avfilter: add maskedmerge filter
authorPaul B Mahol <onemda@gmail.com>
Fri, 25 Sep 2015 20:46:38 +0000 (22:46 +0200)
committerPaul B Mahol <onemda@gmail.com>
Mon, 28 Sep 2015 07:20:58 +0000 (09:20 +0200)
Signed-off-by: Paul B Mahol <onemda@gmail.com>
Changelog
doc/filters.texi
libavfilter/Makefile
libavfilter/allfilters.c
libavfilter/version.h
libavfilter/vf_maskedmerge.c [new file with mode: 0644]

index d5017e1..515d649 100644 (file)
--- a/Changelog
+++ b/Changelog
@@ -12,6 +12,7 @@ version <next>:
 - tremolo filter
 - agate filter
 - chromakey filter
+- maskedmerge filter
 
 
 version 2.8:
index 0fdb677..a5906d6 100644 (file)
@@ -7625,6 +7625,25 @@ lutyuv=y='bitand(val, 128+64+32)'
 @end example
 @end itemize
 
+@section maskedmerge
+
+Merge the first input stream with the second input stream using per pixel
+weights in the third input stream.
+
+A value of 0 in the third stream pixel component means that pixel component
+from first stream is returned unchanged, while maximum value (eg. 255 for
+8-bit videos) means that pixel component from second stream is returned
+unchanged. Intermediate values define the amount of merging between both
+input stream's pixel components.
+
+This filter accepts the following options:
+@table @option
+@item planes
+Set which planes will be processed as bitmap, unprocessed planes will be
+copied from first stream.
+By default value 0xf, all planes will be processed.
+@end table
+
 @section mergeplanes
 
 Merge color channel components from several video streams.
index c70887e..9125443 100644 (file)
@@ -169,6 +169,7 @@ OBJS-$(CONFIG_LUT3D_FILTER)                  += vf_lut3d.o
 OBJS-$(CONFIG_LUT_FILTER)                    += vf_lut.o
 OBJS-$(CONFIG_LUTRGB_FILTER)                 += vf_lut.o
 OBJS-$(CONFIG_LUTYUV_FILTER)                 += vf_lut.o
+OBJS-$(CONFIG_MASKEDMERGE_FILTER)            += vf_maskedmerge.o framesync.o
 OBJS-$(CONFIG_MCDEINT_FILTER)                += vf_mcdeint.o
 OBJS-$(CONFIG_MERGEPLANES_FILTER)            += vf_mergeplanes.o framesync.o
 OBJS-$(CONFIG_MPDECIMATE_FILTER)             += vf_mpdecimate.o
index c21456d..612fc99 100644 (file)
@@ -191,6 +191,7 @@ void avfilter_register_all(void)
     REGISTER_FILTER(LUT,            lut,            vf);
     REGISTER_FILTER(LUTRGB,         lutrgb,         vf);
     REGISTER_FILTER(LUTYUV,         lutyuv,         vf);
+    REGISTER_FILTER(MASKEDMERGE,    maskedmerge,    vf);
     REGISTER_FILTER(MCDEINT,        mcdeint,        vf);
     REGISTER_FILTER(MERGEPLANES,    mergeplanes,    vf);
     REGISTER_FILTER(MPDECIMATE,     mpdecimate,     vf);
index 1426cf8..670ee4e 100644 (file)
@@ -30,7 +30,7 @@
 #include "libavutil/version.h"
 
 #define LIBAVFILTER_VERSION_MAJOR   6
-#define LIBAVFILTER_VERSION_MINOR   8
+#define LIBAVFILTER_VERSION_MINOR   9
 #define LIBAVFILTER_VERSION_MICRO 100
 
 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \
diff --git a/libavfilter/vf_maskedmerge.c b/libavfilter/vf_maskedmerge.c
new file mode 100644 (file)
index 0000000..72f9fb3
--- /dev/null
@@ -0,0 +1,324 @@
+/*
+ * Copyright (c) 2015 Paul B Mahol
+ *
+ * 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
+ */
+
+#include "libavutil/imgutils.h"
+#include "libavutil/pixdesc.h"
+#include "libavutil/opt.h"
+#include "avfilter.h"
+#include "formats.h"
+#include "framesync.h"
+#include "internal.h"
+#include "video.h"
+
+typedef struct MaskedMergeContext {
+    const AVClass *class;
+    int width[4], height[4];
+    int nb_planes;
+    int planes;
+    int max, half, depth;
+    FFFrameSync fs;
+
+    void (*maskedmerge)(struct MaskedMergeContext *s, const AVFrame *base,
+                        const AVFrame *overlay, const AVFrame *mask, AVFrame *out);
+} MaskedMergeContext;
+
+#define OFFSET(x) offsetof(MaskedMergeContext, x)
+#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
+
+static const AVOption maskedmerge_options[] = {
+    { "planes", "set planes", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=0xF}, 0, 0xF, FLAGS },
+    { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(maskedmerge);
+
+static int query_formats(AVFilterContext *ctx)
+{
+    static const enum AVPixelFormat pix_fmts[] = {
+        AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P,
+        AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
+        AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUV420P,
+        AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
+        AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
+        AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
+        AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
+        AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
+        AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA444P9,
+        AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA444P10,
+        AV_PIX_FMT_YUVA420P16, AV_PIX_FMT_YUVA422P16, AV_PIX_FMT_YUVA444P16,
+        AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
+        AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
+        AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP16,
+        AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY16,
+        AV_PIX_FMT_NONE
+    };
+
+    ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
+    return 0;
+}
+
+static int process_frame(FFFrameSync *fs)
+{
+    AVFilterContext *ctx = fs->parent;
+    MaskedMergeContext *s = fs->opaque;
+    AVFilterLink *outlink = ctx->outputs[0];
+    AVFrame *out, *base, *overlay, *mask;
+    int ret;
+
+    if ((ret = ff_framesync_get_frame(&s->fs, 0, &base,    0)) < 0 ||
+        (ret = ff_framesync_get_frame(&s->fs, 1, &overlay, 0)) < 0 ||
+        (ret = ff_framesync_get_frame(&s->fs, 2, &mask,    0)) < 0)
+        return ret;
+
+    if (ctx->is_disabled) {
+        out = av_frame_clone(base);
+        if (!out)
+            return AVERROR(ENOMEM);
+    } else {
+        out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
+        if (!out)
+            return AVERROR(ENOMEM);
+        av_frame_copy_props(out, base);
+
+        s->maskedmerge(s, base, overlay, mask, out);
+    }
+    out->pts = av_rescale_q(base->pts, s->fs.time_base, outlink->time_base);
+
+    return ff_filter_frame(outlink, out);
+}
+
+static void maskedmerge8(MaskedMergeContext *s, const AVFrame *base,
+                         const AVFrame *overlay, const AVFrame *mask, AVFrame *out)
+{
+    int p, x, y;
+
+    for (p = 0; p < s->nb_planes; p++) {
+        const uint8_t *bsrc = base->data[p];
+        const uint8_t *osrc = overlay->data[p];
+        const uint8_t *msrc = mask->data[p];
+        uint8_t *dst = out->data[p];
+
+        if (!((1 << p) & s->planes)) {
+            av_image_copy_plane(dst, out->linesize[p], bsrc, base->linesize[p],
+                                s->width[p], s->height[p]);
+            continue;
+        }
+
+        for (y = 0; y < s->height[p]; y++) {
+            for (x = 0; x < s->width[p]; x++) {
+                dst[x] = ((256 - msrc[x]) * bsrc[x] + msrc[x] * osrc[x] + 128) >> 8;
+            }
+
+            dst  += out->linesize[p];
+            bsrc += base->linesize[p];
+            osrc += overlay->linesize[p];
+            msrc += mask->linesize[p];
+        }
+    }
+}
+
+static void maskedmerge16(MaskedMergeContext *s, const AVFrame *base,
+                          const AVFrame *overlay, const AVFrame *mask, AVFrame *out)
+{
+    const int max = s->max;
+    const int half = s->half;
+    const int shift = s->depth;
+    int p, x, y;
+
+    for (p = 0; p < s->nb_planes; p++) {
+        const uint16_t *bsrc = (const uint16_t *)base->data[p];
+        const uint16_t *osrc = (const uint16_t *)overlay->data[p];
+        const uint16_t *msrc = (const uint16_t *)mask->data[p];
+        uint16_t *dst = (uint16_t *)out->data[p];
+
+        if (!((1 << p) & s->planes)) {
+            av_image_copy_plane(out->data[p], out->linesize[p], base->data[p], base->linesize[p],
+                                s->width[p], s->height[p]);
+            continue;
+        }
+
+        for (y = 0; y < s->height[p]; y++) {
+            for (x = 0; x < s->width[p]; x++) {
+                dst[x] = ((max - msrc[x]) * bsrc[x] + msrc[x] * osrc[x] + half) >> shift;
+            }
+
+            dst  += out->linesize[p]     / 2;
+            bsrc += base->linesize[p]    / 2;
+            osrc += overlay->linesize[p] / 2;
+            msrc += mask->linesize[p]    / 2;
+        }
+    }
+}
+
+static int config_input(AVFilterLink *inlink)
+{
+    AVFilterContext *ctx = inlink->dst;
+    MaskedMergeContext *s = ctx->priv;
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
+    int vsub, hsub;
+
+    s->nb_planes = av_pix_fmt_count_planes(inlink->format);
+
+    hsub = desc->log2_chroma_w;
+    vsub = desc->log2_chroma_h;
+    s->height[1] = s->height[2] = FF_CEIL_RSHIFT(inlink->h, vsub);
+    s->height[0] = s->height[3] = inlink->h;
+    s->width[1]  = s->width[2]  = FF_CEIL_RSHIFT(inlink->w, hsub);
+    s->width[0]  = s->width[3]  = inlink->w;
+
+    s->depth = desc->comp[0].depth;
+    s->max = 1 << s->depth;
+    s->half = s->max / 2;
+
+    if (desc->comp[0].depth == 8)
+        s->maskedmerge = maskedmerge8;
+    else
+        s->maskedmerge = maskedmerge16;
+
+    return 0;
+}
+
+static int config_output(AVFilterLink *outlink)
+{
+    AVFilterContext *ctx = outlink->src;
+    MaskedMergeContext *s = ctx->priv;
+    AVFilterLink *base = ctx->inputs[0];
+    AVFilterLink *overlay = ctx->inputs[1];
+    AVFilterLink *mask = ctx->inputs[2];
+    FFFrameSyncIn *in;
+    int ret;
+
+    if (base->format != overlay->format ||
+        base->format != mask->format) {
+        av_log(ctx, AV_LOG_ERROR, "inputs must be of same pixel format\n");
+        return AVERROR(EINVAL);
+    }
+    if (base->w                       != overlay->w ||
+        base->h                       != overlay->h ||
+        base->sample_aspect_ratio.num != overlay->sample_aspect_ratio.num ||
+        base->sample_aspect_ratio.den != overlay->sample_aspect_ratio.den ||
+        base->w                       != mask->w ||
+        base->h                       != mask->h ||
+        base->sample_aspect_ratio.num != mask->sample_aspect_ratio.num ||
+        base->sample_aspect_ratio.den != mask->sample_aspect_ratio.den) {
+        av_log(ctx, AV_LOG_ERROR, "First input link %s parameters "
+               "(size %dx%d, SAR %d:%d) do not match the corresponding "
+               "second input link %s parameters (%dx%d, SAR %d:%d) "
+               "and/or third input link %s parameters (%dx%d, SAR %d:%d)\n",
+               ctx->input_pads[0].name, base->w, base->h,
+               base->sample_aspect_ratio.num,
+               base->sample_aspect_ratio.den,
+               ctx->input_pads[1].name, overlay->w, overlay->h,
+               overlay->sample_aspect_ratio.num,
+               overlay->sample_aspect_ratio.den,
+               ctx->input_pads[2].name, mask->w, mask->h,
+               mask->sample_aspect_ratio.num,
+               mask->sample_aspect_ratio.den);
+        return AVERROR(EINVAL);
+    }
+
+    outlink->w = base->w;
+    outlink->h = base->h;
+    outlink->time_base = base->time_base;
+    outlink->sample_aspect_ratio = base->sample_aspect_ratio;
+    outlink->frame_rate = base->frame_rate;
+
+    if ((ret = ff_framesync_init(&s->fs, ctx, 3)) < 0)
+        return ret;
+
+    in = s->fs.in;
+    in[0].time_base = base->time_base;
+    in[1].time_base = overlay->time_base;
+    in[2].time_base = mask->time_base;
+    in[0].sync   = 1;
+    in[0].before = EXT_STOP;
+    in[0].after  = EXT_INFINITY;
+    in[1].sync   = 1;
+    in[1].before = EXT_STOP;
+    in[1].after  = EXT_INFINITY;
+    in[2].sync   = 1;
+    in[2].before = EXT_STOP;
+    in[2].after  = EXT_INFINITY;
+    s->fs.opaque   = s;
+    s->fs.on_event = process_frame;
+
+    return ff_framesync_configure(&s->fs);
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
+{
+    MaskedMergeContext *s = inlink->dst->priv;
+    return ff_framesync_filter_frame(&s->fs, inlink, buf);
+}
+
+static int request_frame(AVFilterLink *outlink)
+{
+    MaskedMergeContext *s = outlink->src->priv;
+    return ff_framesync_request_frame(&s->fs, outlink);
+}
+
+static av_cold void uninit(AVFilterContext *ctx)
+{
+    MaskedMergeContext *s = ctx->priv;
+
+    ff_framesync_uninit(&s->fs);
+}
+
+static const AVFilterPad maskedmerge_inputs[] = {
+    {
+        .name         = "base",
+        .type         = AVMEDIA_TYPE_VIDEO,
+        .filter_frame = filter_frame,
+        .config_props = config_input,
+    },
+    {
+        .name         = "overlay",
+        .type         = AVMEDIA_TYPE_VIDEO,
+        .filter_frame = filter_frame,
+    },
+    {
+        .name         = "mask",
+        .type         = AVMEDIA_TYPE_VIDEO,
+        .filter_frame = filter_frame,
+    },
+    { NULL }
+};
+
+static const AVFilterPad maskedmerge_outputs[] = {
+    {
+        .name          = "default",
+        .type          = AVMEDIA_TYPE_VIDEO,
+        .config_props  = config_output,
+        .request_frame = request_frame,
+    },
+    { NULL }
+};
+
+AVFilter ff_vf_maskedmerge = {
+    .name          = "maskedmerge",
+    .description   = NULL_IF_CONFIG_SMALL("Merge first stream with second stream using third stream as mask."),
+    .priv_size     = sizeof(MaskedMergeContext),
+    .uninit        = uninit,
+    .query_formats = query_formats,
+    .inputs        = maskedmerge_inputs,
+    .outputs       = maskedmerge_outputs,
+    .priv_class    = &maskedmerge_class,
+    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
+};