FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libavfilter/vf_owdenoise.c
Date: 2024-11-20 23:03:26
Exec Total Coverage
Lines: 110 139 79.1%
Functions: 10 10 100.0%
Branches: 57 82 69.5%
Line Branch Exec Source
1 /*
2 * Copyright (c) 2007 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (c) 2013 Clément Bœsch <u pkh me>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License along
18 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
19 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
20 */
21
22 // The code written by Michael Niedermayer in 70024b6b47b9eacfe01e8f92349ca9bf1ccd7d5a:libavfilter/vf_owdenoise.c
23 // can also be used under the LGPL due to:
24 // <michaelni> durandal_1707, if you do all the "todo" points from vf_owdenoise.c that are in that file since 2013 then sure i would be more than happy to relicense my part of it to LGPL
25 // <durandal_1707> michaelni: first relicense than work
26
27 /**
28 * @todo try to change to int
29 * @todo try lifting based implementation
30 * @todo optimize optimize optimize
31 * @todo hard thresholding
32 * @todo use QP to decide filter strength
33 * @todo wavelet normalization / least squares optimal signal vs. noise thresholds
34 */
35
36 #include "libavutil/imgutils.h"
37 #include "libavutil/mem.h"
38 #include "libavutil/opt.h"
39 #include "libavutil/mem_internal.h"
40 #include "libavutil/pixdesc.h"
41 #include "avfilter.h"
42 #include "filters.h"
43 #include "video.h"
44
45 typedef struct OWDenoiseContext {
46 const AVClass *class;
47 double luma_strength;
48 double chroma_strength;
49 int depth;
50 float *plane[16+1][4];
51 int linesize;
52 int hsub, vsub;
53 int pixel_depth;
54 } OWDenoiseContext;
55
56 #define OFFSET(x) offsetof(OWDenoiseContext, x)
57 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
58 static const AVOption owdenoise_options[] = {
59 { "depth", "set depth", OFFSET(depth), AV_OPT_TYPE_INT, {.i64 = 8}, 8, 16, FLAGS },
60 { "luma_strength", "set luma strength", OFFSET(luma_strength), AV_OPT_TYPE_DOUBLE, {.dbl = 1.0}, 0, 1000, FLAGS },
61 { "ls", "set luma strength", OFFSET(luma_strength), AV_OPT_TYPE_DOUBLE, {.dbl = 1.0}, 0, 1000, FLAGS },
62 { "chroma_strength", "set chroma strength", OFFSET(chroma_strength), AV_OPT_TYPE_DOUBLE, {.dbl = 1.0}, 0, 1000, FLAGS },
63 { "cs", "set chroma strength", OFFSET(chroma_strength), AV_OPT_TYPE_DOUBLE, {.dbl = 1.0}, 0, 1000, FLAGS },
64 { NULL }
65 };
66
67 AVFILTER_DEFINE_CLASS(owdenoise);
68
69 DECLARE_ALIGNED(8, static const uint8_t, dither)[8][8] = {
70 { 0, 48, 12, 60, 3, 51, 15, 63 },
71 { 32, 16, 44, 28, 35, 19, 47, 31 },
72 { 8, 56, 4, 52, 11, 59, 7, 55 },
73 { 40, 24, 36, 20, 43, 27, 39, 23 },
74 { 2, 50, 14, 62, 1, 49, 13, 61 },
75 { 34, 18, 46, 30, 33, 17, 45, 29 },
76 { 10, 58, 6, 54, 9, 57, 5, 53 },
77 { 42, 26, 38, 22, 41, 25, 37, 21 },
78 };
79
80 static const double coeff[2][5] = {
81 {
82 0.6029490182363579 * M_SQRT2,
83 0.2668641184428723 * M_SQRT2,
84 -0.07822326652898785 * M_SQRT2,
85 -0.01686411844287495 * M_SQRT2,
86 0.02674875741080976 * M_SQRT2,
87 },{
88 1.115087052456994 / M_SQRT2,
89 -0.5912717631142470 / M_SQRT2,
90 -0.05754352622849957 / M_SQRT2,
91 0.09127176311424948 / M_SQRT2,
92 }
93 };
94
95 static const double icoeff[2][5] = {
96 {
97 1.115087052456994 / M_SQRT2,
98 0.5912717631142470 / M_SQRT2,
99 -0.05754352622849957 / M_SQRT2,
100 -0.09127176311424948 / M_SQRT2,
101 },{
102 0.6029490182363579 * M_SQRT2,
103 -0.2668641184428723 * M_SQRT2,
104 -0.07822326652898785 * M_SQRT2,
105 0.01686411844287495 * M_SQRT2,
106 0.02674875741080976 * M_SQRT2,
107 }
108 };
109
110
111 334400 static inline void decompose(float *dst_l, float *dst_h, const float *src,
112 int linesize, int w)
113 {
114 int x, i;
115
2/2
✓ Branch 0 taken 4608000 times.
✓ Branch 1 taken 334400 times.
 4942400 for (x = 0; x < w; x++) {
116 4608000 double sum_l = src[x * linesize] * coeff[0][0];
117 4608000 double sum_h = src[x * linesize] * coeff[1][0];
118
2/2
✓ Branch 0 taken 18432000 times.
✓ Branch 1 taken 4608000 times.
 23040000 for (i = 1; i <= 4; i++) {
119 18432000 const double s = src[avpriv_mirror(x - i, w - 1) * linesize]
120 18432000 + src[avpriv_mirror(x + i, w - 1) * linesize];
121
122 18432000 sum_l += coeff[0][i] * s;
123 18432000 sum_h += coeff[1][i] * s;
124 }
125 4608000 dst_l[x * linesize] = sum_l;
126 4608000 dst_h[x * linesize] = sum_h;
127 }
128 334400 }
129
130 334400 static inline void compose(float *dst, const float *src_l, const float *src_h,
131 int linesize, int w)
132 {
133 int x, i;
134
2/2
✓ Branch 0 taken 4608000 times.
✓ Branch 1 taken 334400 times.
 4942400 for (x = 0; x < w; x++) {
135 4608000 double sum_l = src_l[x * linesize] * icoeff[0][0];
136 4608000 double sum_h = src_h[x * linesize] * icoeff[1][0];
137
2/2
✓ Branch 0 taken 18432000 times.
✓ Branch 1 taken 4608000 times.
 23040000 for (i = 1; i <= 4; i++) {
138 18432000 const int x0 = avpriv_mirror(x - i, w - 1) * linesize;
139 18432000 const int x1 = avpriv_mirror(x + i, w - 1) * linesize;
140
141 18432000 sum_l += icoeff[0][i] * (src_l[x0] + src_l[x1]);
142 18432000 sum_h += icoeff[1][i] * (src_h[x0] + src_h[x1]);
143 }
144 4608000 dst[x * linesize] = (sum_l + sum_h) * 0.5;
145 }
146 334400 }
147
148 114 static inline void decompose2D(float *dst_l, float *dst_h, const float *src,
149 int xlinesize, int ylinesize,
150 int step, int w, int h)
151 {
152 int y, x;
153
2/2
✓ Branch 0 taken 22880 times.
✓ Branch 1 taken 114 times.
 22994 for (y = 0; y < h; y++)
154
2/2
✓ Branch 0 taken 334400 times.
✓ Branch 1 taken 22880 times.
 357280 for (x = 0; x < step; x++)
155 334400 decompose(dst_l + ylinesize*y + xlinesize*x,
156 334400 dst_h + ylinesize*y + xlinesize*x,
157 334400 src + ylinesize*y + xlinesize*x,
158 334400 step * xlinesize, (w - x + step - 1) / step);
159 114 }
160
161 114 static inline void compose2D(float *dst, const float *src_l, const float *src_h,
162 int xlinesize, int ylinesize,
163 int step, int w, int h)
164 {
165 int y, x;
166
2/2
✓ Branch 0 taken 22880 times.
✓ Branch 1 taken 114 times.
 22994 for (y = 0; y < h; y++)
167
2/2
✓ Branch 0 taken 334400 times.
✓ Branch 1 taken 22880 times.
 357280 for (x = 0; x < step; x++)
168 334400 compose(dst + ylinesize*y + xlinesize*x,
169 334400 src_l + ylinesize*y + xlinesize*x,
170 334400 src_h + ylinesize*y + xlinesize*x,
171 334400 step * xlinesize, (w - x + step - 1) / step);
172 114 }
173
174 38 static void decompose2D2(float *dst[4], float *src, float *temp[2],
175 int linesize, int step, int w, int h)
176 {
177 38 decompose2D(temp[0], temp[1], src, 1, linesize, step, w, h);
178 38 decompose2D( dst[0], dst[1], temp[0], linesize, 1, step, h, w);
179 38 decompose2D( dst[2], dst[3], temp[1], linesize, 1, step, h, w);
180 38 }
181
182 38 static void compose2D2(float *dst, float *src[4], float *temp[2],
183 int linesize, int step, int w, int h)
184 {
185 38 compose2D(temp[0], src[0], src[1], linesize, 1, step, h, w);
186 38 compose2D(temp[1], src[2], src[3], linesize, 1, step, h, w);
187 38 compose2D(dst, temp[0], temp[1], 1, linesize, step, w, h);
188 38 }
189
190 6 static void filter(OWDenoiseContext *s,
191 uint8_t *dst, int dst_linesize,
192 const uint8_t *src, int src_linesize,
193 int width, int height, double strength)
194 {
195 6 int x, y, i, j, depth = s->depth;
196
197
4/4
✓ Branch 0 taken 16 times.
✓ Branch 1 taken 12 times.
✓ Branch 2 taken 6 times.
✓ Branch 3 taken 6 times.
 28 while (1<<depth > width || 1<<depth > height)
198 22 depth--;
199
200
1/2
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
 6 if (s->pixel_depth <= 8) {
201
2/2
✓ Branch 0 taken 960 times.
✓ Branch 1 taken 6 times.
 966 for (y = 0; y < height; y++)
202
2/2
✓ Branch 0 taken 230400 times.
✓ Branch 1 taken 960 times.
 231360 for(x = 0; x < width; x++)
203 230400 s->plane[0][0][y*s->linesize + x] = src[y*src_linesize + x];
204 } else {
205 ✗ const uint16_t *src16 = (const uint16_t *)src;
206
207 ✗ src_linesize /= 2;
208 ✗ for (y = 0; y < height; y++)
209 ✗ for(x = 0; x < width; x++)
210 ✗ s->plane[0][0][y*s->linesize + x] = src16[y*src_linesize + x];
211 }
212
213
2/2
✓ Branch 0 taken 38 times.
✓ Branch 1 taken 6 times.
 44 for (i = 0; i < depth; i++)
214 38 decompose2D2(s->plane[i + 1], s->plane[i][0], s->plane[0] + 1, s->linesize, 1<<i, width, height);
215
216
2/2
✓ Branch 0 taken 38 times.
✓ Branch 1 taken 6 times.
 44 for (i = 0; i < depth; i++) {
217
2/2
✓ Branch 0 taken 114 times.
✓ Branch 1 taken 38 times.
 152 for (j = 1; j < 4; j++) {
218
2/2
✓ Branch 0 taken 18720 times.
✓ Branch 1 taken 114 times.
 18834 for (y = 0; y < height; y++) {
219
2/2
✓ Branch 0 taken 4608000 times.
✓ Branch 1 taken 18720 times.
 4626720 for (x = 0; x < width; x++) {
220 4608000 double v = s->plane[i + 1][j][y*s->linesize + x];
221
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4608000 times.
 4608000 if (v > strength) v -= strength;
222
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4608000 times.
 4608000 else if (v < -strength) v += strength;
223 4608000 else v = 0;
224 4608000 s->plane[i + 1][j][x + y*s->linesize] = v;
225 }
226 }
227 }
228 }
229
2/2
✓ Branch 0 taken 38 times.
✓ Branch 1 taken 6 times.
 44 for (i = depth-1; i >= 0; i--)
230 38 compose2D2(s->plane[i][0], s->plane[i + 1], s->plane[0] + 1, s->linesize, 1<<i, width, height);
231
232
1/2
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
 6 if (s->pixel_depth <= 8) {
233
2/2
✓ Branch 0 taken 960 times.
✓ Branch 1 taken 6 times.
 966 for (y = 0; y < height; y++) {
234
2/2
✓ Branch 0 taken 230400 times.
✓ Branch 1 taken 960 times.
 231360 for (x = 0; x < width; x++) {
235 230400 i = s->plane[0][0][y*s->linesize + x] + dither[x&7][y&7]*(1.0/64) + 1.0/128; // yes the rounding is insane but optimal :)
236
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 230400 times.
 230400 if ((unsigned)i > 255U) i = ~(i >> 31);
237 230400 dst[y*dst_linesize + x] = i;
238 }
239 }
240 } else {
241 ✗ uint16_t *dst16 = (uint16_t *)dst;
242
243 ✗ dst_linesize /= 2;
244 ✗ for (y = 0; y < height; y++) {
245 ✗ for (x = 0; x < width; x++) {
246 ✗ i = s->plane[0][0][y*s->linesize + x];
247 ✗ dst16[y*dst_linesize + x] = i;
248 }
249 }
250 }
251 6 }
252
253 2 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
254 {
255 2 AVFilterContext *ctx = inlink->dst;
256 2 OWDenoiseContext *s = ctx->priv;
257 2 AVFilterLink *outlink = ctx->outputs[0];
258 AVFrame *out;
259 2 const int cw = AV_CEIL_RSHIFT(inlink->w, s->hsub);
260 2 const int ch = AV_CEIL_RSHIFT(inlink->h, s->vsub);
261
262
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
 2 if (av_frame_is_writable(in)) {
263 2 out = in;
264
265
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (s->luma_strength > 0)
266 2 filter(s, out->data[0], out->linesize[0], in->data[0], in->linesize[0], inlink->w, inlink->h, s->luma_strength);
267
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (s->chroma_strength > 0) {
268 2 filter(s, out->data[1], out->linesize[1], in->data[1], in->linesize[1], cw, ch, s->chroma_strength);
269 2 filter(s, out->data[2], out->linesize[2], in->data[2], in->linesize[2], cw, ch, s->chroma_strength);
270 }
271 } else {
272 ✗ out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
273 ✗ if (!out) {
274 ✗ av_frame_free(&in);
275 ✗ return AVERROR(ENOMEM);
276 }
277 ✗ av_frame_copy_props(out, in);
278
279 ✗ if (s->luma_strength > 0) {
280 ✗ filter(s, out->data[0], out->linesize[0], in->data[0], in->linesize[0], inlink->w, inlink->h, s->luma_strength);
281 } else {
282 ✗ av_image_copy_plane(out->data[0], out->linesize[0], in ->data[0], in ->linesize[0], inlink->w, inlink->h);
283 }
284 ✗ if (s->chroma_strength > 0) {
285 ✗ filter(s, out->data[1], out->linesize[1], in->data[1], in->linesize[1], cw, ch, s->chroma_strength);
286 ✗ filter(s, out->data[2], out->linesize[2], in->data[2], in->linesize[2], cw, ch, s->chroma_strength);
287 } else {
288 ✗ av_image_copy_plane(out->data[1], out->linesize[1], in ->data[1], in ->linesize[1], inlink->w, inlink->h);
289 ✗ av_image_copy_plane(out->data[2], out->linesize[2], in ->data[2], in ->linesize[2], inlink->w, inlink->h);
290 }
291
292 ✗ if (in->data[3])
293 ✗ av_image_copy_plane(out->data[3], out->linesize[3],
294 ✗ in ->data[3], in ->linesize[3],
295 inlink->w, inlink->h);
296 ✗ av_frame_free(&in);
297 }
298
299 2 return ff_filter_frame(outlink, out);
300 }
301
302 static const enum AVPixelFormat pix_fmts[] = {
303 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P,
304 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P,
305 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV440P,
306 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUVA422P,
307 AV_PIX_FMT_YUVA420P,
308 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
309 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
310 AV_PIX_FMT_YUV440P10,
311 AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV420P12,
312 AV_PIX_FMT_YUV440P12,
313 AV_PIX_FMT_YUV444P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV420P14,
314 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
315 AV_PIX_FMT_NONE
316 };
317
318 1 static int config_input(AVFilterLink *inlink)
319 {
320 int i, j;
321 1 OWDenoiseContext *s = inlink->dst->priv;
322 1 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
323 1 const int h = FFALIGN(inlink->h, 16);
324
325 1 s->hsub = desc->log2_chroma_w;
326 1 s->vsub = desc->log2_chroma_h;
327 1 s->pixel_depth = desc->comp[0].depth;
328
329 1 s->linesize = FFALIGN(inlink->w, 16);
330
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1 times.
 5 for (j = 0; j < 4; j++) {
331
2/2
✓ Branch 0 taken 44 times.
✓ Branch 1 taken 4 times.
 48 for (i = 0; i <= s->depth; i++) {
332 44 s->plane[i][j] = av_malloc_array(s->linesize, h * sizeof(s->plane[0][0][0]));
333
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 44 times.
 44 if (!s->plane[i][j])
334 ✗ return AVERROR(ENOMEM);
335 }
336 }
337 1 return 0;
338 }
339
340 2 static av_cold void uninit(AVFilterContext *ctx)
341 {
342 int i, j;
343 2 OWDenoiseContext *s = ctx->priv;
344
345
2/2
✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
 10 for (j = 0; j < 4; j++)
346
2/2
✓ Branch 0 taken 88 times.
✓ Branch 1 taken 8 times.
 96 for (i = 0; i <= s->depth; i++)
347 88 av_freep(&s->plane[i][j]);
348 2 }
349
350 static const AVFilterPad owdenoise_inputs[] = {
351 {
352 .name = "default",
353 .type = AVMEDIA_TYPE_VIDEO,
354 .filter_frame = filter_frame,
355 .config_props = config_input,
356 },
357 };
358
359 const AVFilter ff_vf_owdenoise = {
360 .name = "owdenoise",
361 .description = NULL_IF_CONFIG_SMALL("Denoise using wavelets."),
362 .priv_size = sizeof(OWDenoiseContext),
363 .uninit = uninit,
364 FILTER_INPUTS(owdenoise_inputs),
365 FILTER_OUTPUTS(ff_video_default_filterpad),
366 FILTER_PIXFMTS_ARRAY(pix_fmts),
367 .priv_class = &owdenoise_class,
368 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
369 };
370