[FFmpeg-devel] [PATCH] lavfi: add kerndeint filter
Clément Bœsch
ubitux at gmail.com
Fri Dec 28 18:18:27 CET 2012
On Fri, Dec 28, 2012 at 04:41:45PM +0100, Stefano Sabatini wrote:
> On date Thursday 2012-10-18 09:50:53 +0200, Jérémy Tran encoded:
> > This is a port of the vf_kerndeint filter (libmpcodecs/vf_kerndeint) by
> > Donal A. Graft (original avisynth plugin author).
> >
> > The filter works fine using YUV colorspace.
> > When using RGB, it is not binary equal with mp=kerndeint (the output video
> > looks fine so this may be a little mistake I did somewhere).
> >
> > I also removed many non-32bits colorspaces since the alogrithm does not
> > seem to be made to work on shorter pixels, is this okay ?
> > The only non-32-bits colorspace left is YUYV422 which is what mp=kerndeint
> > uses, but I don't seem to get the same results (my first thought was
> > because it does not have 32bpp).
> >
> > The FATE test will come after this.
> > ---
> > configure | 1 +
> > doc/filters.texi | 45 ++++++
> > libavfilter/Makefile | 1 +
> > libavfilter/allfilters.c | 1 +
> > libavfilter/vf_kerndeint.c | 377 +++++++++++++++++++++++++++++++++++++++++++++
> > 5 files changed, 425 insertions(+)
> > create mode 100644 libavfilter/vf_kerndeint.c
>
> Updated work in progress.
>
> TODO: add a FATE test, fix the yuyv422 which is broken (same in
> mp=kerndeint), add support to more formats.
> --
> FFmpeg = Faithful & Free Mega Philosofic Erroneous Gorilla
> From ec015bfbea71a0d0de7f367b8c56040731b484c3 Mon Sep 17 00:00:00 2001
> From: Stefano Sabatini <stefasab at gmail.com>
> Date: Thu, 18 Oct 2012 09:50:53 +0200
> Subject: [PATCH] lavfi: add kerndeint filter
> MIME-Version: 1.0
> Content-Type: text/plain; charset=UTF-8
> Content-Transfer-Encoding: 8bit
>
> This is a port of the vf_kerndeint filter (libmpcodecs/vf_kerndeint) by
> Donal A. Graft (original avisynth plugin author), and is based on the
> work by Jérémy Tran <tran.jeremy.av at gmail.com> done for SOCIS 2012.
> ---
> configure | 1 +
> doc/filters.texi | 62 ++++++++
> libavfilter/Makefile | 1 +
> libavfilter/allfilters.c | 1 +
> libavfilter/vf_kerndeint.c | 375 ++++++++++++++++++++++++++++++++++++++++++++
> 5 files changed, 440 insertions(+)
> create mode 100644 libavfilter/vf_kerndeint.c
>
> diff --git a/configure b/configure
> index c8b8c2b..6c489e9 100755
> --- a/configure
> +++ b/configure
> @@ -1986,6 +1986,7 @@ frei0r_src_filter_extralibs='$ldl'
> geq_filter_deps="gpl"
> hqdn3d_filter_deps="gpl"
> hue_filter_deps="gpl"
> +kerndeint_filter_deps="gpl"
Required an entry in LICENSE too.
> movie_filter_deps="avcodec avformat"
> mp_filter_deps="gpl avcodec swscale inline_asm"
> mptestsrc_filter_deps="gpl"
> diff --git a/doc/filters.texi b/doc/filters.texi
> index 765a9b4..9cd6ac1 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -2862,6 +2862,68 @@ 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
> +
> +Deinterlace input video by applying Donald Graft's adaptive kernel
> +deinterling. Work on interlaced parts of a video to produce
> +progressive frames.
> +
> +This filter accepts parameters as a list of @var{key}=@var{value}
> +pairs, separated by ":". The description of the accepted parameters
> +follows.
> +
> +If the key of the first options is omitted, the arguments are
> +interpreted according to the following syntax:
> + at var{thresh}:@var{map}:@var{order}:@var{sharp}:@var{twoway}
> +
> + at table @option
> + at item thresh
> +Set the threshold which affects the filter's tolerance when
> +determining if a pixel line must be processed. Must be an integer in
> +the range [0,255] and defaults to 10. A value of 0 will result in
> +applying the process on every pixels.
> +
> + at item map
> +Must be 0 (ignore pixels exceeding the threshold) or 1 (paint pixels
> +exceeding the threshold white) and defaults to 0.
> +
> + at item order
> +Must be 0 (leave fields alone) or 1 (swap fields) and defaults to 0.
> +
> + at item sharp
> +Must be 0 (disable additional sharpening) or 1 (enable additional
> +sharpening) and defaults to 0.
> +
> + at item twoway
> +Must be 0 (disable twoway sharpening) or 1 (enable twoway sharpening)
> +and defaults to 0.
> +
> + at item deinterlace
> +Disable deinterlacing if set to 0. Default is 0.
nit: I'd say "Enable deinterlacing if set to 1".
> + at end table
> +
> + at subsection Examples
> +
> + at itemize
> + at item
> +Apply default values:
> + at example
> +kerndeint=thresh=10:map=0:order=0:sharp=0:twoway=0
> + at end example
> +
> + at item
> +Enable additional sharpening:
> + at example
> +kerndeint=sharp=1
> + at end example
> +
> + at item
> +Paint processed pixels in white:
> + at example
> +kerndeint=map=1
> + 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 34baa6d..b09b59a 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -112,6 +112,7 @@ 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 88b0979..fb192d6 100644
> --- a/libavfilter/allfilters.c
> +++ b/libavfilter/allfilters.c
> @@ -103,6 +103,7 @@ 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..052d2a6
> --- /dev/null
> +++ b/libavfilter/vf_kerndeint.c
> @@ -0,0 +1,375 @@
> +/*
> + * 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/imgutils.h"
> +#include "libavutil/opt.h"
> +#include "libavutil/pixdesc.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 {
> + const AVClass *class;
> + 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
> + int bpp; ///< bytes per pixel
> +} KerndeintContext;
> +
> +#define OFFSET(x) offsetof(KerndeintContext, x)
> +#define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
> +static const AVOption kerndeint_options[] = {
> + { "thresh", "set the threshold", OFFSET(thresh), AV_OPT_TYPE_INT, { 10 }, 0, 255, FLAGS },
> + { "map", "set the map", OFFSET(map), AV_OPT_TYPE_INT, { 0 }, 0, 1, FLAGS },
> + { "order", "set the order", OFFSET(order), AV_OPT_TYPE_INT, { 0 }, 0, 1, FLAGS },
> + { "sharp", "set the sharpening mode", OFFSET(sharp), AV_OPT_TYPE_INT, { 0 }, 0, 1, FLAGS },
> + { "twoway", "activate twoway", OFFSET(twoway), AV_OPT_TYPE_INT, { 0 }, 0, 1, FLAGS },
> + { "deinterlace", "disable interlacing", OFFSET(do_deinterlace), AV_OPT_TYPE_INT, { 1 }, 0, 1, FLAGS },
> + { NULL }
> +};
> +
> +AVFILTER_DEFINE_CLASS(kerndeint);
> +
> +static av_cold int init(AVFilterContext *ctx, const char *args)
> +{
> + KerndeintContext *kerndeint = ctx->priv;
> + int i;
> + const char const * shorthand[] = { "thresh", "map", "order", "sharp", "twoway", NULL };
> +
> + kerndeint->class = &kerndeint_class;
> + kerndeint->do_deinterlace = 1;
Looks unnecessary
> + for (i = 0; i < 4; ++i)
> + kerndeint->temp_data[i] = NULL;
ditto
> + av_opt_set_defaults(kerndeint);
> +
> + return av_opt_set_from_string(kerndeint, args, shorthand, "=", ":");
> +}
> +
> +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]);
> + av_opt_free(kerndeint);
> +}
> +
> +static int query_formats(AVFilterContext *ctx)
> +{
> + static const enum PixelFormat pix_fmts[] = {
> + PIX_FMT_YUV420P,
> + PIX_FMT_YUVJ420P,
> + PIX_FMT_YUYV422,
> + PIX_FMT_ARGB,
> + 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;
> + kerndeint->bpp = av_get_bits_per_pixel(desc) / 8;
> +
> + for (i = 0; i < 4; ++i)
> + kerndeint->temp_data[i] = av_mallocz(inlink->h * inlink->w * kerndeint->bpp);
> +
Missing alloc check
> + switch (inlink->format) {
> + case PIX_FMT_YUV420P:
> + kerndeint->format = YUV;
> + break;
> + case PIX_FMT_YUYV422:
> + kerndeint->format = YUYV;
> + break;
> + case PIX_FMT_ARGB:
> + kerndeint->format = RGB;
> + break;
> + }
> +
> + return 0;
> +}
> +
> +#define AVWN32(buf, value) *(uint32_t *)(buf) = value
Don't we have safer macro for this?
> +
> +static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *inpic)
> +{
> + KerndeintContext *kerndeint = inlink->dst->priv;
> + AVFilterLink *outlink = inlink->dst->outputs[0];
> + AVFilterBufferRef *outpic;
> + int cw = inlink->w >> kerndeint->hsub;
> + int ch = inlink->h >> kerndeint->vsub;
> + const unsigned char *prvp; ///< Previous field's pixel line number n
> + const unsigned char *prvpp; ///< Previous field's pixel line number (n - 1)
> + const unsigned char *prvpn; ///< Previous field's pixel line number (n + 1)
> + const unsigned char *prvppp; ///< Previous field's pixel line number (n - 2)
> + const unsigned char *prvpnn; ///< Previous field's pixel line number (n + 2)
> + const unsigned char *prvp4p; ///< Previous field's pixel line number (n - 4)
> + const unsigned char *prvp4n; ///< Previous field's pixel line number (n + 4)
> +
> + const unsigned char *srcp; ///< Current field's pixel line number n
> + const unsigned char *srcpp; ///< Current field's pixel line number (n - 1)
> + const unsigned char *srcpn; ///< Current field's pixel line number (n + 1)
> + const unsigned char *srcppp; ///< Current field's pixel line number (n - 2)
> + const unsigned char *srcpnn; ///< Current field's pixel line number (n + 2)
> + const unsigned char *srcp3p; ///< Current field's pixel line number (n - 3)
> + const unsigned char *srcp3n; ///< Current field's pixel line number (n + 3)
> + const unsigned char *srcp4p; ///< Current field's pixel line number (n - 4)
> + const unsigned char *srcp4n; ///< Current field's pixel line number (n + 4)
> +
nit: uint8_t
> + unsigned char *dstp, *dstp_saved;
> + const unsigned char *srcp_saved;
> +
> + 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)
> + return ff_filter_frame(outlink, inpic);
> +
> + outpic = ff_get_video_buffer(outlink, AV_PERM_WRITE|AV_PERM_ALIGN, outlink->w, outlink->h);
> + if (!outpic) {
> + avfilter_unref_bufferp(&inpic);
> + return AVERROR(ENOMEM);
> + }
> + avfilter_copy_buffer_ref_props(outpic, inpic);
> +
> + for (plane = 0; inpic->data[plane] && plane < 4; plane++) {
You should check for plane before dereferencing
> + h = plane == PLANAR_Y ? inlink->h : ch;
> + w = plane == PLANAR_Y ? inlink->w * kerndeint->bpp : cw;
not sure: no need to apply bpp to 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)) {
nit: looks like there is a bunch of pointless ()
> + if (map == 1) {
> + g = x & ~3;
> +
> + if (kerndeint->format == RGB) {
> + AVWN32(dstp + g, 0xffffffff);
> + x = g + 3;
> + } else if (kerndeint->format == YUYV) {
> + AVWN32(dstp + g, 0xeb80eb80);
> + x = g + 3;
> + } else {
> + 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] = 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);
> + }
> +
> + avfilter_unref_buffer(inpic);
> + return ff_filter_frame(outlink, outpic);
> +}
> +
> +AVFilter avfilter_vf_kerndeint = {
> + .name = "kerndeint",
> + .description = NULL_IF_CONFIG_SMALL("Kernel deinterlacer"),
> +
> + .priv_size = sizeof(KerndeintContext),
> +
> + .init = init,
> + .uninit = uninit,
> + .query_formats = query_formats,
> +
> + .inputs = (const AVFilterPad[]) {
> + {
> + .name = "default",
> + .type = AVMEDIA_TYPE_VIDEO,
> + .filter_frame = filter_frame,
> + .config_props = config_props,
> + .min_perms = AV_PERM_READ,
> + },
> + { .name = NULL }
> + },
> + .outputs = (const AVFilterPad[]) {
> + {
> + .name = "default",
> + .type = AVMEDIA_TYPE_VIDEO,
> + },
> + { .name = NULL }
> + },
Please split this like the other filters
> + .priv_class = &kerndeint_class,
> +};
--
Clément B.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 490 bytes
Desc: not available
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20121228/9cded55e/attachment.asc>
More information about the ffmpeg-devel
mailing list