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/imgutils.h"
29 #include "libavutil/opt.h"
30 #include "libavutil/pixdesc.h"
33 #define FF_BUFQUEUE_SIZE 129
34 #include "bufferqueue.h"
40 #define SIZE FF_BUFQUEUE_SIZE
42 typedef struct ATADenoiseContext {
45 float fthra[4], fthrb[4];
55 int linesize[4][SIZE];
59 int (*filter_slice)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
62 #define OFFSET(x) offsetof(ATADenoiseContext, x)
63 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
65 static const AVOption atadenoise_options[] = {
66 { "0a", "set threshold A for 1st plane", OFFSET(fthra[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.02}, 0, 0.3, FLAGS },
67 { "0b", "set threshold B for 1st plane", OFFSET(fthrb[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.04}, 0, 5.0, FLAGS },
68 { "1a", "set threshold A for 2nd plane", OFFSET(fthra[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.02}, 0, 0.3, FLAGS },
69 { "1b", "set threshold B for 2nd plane", OFFSET(fthrb[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.04}, 0, 5.0, FLAGS },
70 { "2a", "set threshold A for 3rd plane", OFFSET(fthra[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.02}, 0, 0.3, FLAGS },
71 { "2b", "set threshold B for 3rd plane", OFFSET(fthrb[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.04}, 0, 5.0, FLAGS },
72 { "s", "set how many frames to use", OFFSET(size), AV_OPT_TYPE_INT, {.i64=9}, 5, SIZE, FLAGS },
73 { "p", "set what planes to filter", OFFSET(planes), AV_OPT_TYPE_FLAGS, {.i64=7}, 0, 15, FLAGS },
77 AVFILTER_DEFINE_CLASS(atadenoise);
79 static int query_formats(AVFilterContext *ctx)
81 static const enum AVPixelFormat pixel_fmts[] = {
88 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P,
89 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P,
90 AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV444P,
91 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
92 AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ444P,
94 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
95 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
97 AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV420P12,
99 AV_PIX_FMT_YUV444P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV420P14,
100 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
101 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
102 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
105 AVFilterFormats *formats = ff_make_format_list(pixel_fmts);
107 return AVERROR(ENOMEM);
108 return ff_set_common_formats(ctx, formats);
111 static av_cold int init(AVFilterContext *ctx)
113 ATADenoiseContext *s = ctx->priv;
115 if (!(s->size & 1)) {
116 av_log(ctx, AV_LOG_WARNING, "size %d is invalid. Must be an odd value, setting it to %d.\n", s->size, s->size|1);
119 s->mid = s->size / 2 + 1;
124 typedef struct ThreadData {
128 static int filter_slice8(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
130 ATADenoiseContext *s = ctx->priv;
131 ThreadData *td = arg;
132 AVFrame *in = td->in;
133 AVFrame *out = td->out;
134 const int size = s->size;
135 const int mid = s->mid;
138 for (p = 0; p < s->nb_planes; p++) {
139 const int h = s->planeheight[p];
140 const int w = s->planewidth[p];
141 const int slice_start = (h * jobnr) / nb_jobs;
142 const int slice_end = (h * (jobnr+1)) / nb_jobs;
143 const uint8_t *src = in->data[p] + slice_start * in->linesize[p];
144 uint8_t *dst = out->data[p] + slice_start * out->linesize[p];
145 const int thra = s->thra[p];
146 const int thrb = s->thrb[p];
147 const uint8_t **data = (const uint8_t **)s->data[p];
148 const int *linesize = (const int *)s->linesize[p];
149 const uint8_t *srcf[SIZE];
151 if (!((1 << p) & s->planes)) {
152 av_image_copy_plane(dst, out->linesize[p], src, in->linesize[p],
153 w, slice_end - slice_start);
157 for (i = 0; i < size; i++)
158 srcf[i] = data[i] + slice_start * linesize[i];
160 for (y = slice_start; y < slice_end; y++) {
161 for (x = 0; x < w; x++) {
162 const int srcx = src[x];
163 unsigned lsumdiff = 0, rsumdiff = 0;
164 unsigned ldiff, rdiff;
169 for (j = mid - 1, i = mid + 1; j >= 0 && i < size; j--, i++) {
172 ldiff = FFABS(srcx - srcjx);
182 rdiff = FFABS(srcx - srcix);
191 dst[x] = sum / (r + l + 1);
194 dst += out->linesize[p];
195 src += in->linesize[p];
197 for (i = 0; i < size; i++)
198 srcf[i] += linesize[i];
205 static int filter_slice16(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
207 ATADenoiseContext *s = ctx->priv;
208 ThreadData *td = arg;
209 AVFrame *in = td->in;
210 AVFrame *out = td->out;
211 const int size = s->size;
212 const int mid = s->mid;
215 for (p = 0; p < s->nb_planes; p++) {
216 const int h = s->planeheight[p];
217 const int w = s->planewidth[p];
218 const int slice_start = (h * jobnr) / nb_jobs;
219 const int slice_end = (h * (jobnr+1)) / nb_jobs;
220 const uint16_t *src = (uint16_t *)(in->data[p] + slice_start * in->linesize[p]);
221 uint16_t *dst = (uint16_t *)(out->data[p] + slice_start * out->linesize[p]);
222 const int thra = s->thra[p];
223 const int thrb = s->thrb[p];
224 const uint8_t **data = (const uint8_t **)s->data[p];
225 const int *linesize = (const int *)s->linesize[p];
226 const uint16_t *srcf[SIZE];
228 if (!((1 << p) & s->planes)) {
229 av_image_copy_plane((uint8_t *)dst, out->linesize[p], (uint8_t *)src, in->linesize[p],
230 w * 2, slice_end - slice_start);
234 for (i = 0; i < s->size; i++)
235 srcf[i] = (const uint16_t *)(data[i] + slice_start * linesize[i]);
237 for (y = slice_start; y < slice_end; y++) {
238 for (x = 0; x < w; x++) {
239 const int srcx = src[x];
240 unsigned lsumdiff = 0, rsumdiff = 0;
241 unsigned ldiff, rdiff;
246 for (j = mid - 1, i = mid + 1; j >= 0 && i < size; j--, i++) {
249 ldiff = FFABS(srcx - srcjx);
259 rdiff = FFABS(srcx - srcix);
268 dst[x] = sum / (r + l + 1);
271 dst += out->linesize[p] / 2;
272 src += in->linesize[p] / 2;
274 for (i = 0; i < size; i++)
275 srcf[i] += linesize[i] / 2;
282 static int config_input(AVFilterLink *inlink)
284 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
285 AVFilterContext *ctx = inlink->dst;
286 ATADenoiseContext *s = ctx->priv;
289 s->nb_planes = desc->nb_components;
291 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
292 s->planeheight[0] = s->planeheight[3] = inlink->h;
293 s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
294 s->planewidth[0] = s->planewidth[3] = inlink->w;
296 depth = desc->comp[0].depth;
298 s->filter_slice = filter_slice8;
300 s->filter_slice = filter_slice16;
302 s->thra[0] = s->fthra[0] * (1 << depth) - 1;
303 s->thra[1] = s->fthra[1] * (1 << depth) - 1;
304 s->thra[2] = s->fthra[2] * (1 << depth) - 1;
305 s->thrb[0] = s->fthrb[0] * (1 << depth) - 1;
306 s->thrb[1] = s->fthrb[1] * (1 << depth) - 1;
307 s->thrb[2] = s->fthrb[2] * (1 << depth) - 1;
312 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
314 AVFilterContext *ctx = inlink->dst;
315 AVFilterLink *outlink = ctx->outputs[0];
316 ATADenoiseContext *s = ctx->priv;
320 if (s->q.available != s->size) {
321 if (s->q.available < s->mid) {
322 for (i = 0; i < s->mid; i++) {
323 out = av_frame_clone(buf);
326 return AVERROR(ENOMEM);
328 ff_bufqueue_add(ctx, &s->q, out);
331 if (s->q.available < s->size) {
332 ff_bufqueue_add(ctx, &s->q, buf);
338 in = ff_bufqueue_peek(&s->q, s->mid);
340 if (!ctx->is_disabled) {
343 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
346 return AVERROR(ENOMEM);
349 for (i = 0; i < s->size; i++) {
350 AVFrame *frame = ff_bufqueue_peek(&s->q, i);
352 s->data[0][i] = frame->data[0];
353 s->data[1][i] = frame->data[1];
354 s->data[2][i] = frame->data[2];
355 s->linesize[0][i] = frame->linesize[0];
356 s->linesize[1][i] = frame->linesize[1];
357 s->linesize[2][i] = frame->linesize[2];
360 td.in = in; td.out = out;
361 ctx->internal->execute(ctx, s->filter_slice, &td, NULL,
362 FFMIN3(s->planeheight[1],
364 ff_filter_get_nb_threads(ctx)));
365 av_frame_copy_props(out, in);
367 out = av_frame_clone(in);
370 return AVERROR(ENOMEM);
374 in = ff_bufqueue_get(&s->q);
376 ff_bufqueue_add(ctx, &s->q, buf);
378 return ff_filter_frame(outlink, out);
381 static int request_frame(AVFilterLink *outlink)
383 AVFilterContext *ctx = outlink->src;
384 ATADenoiseContext *s = ctx->priv;
387 ret = ff_request_frame(ctx->inputs[0]);
389 if (ret == AVERROR_EOF && !ctx->is_disabled && s->available) {
390 AVFrame *buf = av_frame_clone(ff_bufqueue_peek(&s->q, s->available));
392 return AVERROR(ENOMEM);
394 ret = filter_frame(ctx->inputs[0], buf);
401 static av_cold void uninit(AVFilterContext *ctx)
403 ATADenoiseContext *s = ctx->priv;
405 ff_bufqueue_discard_all(&s->q);
408 static const AVFilterPad inputs[] = {
411 .type = AVMEDIA_TYPE_VIDEO,
412 .filter_frame = filter_frame,
413 .config_props = config_input,
418 static const AVFilterPad outputs[] = {
421 .type = AVMEDIA_TYPE_VIDEO,
422 .request_frame = request_frame,
427 AVFilter ff_vf_atadenoise = {
428 .name = "atadenoise",
429 .description = NULL_IF_CONFIG_SMALL("Apply an Adaptive Temporal Averaging Denoiser."),
430 .priv_size = sizeof(ATADenoiseContext),
431 .priv_class = &atadenoise_class,
434 .query_formats = query_formats,
437 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,