2 * Copyright (c) 2015 Paul B Mahol
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * Adaptive Temporal Averaging Denoiser,
24 * based on paper "Video Denoising Based on Adaptive Temporal Averaging" by
25 * David Bartovčak and Miroslav Vrankić
28 #include "libavutil/opt.h"
29 #include "libavutil/pixdesc.h"
32 #define FF_BUFQUEUE_SIZE 129
33 #include "bufferqueue.h"
39 #define SIZE FF_BUFQUEUE_SIZE
41 typedef struct ATADenoiseContext {
44 float fthra[4], fthrb[4];
53 int linesize[4][SIZE];
57 int (*filter_slice)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
60 #define OFFSET(x) offsetof(ATADenoiseContext, x)
61 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
63 static const AVOption atadenoise_options[] = {
64 { "0a", "set threshold A for 1st plane", OFFSET(fthra[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.02}, 0, 0.3, FLAGS },
65 { "0b", "set threshold B for 1st plane", OFFSET(fthrb[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.04}, 0, 5.0, FLAGS },
66 { "1a", "set threshold A for 2nd plane", OFFSET(fthra[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.02}, 0, 0.3, FLAGS },
67 { "1b", "set threshold B for 2nd plane", OFFSET(fthrb[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.04}, 0, 5.0, FLAGS },
68 { "2a", "set threshold A for 3rd plane", OFFSET(fthra[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.02}, 0, 0.3, FLAGS },
69 { "2b", "set threshold B for 3rd plane", OFFSET(fthrb[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.04}, 0, 5.0, FLAGS },
70 { "s", "set how many frames to use", OFFSET(size), AV_OPT_TYPE_INT, {.i64=33}, 5, SIZE, FLAGS },
74 AVFILTER_DEFINE_CLASS(atadenoise);
76 static int query_formats(AVFilterContext *ctx)
78 static const enum AVPixelFormat pixel_fmts[] = {
80 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P,
81 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P,
82 AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV444P,
83 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
84 AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ444P,
86 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
87 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
88 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
89 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
90 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
93 AVFilterFormats *formats = ff_make_format_list(pixel_fmts);
95 return AVERROR(ENOMEM);
96 return ff_set_common_formats(ctx, formats);
99 static av_cold int init(AVFilterContext *ctx)
101 ATADenoiseContext *s = ctx->priv;
103 if (!(s->size & 1)) {
104 av_log(ctx, AV_LOG_ERROR, "size %d is invalid. Must be an odd value.\n", s->size);
105 return AVERROR(EINVAL);
107 s->mid = s->size / 2 + 1;
112 typedef struct ThreadData {
116 static int filter_slice8(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
118 ATADenoiseContext *s = ctx->priv;
119 ThreadData *td = arg;
120 AVFrame *in = td->in;
121 AVFrame *out = td->out;
122 const int size = s->size;
123 const int mid = s->mid;
126 for (p = 0; p < s->nb_planes; p++) {
127 const int h = s->planeheight[p];
128 const int w = s->planewidth[p];
129 const int slice_start = (h * jobnr) / nb_jobs;
130 const int slice_end = (h * (jobnr+1)) / nb_jobs;
131 const uint8_t *src = in->data[p] + slice_start * in->linesize[p];
132 uint8_t *dst = out->data[p] + slice_start * out->linesize[p];
133 const int thra = s->thra[p];
134 const int thrb = s->thrb[p];
135 const uint8_t **data = (const uint8_t **)s->data[p];
136 const int *linesize = (const int *)s->linesize[p];
137 const uint8_t *srcf[SIZE];
139 for (i = 0; i < size; i++)
140 srcf[i] = data[i] + slice_start * linesize[i];
142 for (y = slice_start; y < slice_end; y++) {
143 for (x = 0; x < w; x++) {
144 const int srcx = src[x];
145 unsigned lsumdiff = 0, rsumdiff = 0;
146 unsigned ldiff, rdiff;
151 for (j = mid - 1, i = mid + 1; j >= 0 && i < size; j--, i++) {
154 ldiff = FFABS(srcx - srcjx);
164 rdiff = FFABS(srcx - srcix);
173 dst[x] = sum / (r + l + 1);
176 dst += out->linesize[p];
177 src += in->linesize[p];
179 for (i = 0; i < size; i++)
180 srcf[i] += linesize[i];
187 static int filter_slice16(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
189 ATADenoiseContext *s = ctx->priv;
190 ThreadData *td = arg;
191 AVFrame *in = td->in;
192 AVFrame *out = td->out;
193 const int size = s->size;
194 const int mid = s->mid;
197 for (p = 0; p < s->nb_planes; p++) {
198 const int h = s->planeheight[p];
199 const int w = s->planewidth[p];
200 const int slice_start = (h * jobnr) / nb_jobs;
201 const int slice_end = (h * (jobnr+1)) / nb_jobs;
202 const uint16_t *src = (uint16_t *)(in->data[p] + slice_start * in->linesize[p]);
203 uint16_t *dst = (uint16_t *)(out->data[p] + slice_start * out->linesize[p]);
204 const int thra = s->thra[p];
205 const int thrb = s->thrb[p];
206 const uint8_t **data = (const uint8_t **)s->data[p];
207 const int *linesize = (const int *)s->linesize[p];
208 const uint16_t *srcf[SIZE];
210 for (i = 0; i < s->size; i++)
211 srcf[i] = (const uint16_t *)(data[i] + slice_start * linesize[i]);
213 for (y = slice_start; y < slice_end; y++) {
214 for (x = 0; x < w; x++) {
215 const int srcx = src[x];
216 unsigned lsumdiff = 0, rsumdiff = 0;
217 unsigned ldiff, rdiff;
222 for (j = mid - 1, i = mid + 1; j >= 0 && i < size; j--, i++) {
225 ldiff = FFABS(srcx - srcjx);
235 rdiff = FFABS(srcx - srcix);
244 dst[x] = sum / (r + l + 1);
247 dst += out->linesize[p] / 2;
248 src += in->linesize[p] / 2;
250 for (i = 0; i < size; i++)
251 srcf[i] += linesize[i] / 2;
258 static int config_input(AVFilterLink *inlink)
260 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
261 AVFilterContext *ctx = inlink->dst;
262 ATADenoiseContext *s = ctx->priv;
265 s->nb_planes = desc->nb_components;
267 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
268 s->planeheight[0] = s->planeheight[3] = inlink->h;
269 s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
270 s->planewidth[0] = s->planewidth[3] = inlink->w;
272 depth = desc->comp[0].depth;
274 s->filter_slice = filter_slice8;
276 s->filter_slice = filter_slice16;
278 s->thra[0] = s->fthra[0] * (1 << depth) - 1;
279 s->thra[1] = s->fthra[1] * (1 << depth) - 1;
280 s->thra[2] = s->fthra[2] * (1 << depth) - 1;
281 s->thrb[0] = s->fthrb[0] * (1 << depth) - 1;
282 s->thrb[1] = s->fthrb[1] * (1 << depth) - 1;
283 s->thrb[2] = s->fthrb[2] * (1 << depth) - 1;
288 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
290 AVFilterContext *ctx = inlink->dst;
291 AVFilterLink *outlink = ctx->outputs[0];
292 ATADenoiseContext *s = ctx->priv;
296 if (s->q.available != s->size) {
297 if (s->q.available < s->mid) {
298 for (i = 0; i < s->mid; i++) {
299 out = av_frame_clone(buf);
302 return AVERROR(ENOMEM);
304 ff_bufqueue_add(ctx, &s->q, out);
307 if (s->q.available < s->size) {
308 ff_bufqueue_add(ctx, &s->q, buf);
314 in = ff_bufqueue_peek(&s->q, s->mid);
316 if (!ctx->is_disabled) {
319 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
322 return AVERROR(ENOMEM);
325 for (i = 0; i < s->size; i++) {
326 AVFrame *frame = ff_bufqueue_peek(&s->q, i);
328 s->data[0][i] = frame->data[0];
329 s->data[1][i] = frame->data[1];
330 s->data[2][i] = frame->data[2];
331 s->linesize[0][i] = frame->linesize[0];
332 s->linesize[1][i] = frame->linesize[1];
333 s->linesize[2][i] = frame->linesize[2];
336 td.in = in; td.out = out;
337 ctx->internal->execute(ctx, s->filter_slice, &td, NULL,
338 FFMIN3(s->planeheight[1],
340 ctx->graph->nb_threads));
341 av_frame_copy_props(out, in);
343 out = av_frame_clone(in);
346 return AVERROR(ENOMEM);
350 in = ff_bufqueue_get(&s->q);
352 ff_bufqueue_add(ctx, &s->q, buf);
354 return ff_filter_frame(outlink, out);
357 static int request_frame(AVFilterLink *outlink)
359 AVFilterContext *ctx = outlink->src;
360 ATADenoiseContext *s = ctx->priv;
363 ret = ff_request_frame(ctx->inputs[0]);
365 if (ret == AVERROR_EOF && !ctx->is_disabled && s->available) {
366 AVFrame *buf = av_frame_clone(ff_bufqueue_peek(&s->q, s->available));
368 return AVERROR(ENOMEM);
370 ret = filter_frame(ctx->inputs[0], buf);
377 static av_cold void uninit(AVFilterContext *ctx)
379 ATADenoiseContext *s = ctx->priv;
381 ff_bufqueue_discard_all(&s->q);
384 static const AVFilterPad inputs[] = {
387 .type = AVMEDIA_TYPE_VIDEO,
388 .filter_frame = filter_frame,
389 .config_props = config_input,
394 static const AVFilterPad outputs[] = {
397 .type = AVMEDIA_TYPE_VIDEO,
398 .request_frame = request_frame,
403 AVFilter ff_vf_atadenoise = {
404 .name = "atadenoise",
405 .description = NULL_IF_CONFIG_SMALL("Apply an Adaptive Temporal Averaging Denoiser."),
406 .priv_size = sizeof(ATADenoiseContext),
407 .priv_class = &atadenoise_class,
410 .query_formats = query_formats,
413 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,