[FFmpeg-devel] [PATCH] lavfi: port ow from libmpcodecs.

Clément Bœsch ubitux at gmail.com
Fri May 10 22:10:15 CEST 2013


TODO: Changelog & minor bump

---
Feel free to bikeshed for a new name.

mp=ow will be dropped in the same time.
---
 doc/filters.texi         |  26 ++++
 libavfilter/Makefile     |   1 +
 libavfilter/allfilters.c |   1 +
 libavfilter/vf_ow.c      | 344 +++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 372 insertions(+)
 create mode 100644 libavfilter/vf_ow.c

diff --git a/doc/filters.texi b/doc/filters.texi
index fe38b7f..8b9fb4a 100644
--- a/doc/filters.texi
+++ b/doc/filters.texi
@@ -5103,6 +5103,32 @@ testsrc=s=100x100, split=4 [in0][in1][in2][in3];
 
 @end itemize
 
+ at section ow
+
+Overcomplete Wavelet denoiser.
+
+The filter accepts the following options:
+
+ at table @option
+ at item depth
+Set depth.
+
+Larger depth values will denoise lower frequency components more, but
+slow down filtering
+
+Default is @code{8}.
+
+ at item luma_strength
+Set luma strength.
+
+Default is @code{1.0}
+
+ at item chroma_strength
+Set chroma strength.
+
+Default is @code{1.0}
+ at end table
+
 @section pad
 
 Add paddings to the input image, and place the original input at the
diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index 73a24dd..f6daed1 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -154,6 +154,7 @@ OBJS-$(CONFIG_NULL_FILTER)                   += vf_null.o
 OBJS-$(CONFIG_OCV_FILTER)                    += vf_libopencv.o
 OBJS-$(CONFIG_OPENCL)                        += deshake_opencl.o unsharp_opencl.o
 OBJS-$(CONFIG_OVERLAY_FILTER)                += vf_overlay.o
+OBJS-$(CONFIG_OW_FILTER)                     += vf_ow.o
 OBJS-$(CONFIG_PAD_FILTER)                    += vf_pad.o
 OBJS-$(CONFIG_PERMS_FILTER)                  += f_perms.o
 OBJS-$(CONFIG_PIXDESCTEST_FILTER)            += vf_pixdesctest.o
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index b8f273d..8986547 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -151,6 +151,7 @@ void avfilter_register_all(void)
     REGISTER_FILTER(NULL,           null,           vf);
     REGISTER_FILTER(OCV,            ocv,            vf);
     REGISTER_FILTER(OVERLAY,        overlay,        vf);
+    REGISTER_FILTER(OW,             ow,             vf);
     REGISTER_FILTER(PAD,            pad,            vf);
     REGISTER_FILTER(PERMS,          perms,          vf);
     REGISTER_FILTER(PIXDESCTEST,    pixdesctest,    vf);
diff --git a/libavfilter/vf_ow.c b/libavfilter/vf_ow.c
new file mode 100644
index 0000000..97c386a
--- /dev/null
+++ b/libavfilter/vf_ow.c
@@ -0,0 +1,344 @@
+/*
+ * Copyright (C) 2007 Michael Niedermayer <michaelni at gmx.at>
+ * Copyright (c) 2013 Clément Bœsch
+ *
+ * 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
+ */
+
+/**
+ * @todo try to change to int
+ * @todo try lifting based implementation
+ * @todo optimize optimize optimize
+ * @todo hard tresholding
+ * @todo use QP to decide filter strength
+ * @todo wavelet normalization / least squares optimal signal vs. noise thresholds
+ */
+
+#include "libavutil/imgutils.h"
+#include "libavutil/opt.h"
+#include "libavutil/pixdesc.h"
+#include "avfilter.h"
+#include "internal.h"
+
+typedef struct {
+    const AVClass *class;
+    double luma_strength;
+    double chroma_strength;
+    int depth;
+    float *plane[16+1][4];
+    int linesize;
+    int hsub, vsub;
+} OWContext;
+
+#define OFFSET(x) offsetof(OWContext, x)
+#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
+static const AVOption ow_options[] = {
+    { "depth",           "set depth",           OFFSET(depth),           AV_OPT_TYPE_INT,    {.i64 = 8},    8,   16, FLAGS },
+    { "luma_strength",   "set luma strength",   OFFSET(luma_strength),   AV_OPT_TYPE_DOUBLE, {.dbl = 1.0 }, 0, 1000, FLAGS },
+    { "ls",              "set luma strength",   OFFSET(luma_strength),   AV_OPT_TYPE_DOUBLE, {.dbl = 1.0 }, 0, 1000, FLAGS },
+    { "chroma_strength", "set chroma strength", OFFSET(chroma_strength), AV_OPT_TYPE_DOUBLE, {.dbl = 1.0 }, 0, 1000, FLAGS },
+    { "cs",              "set chroma strength", OFFSET(chroma_strength), AV_OPT_TYPE_DOUBLE, {.dbl = 1.0 }, 0, 1000, FLAGS },
+    { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(ow);
+
+DECLARE_ALIGNED(8, static const uint8_t, dither)[8][8] = {
+    {  0,  48,  12,  60,   3,  51,  15,  63 },
+    { 32,  16,  44,  28,  35,  19,  47,  31 },
+    {  8,  56,   4,  52,  11,  59,   7,  55 },
+    { 40,  24,  36,  20,  43,  27,  39,  23 },
+    {  2,  50,  14,  62,   1,  49,  13,  61 },
+    { 34,  18,  46,  30,  33,  17,  45,  29 },
+    { 10,  58,   6,  54,   9,  57,   5,  53 },
+    { 42,  26,  38,  22,  41,  25,  37,  21 },
+};
+
+#define SQRT2 1.41421356237
+
+static const double coeff[2][5] = {
+    {
+         0.6029490182363579  * SQRT2,
+         0.2668641184428723  * SQRT2,
+        -0.07822326652898785 * SQRT2,
+        -0.01686411844287495 * SQRT2,
+         0.02674875741080976 * SQRT2,
+    },{
+         1.115087052456994   / SQRT2,
+        -0.5912717631142470  / SQRT2,
+        -0.05754352622849957 / SQRT2,
+         0.09127176311424948 / SQRT2,
+    }
+};
+
+static const double icoeff[2][5]={
+    {
+         1.115087052456994   / SQRT2,
+         0.5912717631142470  / SQRT2,
+        -0.05754352622849957 / SQRT2,
+        -0.09127176311424948 / SQRT2,
+    },{
+         0.6029490182363579  * SQRT2,
+        -0.2668641184428723  * SQRT2,
+        -0.07822326652898785 * SQRT2,
+         0.01686411844287495 * SQRT2,
+         0.02674875741080976 * SQRT2,
+    }
+};
+
+static inline int mirror(int x, int w)
+{
+    while((unsigned)x > (unsigned)w) {
+        x = -x;
+        if (x < 0)
+            x += 2 * w;
+    }
+    return x;
+}
+
+static inline void decompose(float *dst_l, float *dst_h, const float *src,
+                             int linesize, int w)
+{
+    int x, i;
+    for (x = 0; x < w; x++) {
+        double sum_l = src[x * linesize] * coeff[0][0];
+        double sum_h = src[x * linesize] * coeff[1][0];
+        for (i = 1; i <= 4; i++) {
+            const double s = src[mirror(x - i, w - 1) * linesize]
+                           + src[mirror(x + i, w - 1) * linesize];
+
+            sum_l += coeff[0][i] * s;
+            sum_h += coeff[1][i] * s;
+        }
+        dst_l[x * linesize] = sum_l;
+        dst_h[x * linesize] = sum_h;
+    }
+}
+
+static inline void compose(float *dst, const float *src_l, const float *src_h,
+                           int linesize, int w)
+{
+    int x, i;
+    for (x = 0; x < w; x++) {
+        double sum_l = src_l[x * linesize] * icoeff[0][0];
+        double sum_h = src_h[x * linesize] * icoeff[1][0];
+        for (i = 1; i <= 4; i++) {
+            const int x0 = mirror(x - i, w - 1) * linesize;
+            const int x1 = mirror(x + i, w - 1) * linesize;
+
+            sum_l += icoeff[0][i] * (src_l[x0] + src_l[x1]);
+            sum_h += icoeff[1][i] * (src_h[x0] + src_h[x1]);
+        }
+        dst[x * linesize] = (sum_l + sum_h) * 0.5;
+    }
+}
+
+static inline void decompose2D(float *dst_l, float *dst_h, const float *src,
+                               int xlinesize, int ylinesize,
+                               int step, int w, int h)
+{
+    int y, x;
+    for (y = 0; y < h; y++)
+        for (x = 0; x < step; x++)
+            decompose(dst_l + ylinesize*y + xlinesize*x,
+                      dst_h + ylinesize*y + xlinesize*x,
+                      src   + ylinesize*y + xlinesize*x,
+                      step * xlinesize, (w - x + step - 1) / step);
+}
+
+static inline void compose2D(float *dst, const float *src_l, const float *src_h,
+                             int xlinesize, int ylinesize,
+                             int step, int w, int h)
+{
+    int y, x;
+    for (y = 0; y < h; y++)
+        for (x = 0; x < step; x++)
+            compose(dst   + ylinesize*y + xlinesize*x,
+                    src_l + ylinesize*y + xlinesize*x,
+                    src_h + ylinesize*y + xlinesize*x,
+                    step * xlinesize, (w - x + step - 1) / step);
+}
+
+static void decompose2D2(float *dst[4], float *src, float *temp[2],
+                         int linesize, int step, int w, int h)
+{
+    decompose2D(temp[0], temp[1], src,     1, linesize, step, w, h);
+    decompose2D( dst[0],  dst[1], temp[0], linesize, 1, step, h, w);
+    decompose2D( dst[2],  dst[3], temp[1], linesize, 1, step, h, w);
+}
+
+static void compose2D2(float *dst, float *src[4], float *temp[2],
+                       int linesize, int step, int w, int h)
+{
+    compose2D(temp[0],  src[0],  src[1], linesize, 1, step, h, w);
+    compose2D(temp[1],  src[2],  src[3], linesize, 1, step, h, w);
+    compose2D(dst,     temp[0], temp[1], 1, linesize, step, w, h);
+}
+
+static void filter(OWContext *s,
+                   uint8_t       *dst, int dst_linesize,
+                   const uint8_t *src, int src_linesize,
+                   int width, int height, double strength)
+{
+    int x, y, i, j, depth = s->depth;
+
+    while (1<<depth > width || 1<<depth > height)
+        depth--;
+
+    for (y = 0; y < height; y++)
+        for(x = 0; x < width; x++)
+            s->plane[0][0][y*s->linesize + x] = src[y*src_linesize + x];
+
+    for (i = 0; i < depth; i++)
+        decompose2D2(s->plane[i + 1], s->plane[i][0], s->plane[0] + 1, s->linesize, 1<<i, width, height);
+
+    for (i = 0; i < depth; i++) {
+        for (j = 1; j < 4; j++) {
+            for (y = 0; y < height; y++) {
+                for (x = 0; x < width; x++) {
+                    double v = s->plane[i + 1][j][y*s->linesize + x];
+                    if      (v >  strength) v -= strength;
+                    else if (v < -strength) v += strength;
+                    else                    v  = 0;
+                    s->plane[i + 1][j][x + y*s->linesize]= v;
+                }
+            }
+        }
+    }
+    for (i = depth-1; i >= 0; i--)
+        compose2D2(s->plane[i][0], s->plane[i + 1], s->plane[0] + 1, s->linesize, 1<<i, width, height);
+
+    for (y = 0; y < height; y++) {
+        for (x = 0; x < width; x++) {
+            i = s->plane[0][0][y*s->linesize + x] + dither[x&7][y&7]*(1.0/64) + 1.0/128; // yes the rounding is insane but optimal :)
+            if ((unsigned)i > 255U) i = ~(i>>31);
+            dst[y*dst_linesize + x] = i;
+        }
+    }
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFrame *in)
+{
+    int direct = 0;
+    AVFilterContext *ctx = inlink->dst;
+    OWContext *s = ctx->priv;
+    AVFilterLink *outlink = ctx->outputs[0];
+    AVFrame *out;
+    const int cw = FF_CEIL_RSHIFT(inlink->w, s->hsub);
+    const int ch = FF_CEIL_RSHIFT(inlink->h, s->vsub);
+
+    if (av_frame_is_writable(in)) {
+        direct = 1;
+        out = in;
+    } else {
+        out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
+        if (!out) {
+            av_frame_free(&in);
+            return AVERROR(ENOMEM);
+        }
+        av_frame_copy_props(out, in);
+    }
+
+    filter(s, out->data[0], out->linesize[0], in->data[0], in->linesize[0], inlink->w, inlink->h, s->luma_strength);
+    filter(s, out->data[1], out->linesize[1], in->data[1], in->linesize[1], cw,        ch,        s->chroma_strength);
+    filter(s, out->data[2], out->linesize[2], in->data[2], in->linesize[2], cw,        ch,        s->chroma_strength);
+
+    if (!direct) {
+        if (in->data[3])
+            av_image_copy_plane(out->data[3], out->linesize[3],
+                                in ->data[3], in ->linesize[3],
+                                inlink->w, inlink->h);
+        av_frame_free(&in);
+    }
+
+    return ff_filter_frame(outlink, out);
+}
+
+static int query_formats(AVFilterContext *ctx)
+{
+    static const enum AVPixelFormat pix_fmts[] = {
+        AV_PIX_FMT_YUV444P,      AV_PIX_FMT_YUV422P,
+        AV_PIX_FMT_YUV420P,      AV_PIX_FMT_YUV411P,
+        AV_PIX_FMT_YUV410P,      AV_PIX_FMT_YUV440P,
+        AV_PIX_FMT_YUVA444P,     AV_PIX_FMT_YUVA422P,
+        AV_PIX_FMT_YUVA420P,
+        AV_PIX_FMT_NONE
+    };
+    ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
+    return 0;
+}
+
+static int config_input(AVFilterLink *inlink)
+{
+    int i, j;
+    OWContext *s = inlink->dst->priv;
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
+    const int h = FFALIGN(inlink->h, 16);
+
+    s->hsub = desc->log2_chroma_w;
+    s->vsub = desc->log2_chroma_h;
+
+    s->linesize = FFALIGN(inlink->w, 16);
+    for (j = 0; j < 4; j++) {
+        for (i = 0; i <= s->depth; i++) {
+            s->plane[i][j] = av_malloc(s->linesize * h * sizeof(s->plane[0][0][0]));
+            if (!s->plane[i][j])
+                return AVERROR(ENOMEM);
+        }
+    }
+    return 0;
+}
+
+static av_cold void uninit(AVFilterContext *ctx)
+{
+    int i, j;
+    OWContext *s = ctx->priv;
+
+    for (j = 0; j < 4; j++)
+        for (i = 0; i <= s->depth; i++)
+            av_freep(&s->plane[i][j]);
+}
+
+static const AVFilterPad ow_inputs[] = {
+    {
+        .name         = "default",
+        .type         = AVMEDIA_TYPE_VIDEO,
+        .filter_frame = filter_frame,
+        .config_props = config_input,
+    },
+    { NULL }
+};
+
+static const AVFilterPad ow_outputs[] = {
+     {
+         .name = "default",
+         .type = AVMEDIA_TYPE_VIDEO,
+     },
+     { NULL }
+};
+
+AVFilter avfilter_vf_ow = {
+    .name          = "ow",
+    .description   = NULL_IF_CONFIG_SMALL("Denoise using wavelets."),
+    .priv_size     = sizeof(OWContext),
+    .uninit        = uninit,
+    .query_formats = query_formats,
+    .inputs        = ow_inputs,
+    .outputs       = ow_outputs,
+    .priv_class    = &ow_class,
+    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE,
+};
-- 
1.8.2.2



More information about the ffmpeg-devel mailing list