null
@@ -0,0 +1 @@
+alphaextract_rgb    b706818ea0d324e0d43adbaef9ab3470
diff --git a/tests/ref/lavfi/alphaextract_yuv b/tests/ref/lavfi/alphaextract_yuv
new file mode 100644 (file)
index 0000000..7bfcf86
--- /dev/null
@@ -0,0 +1 @@
+alphaextract_yuv    430b779a379ad9c38f3b4c190f723cd4
diff --git a/tests/ref/lavfi/alphamerge_rgb b/tests/ref/lavfi/alphamerge_rgb
new file mode 100644 (file)
index 0000000..6dc3b9a
--- /dev/null
@@ -0,0 +1 @@
+alphamerge_rgb      b706818ea0d324e0d43adbaef9ab3470
diff --git a/tests/ref/lavfi/alphamerge_yuv b/tests/ref/lavfi/alphamerge_yuv
new file mode 100644 (file)
index 0000000..ecd7bf8
--- /dev/null
@@ -0,0 +1 @@
+alphamerge_yuv      430b779a379ad9c38f3b4c190f723cd4