[FFmpeg-devel] [PATCH] lavfi: add kerndeint filter

Stefano Sabatini stefasab at gmail.com
Sun Sep 30 11:56:42 CEST 2012


On date Saturday 2012-09-29 13:00:46 +0200, Jérémy Tran encoded:
> This is a port of the vf_kerndeint filter (libmpcodecs/vf_kerndeint) by
> Donald A. Graft (original avisynth plugin author).
> 
> Once this patch is ready to be pushed, a following patch will implement a
> process_command callback to enable or disable the deinterlacing process
> at runtime.
> ---
>  configure                  |   1 +
>  doc/filters.texi           |  42 +++++
>  libavfilter/Makefile       |   1 +
>  libavfilter/allfilters.c   |   1 +
>  libavfilter/vf_kerndeint.c | 385 +++++++++++++++++++++++++++++++++++++++++++++
>  5 files changed, 430 insertions(+)
>  create mode 100644 libavfilter/vf_kerndeint.c
> 
> diff --git a/configure b/configure
> index b669523..772dd2d 100755
> --- a/configure
> +++ b/configure
> @@ -1907,6 +1907,7 @@ frei0r_filter_extralibs='$ldl'
>  frei0r_src_filter_deps="frei0r dlopen"
>  frei0r_src_filter_extralibs='$ldl'
>  hqdn3d_filter_deps="gpl"
> +kerndeint_filter_deps="gpl"
>  movie_filter_deps="avcodec avformat"
>  mp_filter_deps="gpl avcodec swscale postproc inline_asm"
>  mptestsrc_filter_deps="gpl"
> diff --git a/doc/filters.texi b/doc/filters.texi
> index 916aa6f..bb8a1a0 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -2431,6 +2431,48 @@ If a parameter is omitted, it is kept at its current value.
>  Interlaceing detect filter. This filter tries to detect if the input is
>  interlaced or progressive. Top or bottom field first.
>  
> + at section kerndeint
> +
> +Donald Graft's adaptive kernel deinterlacer. Work on interlaced parts of a video
> +to produce progressive frames.
> +
> +This @var{kerndeint} filter accepts the following optional parameters:
> +[@var{threshold}[:@var{map}[:@var{order}[:@var{sharp}[:@var{twoway}]]]]]
> +
> + at var{threshold} must be an integer in the range [0,255] and defaults to 10.
> +Affect the filter's tolerance when determining if a pixel line must be
> +processed. A value of 0 will result in applying the process on every pixels.
> +
> + at var{map} must be 0 (ignore pixels exceeding the threshold) or 1 (paint pixels
> +exceeding the threshold white) and defaults to 0.
> +
> + at var{order} must be 0 (leave fields alone) or 1 (swap fields) and defaults to 0.
> +
> + at var{sharp} must be 0 (disable additional sharpening) or 1 (enable additional
> +sharpening) and defaults to 0.
> +
> + at var{twoway} must be 0 (disable twoway sharpening) or 1 (enable twoway
> +sharpening) and defaults to 0.
> +

> +Some examples follow:

> + at itemize
> + at item Those two commands are equivalent

@item
Those two commands are equivalent

or the man output will be mangled, same below.

> + at example
> +kerndeint
> +kerndeint=10:0:0:0:0
> + at end example
> +
> + at item Enable additional sharpening
> + at example
> +kerndeint=10:0:0:1:0
> + at end example
> +
> + at item Paint processed pixels in white
> + at example
> +kerndeint=10:1:0:0:0
> + at end example
> + at end itemize

> +
>  @section lut, lutrgb, lutyuv
>  
>  Compute a look-up table for binding each pixel component input value
> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> index ed1b9b5..2f549cc 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -108,6 +108,7 @@ OBJS-$(CONFIG_HFLIP_FILTER)                  += vf_hflip.o
>  OBJS-$(CONFIG_HQDN3D_FILTER)                 += vf_hqdn3d.o
>  OBJS-$(CONFIG_HUE_FILTER)                    += vf_hue.o
>  OBJS-$(CONFIG_IDET_FILTER)                   += vf_idet.o
> +OBJS-$(CONFIG_KERNDEINT_FILTER)              += vf_kerndeint.o
>  OBJS-$(CONFIG_LUT_FILTER)                    += vf_lut.o
>  OBJS-$(CONFIG_LUTRGB_FILTER)                 += vf_lut.o
>  OBJS-$(CONFIG_LUTYUV_FILTER)                 += vf_lut.o
> diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
> index 49f166d..e72e95a 100644
> --- a/libavfilter/allfilters.c
> +++ b/libavfilter/allfilters.c
> @@ -100,6 +100,7 @@ void avfilter_register_all(void)
>      REGISTER_FILTER (HQDN3D,      hqdn3d,      vf);
>      REGISTER_FILTER (HUE,         hue,         vf);
>      REGISTER_FILTER (IDET,        idet,        vf);
> +    REGISTER_FILTER (KERNDEINT,   kerndeint,   vf);
>      REGISTER_FILTER (LUT,         lut,         vf);
>      REGISTER_FILTER (LUTRGB,      lutrgb,      vf);
>      REGISTER_FILTER (LUTYUV,      lutyuv,      vf);
> diff --git a/libavfilter/vf_kerndeint.c b/libavfilter/vf_kerndeint.c
> new file mode 100644
> index 0000000..5ea0a8d
> --- /dev/null
> +++ b/libavfilter/vf_kerndeint.c
> @@ -0,0 +1,385 @@
> +/*
> + * Copyright (c) 2012 Jeremy Tran
> + * Copyright (c) 2004 Tobias Diedrich
> + * Copyright (c) 2003 Donald A. Graft
> + *
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either version 2 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 General Public License for more details.
> + *
> + * You should have received a copy of the GNU 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
> + * Kernel Deinterlacer
> + * Ported from MPlayer libmpcodecs/vf_kerndeint.c.
> + */
> +
> +#include "libavutil/pixdesc.h"
> +#include "libavutil/imgutils.h"
> +
> +#include "avfilter.h"
> +#include "formats.h"
> +#include "internal.h"
> +
> +#define PLANAR_Y 0
> +#define PLANAR_U 1
> +#define PLANAR_V 2
> +
> +#define RGB  0
> +#define YUYV 1
> +#define YUV  2
> +
> +typedef struct {
> +    int     frame;
> +    int     map;
> +    int     order;
> +    int     thresh;
> +    int     sharp;
> +    int     twoway;
> +    int     do_deinterlace;
> +    int     hsub;
> +    int     vsub;
> +    int     format;        ///< Pixel format: RGB, YUYV or YUV
> +    uint8_t *temp_data[4]; ///< Temporary video plane data buffer
> +} KerndeintContext;
> +
> +#define CHECK_PARAM(param, min, max)                                \
> +    do {                                                            \
> +        if (param < min || param > max) {                           \
> +            av_log(ctx, AV_LOG_ERROR,                               \
> +                   "Invalid " #param " value : must be an integer " \
> +                   "between %d and %d\n",                           \
> +                   min, max);                                       \
> +            ret = AVERROR(EINVAL);                                  \
> +        }                                                           \
> +    } while (0)
> +
> +static av_cold int init(AVFilterContext *ctx, const char *args)
> +{
> +    KerndeintContext *kerndeint = ctx->priv;
> +    int ret = 0, i;
> +    char c1 = ':', c2 = ':', c3 = ':', c4 = ':';
> +
> +    // default values
> +    kerndeint->thresh = 10;
> +    kerndeint->map    = 0;
> +    kerndeint->order  = 0;
> +    kerndeint->sharp  = 0;
> +    kerndeint->twoway = 0;
> +
> +    kerndeint->do_deinterlace = 1;
> +    for (i = 0; i < 4; ++i)
> +        kerndeint->temp_data[i] = NULL;
> +

> +    if (args)
> +        sscanf(args, "%d%c%d%c%d%c%d%c%d",
> +               &kerndeint->thresh, &c1, &kerndeint->map,   &c2,
> +               &kerndeint->order,  &c3, &kerndeint->sharp, &c4, &kerndeint->twoway);
> +
> +    if (c1 != ':' || c2 != ':' || c3 != ':' || c4 != ':') {
> +        av_log(ctx, AV_LOG_ERROR,
> +               "Invalid syntax, must be: "
> +               "kerndeint[=threshold[:map[:order[:sharp[:twoway]]]]]\n");
> +        return AVERROR(EINVAL);
> +    }
> +
> +    CHECK_PARAM(kerndeint->thresh, 0, 255);
> +    CHECK_PARAM(kerndeint->map,    0, 1);
> +    CHECK_PARAM(kerndeint->order,  0, 1);
> +    CHECK_PARAM(kerndeint->sharp,  0, 1);
> +    CHECK_PARAM(kerndeint->twoway, 0, 1);

This could be simplified making use of av_opt_set_from_string().

> +
> +    return ret;
> +}
> +
> +static av_cold void uninit(AVFilterContext *ctx)
> +{
> +    KerndeintContext *kerndeint = ctx->priv;
> +    int i;
> +
> +    for (i = 0; i < 4; ++i)
> +        av_free(kerndeint->temp_data[i]);
> +}
> +
> +static int query_formats(AVFilterContext *ctx)
> +{
> +    static const enum PixelFormat pix_fmts[] = {
> +        PIX_FMT_YUV444P,      PIX_FMT_YUV422P,
> +        PIX_FMT_YUV420P,      PIX_FMT_YUV411P,
> +        PIX_FMT_YUV410P,      PIX_FMT_YUV440P,
> +        PIX_FMT_YUYV422,
> +        PIX_FMT_RGB24,        PIX_FMT_BGR24,
> +        PIX_FMT_NONE
> +    };
> +
> +    ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
> +
> +    return 0;
> +}
> +
> +static int config_props(AVFilterLink *inlink)
> +{
> +    KerndeintContext *kerndeint = inlink->dst->priv;
> +    const AVPixFmtDescriptor *desc = &av_pix_fmt_descriptors[inlink->format];
> +    int i;
> +
> +    kerndeint->hsub = desc->log2_chroma_w;
> +    kerndeint->vsub = desc->log2_chroma_h;
> +
> +    for (i = 0; i < 4; ++i)
> +        kerndeint->temp_data[i] = av_mallocz(inlink->h * inlink->w);
> +
> +    switch (inlink->format) {
> +        case PIX_FMT_YUV444P:
> +        case PIX_FMT_YUV422P:
> +        case PIX_FMT_YUV420P:
> +        case PIX_FMT_YUV411P:
> +        case PIX_FMT_YUV410P:
> +        case PIX_FMT_YUV440P:
> +            kerndeint->format = YUV;
> +            break;
> +        case PIX_FMT_YUYV422:
> +            kerndeint->format = YUYV;
> +            break;
> +        case PIX_FMT_RGB24:
> +        case PIX_FMT_BGR24:
> +            kerndeint->format = RGB;
> +            break;
> +    }
> +
> +    return 0;
> +}
> +
> +static int end_frame(AVFilterLink *inlink)
> +{
> +    KerndeintContext *kerndeint = inlink->dst->priv;
> +    AVFilterBufferRef *inpic  = inlink->cur_buf;
> +    AVFilterBufferRef *outpic = inlink->dst->outputs[0]->out_buf;
> +    int cw = inlink->w >> kerndeint->hsub;
> +    int ch = inlink->h >> kerndeint->vsub;

> +    const unsigned char *prvp, *prvpp, *prvpn, *prvpnn, *prvppp, *prvp4p, *prvp4n;
> +    const unsigned char *srcp_saved;
> +    const unsigned char *srcp, *srcpp, *srcpn, *srcpnn, *srcppp, *srcp3p, *srcp3n, *srcp4p, *srcp4n;
> +    unsigned char *dstp, *dstp_saved;

Could you add some doxy here (or maybe using less obfuscated names)?
Yes I know this was in the original code.

> +    int src_linesize;
> +    int psrc_linesize;
> +    int dst_linesize;
> +    int x, y, plane, g;
> +    int n = kerndeint->frame++;
> +    int val, hi, lo, w, h;
> +    double valf;
> +    const int threshold = kerndeint->thresh;
> +    const int order     = kerndeint->order;
> +    const int map       = kerndeint->map;
> +    const int sharp     = kerndeint->sharp;
> +    const int twoway    = kerndeint->twoway;
> +
> +    if (kerndeint->do_deinterlace) {
> +        for (plane = 0; inpic->data[plane] && plane < 4; plane++) {
> +            h = plane == PLANAR_Y ? inlink->h : ch;
> +            w = plane == PLANAR_Y ? inlink->w : cw;
> +
> +            srcp = srcp_saved = inpic->data[plane];
> +            src_linesize      = inpic->linesize[plane];
> +            psrc_linesize     = outpic->linesize[plane];
> +            dstp = dstp_saved = outpic->data[plane];
> +            dst_linesize      = outpic->linesize[plane];
> +            srcp              = srcp_saved + (1 - order) * src_linesize;
> +            dstp              = dstp_saved + (1 - order) * dst_linesize;
> +
> +            for (y = 0; y < h; y += 2) {
> +                memcpy(dstp, srcp, w);
> +                srcp += 2 * src_linesize;
> +                dstp += 2 * dst_linesize;
> +            }
> +
> +            // Copy through the lines that will be missed below.
> +            memcpy(dstp_saved + order           * dst_linesize, srcp_saved + (1 - order)     * src_linesize, w);
> +            memcpy(dstp_saved + (2 + order)     * dst_linesize, srcp_saved + (3 - order)     * src_linesize, w);
> +            memcpy(dstp_saved + (h - 2 + order) * dst_linesize, srcp_saved + (h - 1 - order) * src_linesize, w);
> +            memcpy(dstp_saved + (h - 4 + order) * dst_linesize, srcp_saved + (h - 3 - order) * src_linesize, w);
> +
> +            /* For the other field choose adaptively between using the previous field
> +               or the interpolant from the current field. */
> +            prvp   = kerndeint->temp_data[plane] + 5 * psrc_linesize - (1 - order) * psrc_linesize;
> +            prvpp  = prvp - psrc_linesize;
> +            prvppp = prvp - 2 * psrc_linesize;
> +            prvp4p = prvp - 4 * psrc_linesize;
> +            prvpn  = prvp + psrc_linesize;
> +            prvpnn = prvp + 2 * psrc_linesize;
> +            prvp4n = prvp + 4 * psrc_linesize;
> +
> +            srcp   = srcp_saved + 5 * src_linesize - (1 - order) * src_linesize;
> +            srcpp  = srcp - src_linesize;
> +            srcppp = srcp - 2 * src_linesize;
> +            srcp3p = srcp - 3 * src_linesize;
> +            srcp4p = srcp - 4 * src_linesize;
> +
> +            srcpn  = srcp + src_linesize;
> +            srcpnn = srcp + 2 * src_linesize;
> +            srcp3n = srcp + 3 * src_linesize;
> +            srcp4n = srcp + 4 * src_linesize;
> +
> +            dstp   = dstp_saved  + 5 * dst_linesize - (1 - order) * dst_linesize;
> +
> +            for (y = 5 - (1 - order); y <= h - 5 - (1 - order); y += 2) {
> +                for (x = 0; x < w; x++) {
> +                    if ((threshold == 0) || (n == 0) ||
> +                        (abs((int)prvp[x]  - (int)srcp[x])  > threshold) ||
> +                        (abs((int)prvpp[x] - (int)srcpp[x]) > threshold) ||
> +                        (abs((int)prvpn[x] - (int)srcpn[x]) > threshold)) {
> +                        if (map == 1) {
> +                            g = x & ~3;
> +

> +                            if (kerndeint->format == RGB) {
> +                                dstp[g++] = 255;
> +                                dstp[g++] = 255;
> +                                dstp[g++] = 255;
> +                                dstp[g] = 255;

possibly faster:
         *(uint32_t *)(dstp + g) = 0xffffffff;
         x = g+4;

Also I'm not sure that this works with RGB24.

> +                                x = g;
> +                            } else if (kerndeint->format == YUYV) {

> +                                dstp[g++] = 235;
> +                                dstp[g++] = 128;
> +                                dstp[g++] = 235;
> +                                dstp[g] = 128;
> +                                x = g;

same here:
     AVWN32(dstp + g, 0xeb80eb80);
     x = g+4;

> +                            } else {
> +                                if (plane == PLANAR_Y)
> +                                    dstp[x] = 235;
> +                                else
> +                                    dstp[x] = 128;

Nit: dstp[x] = plane == PLANAR_Y ? 235 : 128;


> +                            }
> +                        } else {
> +                            if (kerndeint->format == RGB) {
> +                                hi = 255;
> +                                lo = 0;
> +                            } else if (kerndeint->format == YUYV) {
> +                                hi = (x & 1) ? 240 : 235;
> +                                lo = 16;
> +                            } else {
> +                                hi = (plane == PLANAR_Y) ? 235 : 240;
> +                                lo = 16;
> +                            }
> +
> +                            if (sharp == 1) {
> +                                if (twoway == 1) {
> +                                    valf =   0.526 * ((int)srcpp[x] + (int)srcpn[x])
> +                                           + 0.170 * ((int)srcp[x] + (int)prvp[x])
> +                                           - 0.116 * ((int)srcppp[x]
> +                                                      + (int)srcpnn[x]
> +                                                      + (int)prvppp[x]
> +                                                      + (int)prvpnn[x])
> +                                           - 0.026 * ((int)srcp3p[x] + (int)srcp3n[x])
> +                                           + 0.031 * ((int)srcp4p[x]
> +                                                      + (int)srcp4n[x]
> +                                                      + (int)prvp4p[x]
> +                                                      + (int)prvp4n[x]);
> +                                } else {
> +                                    valf =   0.526 * ((int)srcpp[x] + (int)srcpn[x])
> +                                           + 0.170 * ((int)prvp[x])
> +                                           - 0.116 * ((int)prvppp[x] + (int)prvpnn[x])
> +                                           - 0.026 * ((int)srcp3p[x] + (int)srcp3n[x])
> +                                           + 0.031 * ((int)prvp4p[x] + (int)prvp4p[x]);
> +                                }
> +
> +                                valf = av_clip(valf, lo, hi);
> +                                dstp[x] = (int) valf;
> +                            } else {
> +                                if (twoway == 1) {
> +                                    val = (8 * ((int)srcpp[x] + (int)srcpn[x])
> +                                           + 2 * ((int)srcp[x] + (int)prvp[x])
> +                                           - (int)(srcppp[x]) - (int)(srcpnn[x])
> +                                           - (int)(prvppp[x]) - (int)(prvpnn[x]))
> +                                          >> 4;
> +                                } else {
> +                                    val = (8 * ((int)srcpp[x]
> +                                           + (int)srcpn[x])
> +                                           + 2 * ((int)prvp[x])
> +                                           - (int)(prvppp[x]) - (int)(prvpnn[x]))
> +                                          >> 4;
> +                                }
> +
> +                                val = av_clip(val, lo, hi);
> +                                dstp[x] = (int) val;
> +                            }
> +                        }
> +                    } else {
> +                        dstp[x] = srcp[x];
> +                    }
> +                }
> +                prvp   += 2 * psrc_linesize;
> +                prvpp  += 2 * psrc_linesize;
> +                prvppp += 2 * psrc_linesize;
> +                prvpn  += 2 * psrc_linesize;
> +                prvpnn += 2 * psrc_linesize;
> +                prvp4p += 2 * psrc_linesize;
> +                prvp4n += 2 * psrc_linesize;
> +                srcp   += 2 * src_linesize;
> +                srcpp  += 2 * src_linesize;
> +                srcppp += 2 * src_linesize;
> +                srcp3p += 2 * src_linesize;
> +                srcp4p += 2 * src_linesize;
> +                srcpn  += 2 * src_linesize;
> +                srcpnn += 2 * src_linesize;
> +                srcp3n += 2 * src_linesize;
> +                srcp4n += 2 * src_linesize;
> +                dstp   += 2 * dst_linesize;
> +            }
> +
> +            srcp = inpic->data[plane];
> +            dstp = kerndeint->temp_data[plane];
> +            av_image_copy_plane(dstp, psrc_linesize, srcp, src_linesize, w, h);
> +        }
> +    } else {
> +        for (plane = 0; inpic->data[plane] && plane < 4; plane++) {
> +            h = plane == PLANAR_Y ? inlink->h : ch;
> +            w = plane == PLANAR_Y ? inlink->w : cw;
> +            av_image_copy_plane(outpic->data[plane], outpic->linesize[plane],
> +                                inpic->data[plane],  inpic->linesize[plane],
> +                                w, h);
> +        }
> +    }
> +
> +    return ff_end_frame(inlink->dst->outputs[0]);
> +}

It would be useful to add a test for it (checking all the supported
pixel formats), possibly making use of tinterlace.

Seems fine otherwise, assuming the output is binary exact with
mp=kerndeint.
-- 
FFmpeg = Fostering Fascinating Mastodontic Portentous Erudite Genius


More information about the ffmpeg-devel mailing list