[FFmpeg-devel] rectification filter
Timothy Gu
timothygu99 at gmail.com
Sun Aug 3 19:11:04 CEST 2014
On 08/03/2014 08:39 AM, Daniel Oberhoff wrote:
>
> Hello all,
>
> I updated the patch trying to incorporate all review feedback. I also got consent from the original author to put this filter under LGPL, and thus have it compiled in by default. I also conversed with Cyrille from Krita
> and he doesn’t see any more copyright issues, as at that point it boils down to the use of a well known algorithm, in a straight-forward implementation.
>
> Compared to the previous patch I renamed the filter to lenscorrection, since that is the name in the frei0r suite, and it will make using this instead of the frei0r one as easy as stating
>
> lenscorrection=...
>
> instead of
>
> frei0r=lenscorrection:...
>
> Last but not least I adapted the format list, inspected results on all formats, and supplied a fate test. Release notes and docs where also adapted. Please check if this is ok to push now.
>
> From dc552ae06a41725988250896327af2cceee1b812 Mon Sep 17 00:00:00 2001
> From: Daniel Oberhoff <daniel at danieloberhoff.de>
> Date: Mon, 28 Jul 2014 23:58:12 +0200
> Subject: [PATCH] ported lenscorrection filter from frei0r
avfilter: port lenscorrection filter from frei0r
>
> ---
> Changelog | 2 +-
> doc/filters.texi | 36 +++++
> libavfilter/Makefile | 1 +
> libavfilter/allfilters.c | 1 +
> libavfilter/version.h | 4 +-
> libavfilter/vf_lenscorrection.c | 208 +++++++++++++++++++++++++++
> tests/fate/filter-video.mak | 3 +
> tests/ref/fate/filter-pixfmts-lenscorrection | 8 ++
> 8 files changed, 260 insertions(+), 3 deletions(-)
> create mode 100644 libavfilter/vf_lenscorrection.c
> create mode 100644 tests/ref/fate/filter-pixfmts-lenscorrection
If you would like to maintain this filter, you can choose to add your
name to the MAINTAINERS file.
>
> diff --git a/Changelog b/Changelog
> index 067f72a..3c1ee51 100644
> --- a/Changelog
> +++ b/Changelog
> @@ -2,7 +2,7 @@ Entries are sorted chronologically from oldest to youngest within each release,
> releases are sorted from youngest to oldest.
>
> version <next>:
> -
> +- ported lenscorrection filter from frei0r filter
Just say frei0r and drop the "filter"
>
> version 2.3:
> - AC3 fixed-point decoding
> diff --git a/doc/filters.texi b/doc/filters.texi
> index c5caa77..2af311a 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -5532,6 +5532,42 @@ kerndeint=map=1
> @end example
> @end itemize
>
> + at section lenscorrection
> +
Add an one-line description here:
Compensates for lens distortion.
or
Correct lens distortion.
> +This filter can be used to correct for radial distortion as can result from the use
> +of wide angle lenses, and thereby re-rectify the image. To find the right parameters
> +one can use tools available for example as part of opencv or simply trial-and-error.
OpenCV
Can you offer some examples on how to find the right parameters? FFmpeg
has support for libopencv. Can you use that?
> +Note that effectively the same filter is available in the open-source tools Krita and
> +Digikam from the KDE project.
Add an empty line here to make the two paragraphs separate on the built
documentation.
> +In contrast to the vignette filter, which can also be used to compensate lens errors,
Use @ref{vignette} here, and add @anchor{vignette} before @section
vignette later in the file.
> +this filter corrects the distortion of the image, whereas vignette corrects the
> +brightness distribution, so you may want to use both filters together in certain
> +cases, though you will have to take care of ordering, i.e. wether vignette should
whether
I also think you should probably use regular English and not filter
names here, which makes the documentation more readable. So it's
"vignetting" here, and "lens correction" below.
> +be applied before or after lenscorrection.
> +
Add `@subsection Options` here.
> +The filter accepts the following options:
> +
> + at table @option
> + at item cx
> +Relative x-coordinate of the focal point of the image, and thereby the center of the
> +distrortion. This value has a range [0,1] and is expressed as fractions of the image
> +width.
> + at item cy
> +Relative y-coordinate of the focal point of the image, and thereby the center of the
> +distrortion. This value has a range [0,1] and is expressed as fractions of the image
> +height.
> + at item k1
> +Coefficient of the quadratic correction term. 0.5 means no correction.
> + at item k2
> +Coefficient of the double quadratic correction term. 0.5 means no correction.
> + at end table
> +
> +The formula that generates the correction is:
> +
> +r_src = r_tgt * (1 + (k1 - 0.5) * (r_tgt/r_0)^2 + (k2 - 0.5) * (r_tgt/r_0)^4)
Please add @var{} to all of the variables, or, if you prefer, surround
this block with @example and @end example.
> +
> +where r_0 is halve of the image diagonal.
> +
> @anchor{lut3d}
> @section lut3d
>
> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> index 0f54381..f74defa 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -162,6 +162,7 @@ OBJS-$(CONFIG_PIXDESCTEST_FILTER) += vf_pixdesctest.o
> OBJS-$(CONFIG_PP_FILTER) += vf_pp.o
> OBJS-$(CONFIG_PSNR_FILTER) += vf_psnr.o dualinput.o framesync.o
> OBJS-$(CONFIG_PULLUP_FILTER) += vf_pullup.o
> +OBJS-$(CONFIG_LENSCORRECTION_FILTER) += vf_lenscorrection.o
> OBJS-$(CONFIG_REMOVELOGO_FILTER) += bbox.o lswsutils.o lavfutils.o vf_removelogo.o
> OBJS-$(CONFIG_ROTATE_FILTER) += vf_rotate.o
> OBJS-$(CONFIG_SEPARATEFIELDS_FILTER) += vf_separatefields.o
> diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
> index 1877557..b1d6ff5 100644
> --- a/libavfilter/allfilters.c
> +++ b/libavfilter/allfilters.c
> @@ -156,6 +156,7 @@ void avfilter_register_all(void)
> REGISTER_FILTER(INTERLACE, interlace, vf);
> REGISTER_FILTER(INTERLEAVE, interleave, vf);
> REGISTER_FILTER(KERNDEINT, kerndeint, vf);
> + REGISTER_FILTER(LENSCORRECTION, lenscorrection, vf);
> REGISTER_FILTER(LUT3D, lut3d, vf);
> REGISTER_FILTER(LUT, lut, vf);
> REGISTER_FILTER(LUTRGB, lutrgb, vf);
> diff --git a/libavfilter/version.h b/libavfilter/version.h
> index 1a43dc5..47bac78 100644
> --- a/libavfilter/version.h
> +++ b/libavfilter/version.h
> @@ -30,8 +30,8 @@
> #include "libavutil/version.h"
>
> #define LIBAVFILTER_VERSION_MAJOR 4
> -#define LIBAVFILTER_VERSION_MINOR 11
> -#define LIBAVFILTER_VERSION_MICRO 102
> +#define LIBAVFILTER_VERSION_MINOR 12
> +#define LIBAVFILTER_VERSION_MICRO 100
>
> #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \
> LIBAVFILTER_VERSION_MINOR, \
> diff --git a/libavfilter/vf_lenscorrection.c b/libavfilter/vf_lenscorrection.c
> new file mode 100644
> index 0000000..1aad94c
> --- /dev/null
> +++ b/libavfilter/vf_lenscorrection.c
> @@ -0,0 +1,208 @@
> +/*
> + * Copyright (c) 2014 Daniel Oberhoff
> + * Copyright (C) 2007 Richard Spindler (author of frei0r plugin from which this was derived)
Switch these two.
> + *
> + * 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
> + * Lenscorrection filter, algorithm from the frei0r plugin with the same name
> +*/
> +#include <stdlib.h>
> +#include <math.h>
> +
> +#include "libavutil/opt.h"
> +#include "libavutil/intreadwrite.h"
> +#include "libavutil/pixdesc.h"
> +
> +#include "avfilter.h"
> +#include "internal.h"
> +#include "video.h"
> +
> +typedef struct LenscorrectionCtx {
> + const AVClass* av_class;
> + unsigned int width;
> + unsigned int height;
> + int hsub, vsub;
> + int nb_planes;
> + double cx, cy, k1, k2;
> +} LenscorrectionCtx;
In FFmpeg, we indent by four spaces.
> +
> +#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
> +static const AVOption Lenscorrection_options[] = {
> + { "cx", "set relative center x", offsetof(LenscorrectionCtx, cx), AV_OPT_TYPE_DOUBLE, {.dbl=0.5}, 0, 1, .flags=FLAGS },
> + { "cy", "set relative center y", offsetof(LenscorrectionCtx, cy), AV_OPT_TYPE_DOUBLE, {.dbl=0.5}, 0, 1, .flags=FLAGS },
> + { "k1", "set quadratic distortion factor", offsetof(LenscorrectionCtx, k1), AV_OPT_TYPE_DOUBLE, {.dbl=0.5}, 0, 1, .flags=FLAGS },
> + { "k2", "set double quadratic distortion factor", offsetof(LenscorrectionCtx, k2), AV_OPT_TYPE_DOUBLE, {.dbl=0.5}, 0, 1, .flags=FLAGS },
You said 0.5 means no correction. Maybe use a saner default?
> + { NULL }
> +};
We usually use lower-case for names of AVOption arrays and other
structures. So it's lenscorrection_options[].
> +
> +AVFILTER_DEFINE_CLASS(Lenscorrection);
here too
> +
> +static av_cold int init(AVFilterContext *ctx)
> +{
> + return 0;
> +}
> +
> +static av_cold void uninit(AVFilterContext *ctx)
> +{
> +}
As Paul said, drop these two.
> +
> +typedef struct ThreadData {
> + AVFrame *in, *out;
> + float w, h;
> + int plane;
> + float xcenter, ycenter;
> + float k1, k2;
> +} ThreadData;
> +
> +static int filter_slice(AVFilterContext *ctx, void *arg, int job, int nb_jobs)
> +{
> + ThreadData *td = (ThreadData*)arg;
> + AVFrame *in = td->in;
> + AVFrame *out = td->out;
> +
> + const float w = td->w, h = td->h;
> + const float xcenter = td->xcenter;
> + const float ycenter = td->ycenter;
> + const float r2inv = 4.0 / (w * w + h * h);
> + const float k1 = td->k1 - 0.5;
> + const float k2 = td->k2 - 0.5;
> + const int start = (h * job ) / nb_jobs;
> + const int end = (h * (job+1)) / nb_jobs;
> + const int plane = td->plane;
> + const int inlinesize = in->linesize[plane];
> + const int outlinesize = out->linesize[plane];
> + const uint8_t *indata = in->data[plane];
> + uint8_t *outrow = out->data[plane] + start * outlinesize;
> + int i;
> + for (i = start; i < end; i++, outrow += outlinesize) {
> + const float off_y = i - ycenter;
> + const float off_y2 = off_y * off_y;
> + uint8_t *out = outrow;
> + int j;
> + for (j = 0; j < w; j++) {
> + const float off_x = j - xcenter;
> + const float r2 = (off_x * off_x + off_y2) * r2inv;
> + const float radius_mult = 1.0f + r2 * k1 + r2 * r2 * k2;
> + const int x = xcenter + radius_mult * off_x + 0.5f;
> + const int y = ycenter + radius_mult * off_y + 0.5f;
> + const char isvalid = x > 0 && x < w - 1 && y > 0 && y < h - 1;
> + *out++ = isvalid ? indata[y * inlinesize + x] : 0;
> + }
> + }
> + return 0;
> +}
> +
> +static int query_formats(AVFilterContext *ctx)
> +{
> + static enum PixelFormat pix_fmts[] = {
Use AVPixelFormat. PixelFormat is deprecated.
> + AV_PIX_FMT_YUV410P,
> + AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUVJ444P,
> + AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUVJ420P,
> + AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUVA420P,
> + AV_PIX_FMT_YUV422P,
> + AV_PIX_FMT_NONE
> + };
> +
> + ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
> + return 0;
> +}
> +
> +static int config_props(AVFilterLink *outlink)
> +{
> + AVFilterContext* ctx = outlink->src;
> + LenscorrectionCtx* rect = ctx->priv;
We usually write * as attached to the variable name, which reduces
ambiguity (int* a, b; vs. int *a, *b). So it's
AVFilterContext *ctx = outlink->src;
LenscorrectionCtx *rect = ctx->priv;
> + AVFilterLink *inlink = ctx->inputs[0];
> + const AVPixFmtDescriptor *pixdesc = av_pix_fmt_desc_get(inlink->format);
> + rect->hsub = pixdesc->log2_chroma_w;
> + rect->vsub = pixdesc->log2_chroma_h;
> + outlink->w = rect->width = inlink->w;
> + outlink->h = rect->height = inlink->h;
> + rect->nb_planes = av_pix_fmt_count_planes(inlink->format);
> + return 0;
> +}
> +
> +static int filter_frame(AVFilterLink *inlink, AVFrame *in)
> +{
> + AVFilterContext *ctx = inlink->dst;
> + AVFilterLink *outlink = ctx->outputs[0];
> + LenscorrectionCtx *rect = (LenscorrectionCtx*)ctx->priv;
> + AVFrame* out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
ditto
> + int plane;
> +
> + if (!out) {
> + av_frame_free(&in);
> + return AVERROR(ENOMEM);
> + }
> +
> + av_frame_copy_props(out, in);
> +
> + for (plane = 0; plane < rect->nb_planes; ++plane) {
> + int hsub = plane == 1 || plane == 2 ? rect->hsub : 0;
> + int vsub = plane == 1 || plane == 2 ? rect->vsub : 0;
> + float hdiv = 1 << hsub;
> + float vdiv = 1 << vsub;
> + float w = rect->width / hdiv;
> + float h = rect->height / vdiv;
> + ThreadData td = { .in = in, .out = out,
> + .w = w,
> + .h = h,
> + .xcenter = rect->cx * w,
> + .ycenter = rect->cy * h,
> + .k1 = rect->k1,
> + .k2 = rect->k2,
> + .plane = plane};
We usually write designated initializers like this:
ThreadData td = {
.in = in,
.out = out,
.w = w,
.h = h,
.xcenter = rect->cx * w,
.ycenter = rect->cy * h,
.k1 = rect->k1,
.k2 = rect->k2,
.plane = plane
};
which makes it slightly more readable.
> + ctx->internal->execute(ctx, filter_slice, &td, NULL, FFMIN(h, ctx->graph->nb_threads));
> + }
> +
> + av_frame_free(&in);
> + return ff_filter_frame(outlink, out);
> +}
> +
> +static const AVFilterPad Lenscorrection_inputs[] = {
> + {
> + .name = "default",
> + .type = AVMEDIA_TYPE_VIDEO,
> + .filter_frame = filter_frame,
> + },
> + { NULL }
> +};
> +
> +static const AVFilterPad Lenscorrection_outputs[] = {
> + {
> + .name = "default",
> + .type = AVMEDIA_TYPE_VIDEO,
> + .config_props = config_props,
> + },
> + { NULL }
> +};
capitalization of variable names
> +
> +AVFilter ff_vf_lenscorrection = {
> + .name = "lenscorrection",
> + .description = NULL_IF_CONFIG_SMALL("Rectify the image by correcting for lens distortion."),
> + .priv_size = sizeof(LenscorrectionCtx),
> + .init = init,
> + .uninit = uninit,
> + .query_formats = query_formats,
> + .inputs = Lenscorrection_inputs,
> + .outputs = Lenscorrection_outputs,
> + .priv_class = &Lenscorrection_class,
> + .flags = AVFILTER_FLAG_SLICE_THREADS,
> +};
> +
> \ No newline at end of file
> diff --git a/tests/fate/filter-video.mak b/tests/fate/filter-video.mak
> index d87129b..cc5a71b 100644
> --- a/tests/fate/filter-video.mak
> +++ b/tests/fate/filter-video.mak
> @@ -331,6 +331,9 @@ fate-filter-pixfmts-il: CMD = pixfmts "luma_mode=d:chroma_mode=d:alpha_mode=d
> FATE_FILTER_PIXFMTS-$(CONFIG_KERNDEINT_FILTER) += fate-filter-pixfmts-kerndeint
> fate-filter-pixfmts-kerndeint: CMD = pixfmts "" "tinterlace=interleave_top,"
>
> +FATE_FILTER_PIXFMTS-$(CONFIG_LENSCORRECTION_FILTER) += fate-filter-pixfmts-lenscorrection
> +fate-filter-pixfmts-lenscorrection: CMD = pixfmts "0.6:0.4:0.65:0.4"
> +
> FATE_FILTER_PIXFMTS-$(CONFIG_LUT_FILTER) += fate-filter-pixfmts-lut
> fate-filter-pixfmts-lut: CMD = pixfmts "c0=2*val:c1=2*val:c2=val/2:c3=negval+40"
>
> diff --git a/tests/ref/fate/filter-pixfmts-lenscorrection b/tests/ref/fate/filter-pixfmts-lenscorrection
> new file mode 100644
> index 0000000..18af6fb
> --- /dev/null
> +++ b/tests/ref/fate/filter-pixfmts-lenscorrection
> @@ -0,0 +1,8 @@
> +yuv410p e7d59dbdb1afab7e2a8f770d563e28c4
> +yuv420p bc45b4762d5271410ff825317c85af64
> +yuv422p 5cce0c299322634d65e6b32c976e2c12
> +yuv444p 03de9a93ab3045a523b234ea93f21c91
> +yuva420p d1fa6735c4e7fbbf3a501cec1f0b4ac1
> +yuva444p 713ddf5861d3df11c70a242a13c5e92e
> +yuvj420p 1d5cccaf4ef568ae9fa36f9a28e71c34
> +yuvj444p aef1db29848e3b1dcaf4309255c38cbd
>
More information about the ffmpeg-devel
mailing list