diff --git a/Changelog b/Changelog index 6fab938e0f..7c971fce63 100644 --- a/Changelog +++ b/Changelog @@ -19,6 +19,7 @@ version : - bilateral filter - maskedmin and maskedmax filters - VDPAU VP9 hwaccel +- median filter version 4.2: diff --git a/doc/filters.texi b/doc/filters.texi index 5b8062d806..c4477e6677 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -12805,6 +12805,21 @@ Higher values should result in a smoother motion vector field but less optimal individual vectors. Default value is 1. @end table +@section median + +Pick median pixel from certain rectangle defined by radius. + +This filter accepts the following options: + +@table @option +@item radius +Set horizontal and vertical radius size. Default value is @code{1}. +Allowed range is integer from 1 to 127. + +@item planes +Set which planes to process. Default is @code{15}, which is all available planes. +@end table + @section mergeplanes Merge color channel components from several video streams. diff --git a/libavfilter/Makefile b/libavfilter/Makefile index bae566d433..2080eed559 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -297,6 +297,7 @@ OBJS-$(CONFIG_MASKEDMERGE_FILTER) += vf_maskedmerge.o framesync.o OBJS-$(CONFIG_MASKEDMIN_FILTER) += vf_maskedminmax.o framesync.o OBJS-$(CONFIG_MASKFUN_FILTER) += vf_maskfun.o OBJS-$(CONFIG_MCDEINT_FILTER) += vf_mcdeint.o +OBJS-$(CONFIG_MEDIAN_FILTER) += vf_median.o OBJS-$(CONFIG_MERGEPLANES_FILTER) += vf_mergeplanes.o framesync.o OBJS-$(CONFIG_MESTIMATE_FILTER) += vf_mestimate.o motion_estimation.o OBJS-$(CONFIG_METADATA_FILTER) += f_metadata.o diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index d136734338..cb609067b6 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -282,6 +282,7 @@ extern AVFilter ff_vf_maskedmerge; extern AVFilter ff_vf_maskedmin; extern AVFilter ff_vf_maskfun; extern AVFilter ff_vf_mcdeint; +extern AVFilter ff_vf_median; extern AVFilter ff_vf_mergeplanes; extern AVFilter ff_vf_mestimate; extern AVFilter ff_vf_metadata; diff --git a/libavfilter/median.h b/libavfilter/median.h new file mode 100644 index 0000000000..bb07fddd5d --- /dev/null +++ b/libavfilter/median.h @@ -0,0 +1,54 @@ +/* + * Copyright (c) 2019 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 + * Redistribution and use in source and binary forms, with or without modification, + * are permitted provided that the following conditions are met: + */ + +#ifndef AVFILTER_MEDIAN_H +#define AVFILTER_MEDIAN_H + +#include "avfilter.h" + +typedef struct MedianContext { + const AVClass *class; + + int planes; + int radius; + + int planewidth[4]; + int planeheight[4]; + int depth; + int nb_planes; + int nb_threads; + + uint16_t **coarse, **fine; + int coarse_size, fine_size; + int bins; + int t; + + void (*hadd)(uint16_t *dst, const uint16_t *src, int bins); + void (*hsub)(uint16_t *dst, const uint16_t *src, int bins); + void (*hmuladd)(uint16_t *dst, const uint16_t *src, int f, int bins); + + void (*filter_plane)(AVFilterContext *ctx, const uint8_t *ssrc, int src_linesize, + uint8_t *ddst, int dst_linesize, int width, int height, + int slice_h_start, int slice_h_end, int jobnr); +} MedianContext; + +#endif diff --git a/libavfilter/median_template.c b/libavfilter/median_template.c new file mode 100644 index 0000000000..6953a23a5b --- /dev/null +++ b/libavfilter/median_template.c @@ -0,0 +1,159 @@ +/* + * Copyright (c) 2019 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 + * Redistribution and use in source and binary forms, with or without modification, + * are permitted provided that the following conditions are met: + */ + +#include "libavutil/avassert.h" +#include "avfilter.h" +#include "formats.h" +#include "internal.h" +#include "video.h" + +#undef pixel +#if DEPTH == 8 +#define pixel uint8_t +#else +#define pixel uint16_t +#endif + +#undef htype +#define htype uint16_t + +#undef fn +#undef fn2 +#undef fn3 +#define SHIFT ((DEPTH + 1) / 2) +#define BINS (1 << SHIFT) +#define MASK (BINS - 1) +#define fn3(a,b) a##_##b +#define fn2(a,b) fn3(a,b) +#define fn(a) fn2(a, DEPTH) + +#define PICK_COARSE_BIN(x, y) (BINS * (x) + ((y) >> SHIFT)) +#define PICK_FINE_BIN(x, y, z) (BINS * ((x) * ((y) >> SHIFT) + (z)) + ((y) & MASK)) + +static void fn(filter_plane)(AVFilterContext *ctx, const uint8_t *ssrc, int src_linesize, + uint8_t *ddst, int dst_linesize, int width, int height, + int slice_h_start, int slice_h_end, int jobnr) +{ + MedianContext *s = ctx->priv; + htype *ccoarse = s->coarse[jobnr]; + htype *cfine = s->fine[jobnr]; + const int radius = s->radius; + const int t = s->t; + const pixel *src = (const pixel *)ssrc; + pixel *dst = (pixel *)ddst; + const pixel *srcp; + const pixel *p; + + src_linesize /= sizeof(pixel); + dst_linesize /= sizeof(pixel); + + memset(cfine, 0, s->fine_size * sizeof(*cfine)); + memset(ccoarse, 0, s->coarse_size * sizeof(*ccoarse)); + + srcp = src + FFMAX(0, slice_h_start - radius) * src_linesize; + if (jobnr == 0) { + for (int i = 0; i < width; i++) { + cfine[PICK_FINE_BIN(width, srcp[i], i)] += radius + 1; + ccoarse[PICK_COARSE_BIN(i, srcp[i])] += radius + 1; + } + } + + srcp = src + FFMAX(0, slice_h_start - radius - (jobnr != 0)) * src_linesize; + for (int i = 0; i < radius + (jobnr != 0) * (1 + radius); i++) { + for (int j = 0; j < width; j++) { + cfine[PICK_FINE_BIN(width, srcp[j], j)]++; + ccoarse[PICK_COARSE_BIN(j, srcp[j])]++; + } + srcp += src_linesize; + } + + srcp = src; + + for (int i = slice_h_start; i < slice_h_end; i++) { + htype coarse[BINS] = { 0 }; + htype fine[BINS][BINS] = { { 0 } }; + htype luc[BINS] = { 0 }; + + p = srcp + src_linesize * FFMAX(0, i - radius - 1); + for (int j = 0; j < width; j++) { + cfine[PICK_FINE_BIN(width, p[j], j)]--; + ccoarse[PICK_COARSE_BIN(j, p[j])]--; + } + + p = srcp + src_linesize * FFMIN(height - 1, i + radius); + for (int j = 0; j < width; j++) { + cfine[PICK_FINE_BIN(width, p[j], j)]++; + ccoarse[PICK_COARSE_BIN(j, p[j])]++; + } + + s->hmuladd(coarse, &ccoarse[0], radius, BINS); + for (int j = 0; j < radius; j++) + s->hadd(coarse, &ccoarse[BINS * j], BINS); + for (int k = 0; k < BINS; k++) + s->hmuladd(&fine[k][0], &cfine[BINS * width * k], 2 * radius + 1, BINS); + + for (int j = 0; j < width; j++) { + int sum = 0, k, b; + htype *segment; + + s->hadd(coarse, &ccoarse[BINS * FFMIN(j + radius, width - 1)], BINS); + + for (k = 0; k < BINS; k++) { + sum += coarse[k]; + if (sum > t) { + sum -= coarse[k]; + break; + } + } + av_assert0(k < BINS); + + if (luc[k] <= j - radius) { + memset(&fine[k], 0, BINS * sizeof(htype)); + for (luc[k] = j - radius; luc[k] < FFMIN(j + radius + 1, width); luc[k]++) + s->hadd(fine[k], &cfine[BINS * (width * k + luc[k])], BINS); + if (luc[k] < j + radius + 1) { + s->hmuladd(&fine[k][0], &cfine[BINS * (width * k + width - 1)], j + radius + 1 - width, BINS); + luc[k] = j + radius + 1; + } + } else { + for (; luc[k] < j + radius + 1; luc[k]++) { + s->hsub(fine[k], &cfine[BINS * (width * k + FFMAX(luc[k] - 2 * radius - 1, 0))], BINS); + s->hadd(fine[k], &cfine[BINS * (width * k + FFMIN(luc[k], width - 1))], BINS); + } + } + + s->hsub(coarse, &ccoarse[BINS * FFMAX(j - radius, 0)], BINS); + + segment = fine[k]; + for (b = 0; b < BINS; b++) { + sum += segment[b]; + if (sum > t) { + dst[j] = BINS * k + b; + break; + } + } + av_assert0(b < BINS); + } + + dst += dst_linesize; + } +} diff --git a/libavfilter/version.h b/libavfilter/version.h index 58d57ae95f..2e3ff53b20 100644 --- a/libavfilter/version.h +++ b/libavfilter/version.h @@ -30,7 +30,7 @@ #include "libavutil/version.h" #define LIBAVFILTER_VERSION_MAJOR 7 -#define LIBAVFILTER_VERSION_MINOR 65 +#define LIBAVFILTER_VERSION_MINOR 66 #define LIBAVFILTER_VERSION_MICRO 100 diff --git a/libavfilter/vf_median.c b/libavfilter/vf_median.c new file mode 100644 index 0000000000..98296bcc2b --- /dev/null +++ b/libavfilter/vf_median.c @@ -0,0 +1,271 @@ +/* + * Copyright (c) 2019 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 + * Redistribution and use in source and binary forms, with or without modification, + * are permitted provided that the following conditions are met: + */ + +#include "libavutil/avassert.h" +#include "libavutil/imgutils.h" +#include "libavutil/opt.h" +#include "libavutil/pixdesc.h" +#include "avfilter.h" +#include "formats.h" +#include "internal.h" +#include "median.h" +#include "video.h" + +#define DEPTH 8 +#include "median_template.c" + +#undef DEPTH +#define DEPTH 9 +#include "median_template.c" + +#undef DEPTH +#define DEPTH 10 +#include "median_template.c" + +#undef DEPTH +#define DEPTH 12 +#include "median_template.c" + +#undef DEPTH +#define DEPTH 14 +#include "median_template.c" + +#undef DEPTH +#define DEPTH 16 +#include "median_template.c" + +#define OFFSET(x) offsetof(MedianContext, x) +#define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM + +static const AVOption median_options[] = { + { "radius", "set median radius", OFFSET(radius), AV_OPT_TYPE_INT, {.i64=1}, 1, 127, FLAGS }, + { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=0xF}, 0, 0xF, FLAGS }, + { NULL } +}; + +AVFILTER_DEFINE_CLASS(median); + +static void hadd(htype *dst, const htype *src, int bins) +{ + for (int i = 0; i < bins; i++) + dst[i] += src[i]; +} + +static void hsub(htype *dst, const htype *src, int bins) +{ + for (int i = 0; i < bins; i++) + dst[i] -= src[i]; +} + +static void hmuladd(htype *dst, const htype *src, int f, int bins) +{ + for (int i = 0; i < bins; i++) + dst[i] += f * src[i]; +} + +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_GBRP, AV_PIX_FMT_GBRAP, AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, + AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9, AV_PIX_FMT_GBRP9, + AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA444P9, + AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10, + AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12, + AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14, + AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16, + 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_GBRP10, AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16, + AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16, + AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16, + AV_PIX_FMT_NONE + }; + + return ff_set_common_formats(ctx, ff_make_format_list(pix_fmts)); +} + +static int config_input(AVFilterLink *inlink) +{ + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); + MedianContext *s = inlink->dst->priv; + + s->depth = desc->comp[0].depth; + s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w); + s->planewidth[0] = s->planewidth[3] = inlink->w; + s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h); + s->planeheight[0] = s->planeheight[3] = inlink->h; + + s->nb_planes = av_pix_fmt_count_planes(inlink->format); + s->t = 2 * s->radius * s->radius + 2 * s->radius; + + for (int i = 0; i < s->nb_planes; i++) { + if (!(s->planes & (1 << i))) + continue; + + if (s->planewidth[i] < s->radius * 2 + 1) { + av_log(inlink->dst, AV_LOG_ERROR, "The %d plane width %d must be not less than %d\n", i, s->planewidth[i], s->radius * 2 + 1); + return AVERROR(EINVAL); + } + + if (s->planeheight[i] < s->radius * 2 + 1) { + av_log(inlink->dst, AV_LOG_ERROR, "The %d plane height %d must be not less than %d\n", i, s->planeheight[i], s->radius * 2 + 1); + return AVERROR(EINVAL); + } + } + + s->nb_threads = FFMAX(1, FFMIN(s->planeheight[1] / (s->radius + 1), ff_filter_get_nb_threads(inlink->dst))); + s->bins = 1 << ((s->depth + 1) / 2); + s->fine_size = s->bins * s->bins * inlink->w; + s->coarse_size = s->bins * inlink->w; + s->coarse = av_calloc(s->nb_threads, sizeof(*s->coarse)); + s->fine = av_calloc(s->nb_threads, sizeof(*s->fine)); + if (!s->coarse || !s->fine) + return AVERROR(ENOMEM); + for (int i = 0; i < s->nb_threads; i++) { + s->coarse[i] = av_malloc_array(s->coarse_size, sizeof(**s->coarse)); + s->fine[i] = av_malloc_array(s->fine_size, sizeof(**s->fine)); + if (!s->coarse[i] || !s->fine[i]) + return AVERROR(ENOMEM); + } + + s->hadd = hadd; + s->hsub = hsub; + s->hmuladd = hmuladd; + + switch (s->depth) { + case 8: s->filter_plane = filter_plane_8; break; + case 9: s->filter_plane = filter_plane_9; break; + case 10: s->filter_plane = filter_plane_10; break; + case 12: s->filter_plane = filter_plane_12; break; + case 14: s->filter_plane = filter_plane_14; break; + case 16: s->filter_plane = filter_plane_16; break; + } + + return 0; +} + +typedef struct ThreadData { + AVFrame *in, *out; +} ThreadData; + +static int filter_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) +{ + MedianContext *s = ctx->priv; + ThreadData *td = arg; + AVFrame *in = td->in; + AVFrame *out = td->out; + + for (int plane = 0; plane < s->nb_planes; plane++) { + const int h = s->planeheight[plane]; + const int w = s->planewidth[plane]; + const int slice_h_start = (h * jobnr) / nb_jobs; + const int slice_h_end = (h * (jobnr+1)) / nb_jobs; + + if (!(s->planes & (1 << plane))) { + av_image_copy_plane(out->data[plane] + slice_h_start * out->linesize[plane], + out->linesize[plane], + in->data[plane] + slice_h_start * in->linesize[plane], + in->linesize[plane], + w * ((s->depth + 7) / 8), + slice_h_end - slice_h_start); + continue; + } + + s->filter_plane(ctx, in->data[plane], + in->linesize[plane], + out->data[plane] + slice_h_start * out->linesize[plane], + out->linesize[plane], w, h, + slice_h_start, slice_h_end, jobnr); + } + + return 0; +} + +static int filter_frame(AVFilterLink *inlink, AVFrame *in) +{ + AVFilterContext *ctx = inlink->dst; + MedianContext *s = ctx->priv; + AVFilterLink *outlink = ctx->outputs[0]; + ThreadData td; + AVFrame *out; + + 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); + + td.in = in; td.out = out; + ctx->internal->execute(ctx, filter_slice, &td, NULL, s->nb_threads); + + av_frame_free(&in); + return ff_filter_frame(outlink, out); +} + +static av_cold void uninit(AVFilterContext *ctx) +{ + MedianContext *s = ctx->priv; + + for (int i = 0; i < s->nb_threads && s->coarse && s->fine; i++) { + av_freep(&s->coarse[i]); + av_freep(&s->fine[i]); + } + + av_freep(&s->coarse); + av_freep(&s->fine); +} + +static const AVFilterPad median_inputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .config_props = config_input, + .filter_frame = filter_frame, + }, + { NULL } +}; + +static const AVFilterPad median_outputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + }, + { NULL } +}; + +AVFilter ff_vf_median = { + .name = "median", + .description = NULL_IF_CONFIG_SMALL("Apply Median filter."), + .priv_size = sizeof(MedianContext), + .priv_class = &median_class, + .uninit = uninit, + .query_formats = query_formats, + .inputs = median_inputs, + .outputs = median_outputs, + .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS, +};