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/opt.h"
38 #include "libavutil/mem_internal.h"
39 #include "libavutil/pixdesc.h"
40 #include "avfilter.h"
41 #include "internal.h"
42
43 typedef struct OWDenoiseContext {
44 const AVClass *class;
45 double luma_strength;
46 double chroma_strength;
47 int depth;
48 float *plane[16+1][4];
49 int linesize;
50 int hsub, vsub;
51 int pixel_depth;
52 } OWDenoiseContext;
53
54 #define OFFSET(x) offsetof(OWDenoiseContext, x)
55 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
56 static const AVOption owdenoise_options[] = {
57 { "depth", "set depth", OFFSET(depth), AV_OPT_TYPE_INT, {.i64 = 8}, 8, 16, FLAGS },
58 { "luma_strength", "set luma strength", OFFSET(luma_strength), AV_OPT_TYPE_DOUBLE, {.dbl = 1.0}, 0, 1000, FLAGS },
59 { "ls", "set luma strength", OFFSET(luma_strength), AV_OPT_TYPE_DOUBLE, {.dbl = 1.0}, 0, 1000, FLAGS },
60 { "chroma_strength", "set chroma strength", OFFSET(chroma_strength), AV_OPT_TYPE_DOUBLE, {.dbl = 1.0}, 0, 1000, FLAGS },
61 { "cs", "set chroma strength", OFFSET(chroma_strength), AV_OPT_TYPE_DOUBLE, {.dbl = 1.0}, 0, 1000, FLAGS },
62 { NULL }
63 };
64
65 AVFILTER_DEFINE_CLASS(owdenoise);
66
67 DECLARE_ALIGNED(8, static const uint8_t, dither)[8][8] = {
68 { 0, 48, 12, 60, 3, 51, 15, 63 },
69 { 32, 16, 44, 28, 35, 19, 47, 31 },
70 { 8, 56, 4, 52, 11, 59, 7, 55 },
71 { 40, 24, 36, 20, 43, 27, 39, 23 },
72 { 2, 50, 14, 62, 1, 49, 13, 61 },
73 { 34, 18, 46, 30, 33, 17, 45, 29 },
74 { 10, 58, 6, 54, 9, 57, 5, 53 },
75 { 42, 26, 38, 22, 41, 25, 37, 21 },
76 };
77
78 static const double coeff[2][5] = {
79 {
80 0.6029490182363579 * M_SQRT2,
81 0.2668641184428723 * M_SQRT2,
82 -0.07822326652898785 * M_SQRT2,
83 -0.01686411844287495 * M_SQRT2,
84 0.02674875741080976 * M_SQRT2,
85 },{
86 1.115087052456994 / M_SQRT2,
87 -0.5912717631142470 / M_SQRT2,
88 -0.05754352622849957 / M_SQRT2,
89 0.09127176311424948 / M_SQRT2,
90 }
91 };
92
93 static const double icoeff[2][5] = {
94 {
95 1.115087052456994 / M_SQRT2,
96 0.5912717631142470 / M_SQRT2,
97 -0.05754352622849957 / M_SQRT2,
98 -0.09127176311424948 / M_SQRT2,
99 },{
100 0.6029490182363579 * M_SQRT2,
101 -0.2668641184428723 * M_SQRT2,
102 -0.07822326652898785 * M_SQRT2,
103 0.01686411844287495 * M_SQRT2,
104 0.02674875741080976 * M_SQRT2,
105 }
106 };
107
108
decompose(float * dst_l,float * dst_h,const float * src,int linesize,int w)109 static inline void decompose(float *dst_l, float *dst_h, const float *src,
110 int linesize, int w)
111 {
112 int x, i;
113 for (x = 0; x < w; x++) {
114 double sum_l = src[x * linesize] * coeff[0][0];
115 double sum_h = src[x * linesize] * coeff[1][0];
116 for (i = 1; i <= 4; i++) {
117 const double s = src[avpriv_mirror(x - i, w - 1) * linesize]
118 + src[avpriv_mirror(x + i, w - 1) * linesize];
119
120 sum_l += coeff[0][i] * s;
121 sum_h += coeff[1][i] * s;
122 }
123 dst_l[x * linesize] = sum_l;
124 dst_h[x * linesize] = sum_h;
125 }
126 }
127
compose(float * dst,const float * src_l,const float * src_h,int linesize,int w)128 static inline void compose(float *dst, const float *src_l, const float *src_h,
129 int linesize, int w)
130 {
131 int x, i;
132 for (x = 0; x < w; x++) {
133 double sum_l = src_l[x * linesize] * icoeff[0][0];
134 double sum_h = src_h[x * linesize] * icoeff[1][0];
135 for (i = 1; i <= 4; i++) {
136 const int x0 = avpriv_mirror(x - i, w - 1) * linesize;
137 const int x1 = avpriv_mirror(x + i, w - 1) * linesize;
138
139 sum_l += icoeff[0][i] * (src_l[x0] + src_l[x1]);
140 sum_h += icoeff[1][i] * (src_h[x0] + src_h[x1]);
141 }
142 dst[x * linesize] = (sum_l + sum_h) * 0.5;
143 }
144 }
145
decompose2D(float * dst_l,float * dst_h,const float * src,int xlinesize,int ylinesize,int step,int w,int h)146 static inline void decompose2D(float *dst_l, float *dst_h, const float *src,
147 int xlinesize, int ylinesize,
148 int step, int w, int h)
149 {
150 int y, x;
151 for (y = 0; y < h; y++)
152 for (x = 0; x < step; x++)
153 decompose(dst_l + ylinesize*y + xlinesize*x,
154 dst_h + ylinesize*y + xlinesize*x,
155 src + ylinesize*y + xlinesize*x,
156 step * xlinesize, (w - x + step - 1) / step);
157 }
158
compose2D(float * dst,const float * src_l,const float * src_h,int xlinesize,int ylinesize,int step,int w,int h)159 static inline void compose2D(float *dst, const float *src_l, const float *src_h,
160 int xlinesize, int ylinesize,
161 int step, int w, int h)
162 {
163 int y, x;
164 for (y = 0; y < h; y++)
165 for (x = 0; x < step; x++)
166 compose(dst + ylinesize*y + xlinesize*x,
167 src_l + ylinesize*y + xlinesize*x,
168 src_h + ylinesize*y + xlinesize*x,
169 step * xlinesize, (w - x + step - 1) / step);
170 }
171
decompose2D2(float * dst[4],float * src,float * temp[2],int linesize,int step,int w,int h)172 static void decompose2D2(float *dst[4], float *src, float *temp[2],
173 int linesize, int step, int w, int h)
174 {
175 decompose2D(temp[0], temp[1], src, 1, linesize, step, w, h);
176 decompose2D( dst[0], dst[1], temp[0], linesize, 1, step, h, w);
177 decompose2D( dst[2], dst[3], temp[1], linesize, 1, step, h, w);
178 }
179
compose2D2(float * dst,float * src[4],float * temp[2],int linesize,int step,int w,int h)180 static void compose2D2(float *dst, float *src[4], float *temp[2],
181 int linesize, int step, int w, int h)
182 {
183 compose2D(temp[0], src[0], src[1], linesize, 1, step, h, w);
184 compose2D(temp[1], src[2], src[3], linesize, 1, step, h, w);
185 compose2D(dst, temp[0], temp[1], 1, linesize, step, w, h);
186 }
187
filter(OWDenoiseContext * s,uint8_t * dst,int dst_linesize,const uint8_t * src,int src_linesize,int width,int height,double strength)188 static void filter(OWDenoiseContext *s,
189 uint8_t *dst, int dst_linesize,
190 const uint8_t *src, int src_linesize,
191 int width, int height, double strength)
192 {
193 int x, y, i, j, depth = s->depth;
194
195 while (1<<depth > width || 1<<depth > height)
196 depth--;
197
198 if (s->pixel_depth <= 8) {
199 for (y = 0; y < height; y++)
200 for(x = 0; x < width; x++)
201 s->plane[0][0][y*s->linesize + x] = src[y*src_linesize + x];
202 } else {
203 const uint16_t *src16 = (const uint16_t *)src;
204
205 src_linesize /= 2;
206 for (y = 0; y < height; y++)
207 for(x = 0; x < width; x++)
208 s->plane[0][0][y*s->linesize + x] = src16[y*src_linesize + x];
209 }
210
211 for (i = 0; i < depth; i++)
212 decompose2D2(s->plane[i + 1], s->plane[i][0], s->plane[0] + 1, s->linesize, 1<<i, width, height);
213
214 for (i = 0; i < depth; i++) {
215 for (j = 1; j < 4; j++) {
216 for (y = 0; y < height; y++) {
217 for (x = 0; x < width; x++) {
218 double v = s->plane[i + 1][j][y*s->linesize + x];
219 if (v > strength) v -= strength;
220 else if (v < -strength) v += strength;
221 else v = 0;
222 s->plane[i + 1][j][x + y*s->linesize] = v;
223 }
224 }
225 }
226 }
227 for (i = depth-1; i >= 0; i--)
228 compose2D2(s->plane[i][0], s->plane[i + 1], s->plane[0] + 1, s->linesize, 1<<i, width, height);
229
230 if (s->pixel_depth <= 8) {
231 for (y = 0; y < height; y++) {
232 for (x = 0; x < width; x++) {
233 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 :)
234 if ((unsigned)i > 255U) i = ~(i >> 31);
235 dst[y*dst_linesize + x] = i;
236 }
237 }
238 } else {
239 uint16_t *dst16 = (uint16_t *)dst;
240
241 dst_linesize /= 2;
242 for (y = 0; y < height; y++) {
243 for (x = 0; x < width; x++) {
244 i = s->plane[0][0][y*s->linesize + x];
245 dst16[y*dst_linesize + x] = i;
246 }
247 }
248 }
249 }
250
filter_frame(AVFilterLink * inlink,AVFrame * in)251 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
252 {
253 AVFilterContext *ctx = inlink->dst;
254 OWDenoiseContext *s = ctx->priv;
255 AVFilterLink *outlink = ctx->outputs[0];
256 AVFrame *out;
257 const int cw = AV_CEIL_RSHIFT(inlink->w, s->hsub);
258 const int ch = AV_CEIL_RSHIFT(inlink->h, s->vsub);
259
260 if (av_frame_is_writable(in)) {
261 out = in;
262
263 if (s->luma_strength > 0)
264 filter(s, out->data[0], out->linesize[0], in->data[0], in->linesize[0], inlink->w, inlink->h, s->luma_strength);
265 if (s->chroma_strength > 0) {
266 filter(s, out->data[1], out->linesize[1], in->data[1], in->linesize[1], cw, ch, s->chroma_strength);
267 filter(s, out->data[2], out->linesize[2], in->data[2], in->linesize[2], cw, ch, s->chroma_strength);
268 }
269 } else {
270 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
271 if (!out) {
272 av_frame_free(&in);
273 return AVERROR(ENOMEM);
274 }
275 av_frame_copy_props(out, in);
276
277 if (s->luma_strength > 0) {
278 filter(s, out->data[0], out->linesize[0], in->data[0], in->linesize[0], inlink->w, inlink->h, s->luma_strength);
279 } else {
280 av_image_copy_plane(out->data[0], out->linesize[0], in ->data[0], in ->linesize[0], inlink->w, inlink->h);
281 }
282 if (s->chroma_strength > 0) {
283 filter(s, out->data[1], out->linesize[1], in->data[1], in->linesize[1], cw, ch, s->chroma_strength);
284 filter(s, out->data[2], out->linesize[2], in->data[2], in->linesize[2], cw, ch, s->chroma_strength);
285 } else {
286 av_image_copy_plane(out->data[1], out->linesize[1], in ->data[1], in ->linesize[1], inlink->w, inlink->h);
287 av_image_copy_plane(out->data[2], out->linesize[2], in ->data[2], in ->linesize[2], inlink->w, inlink->h);
288 }
289
290 if (in->data[3])
291 av_image_copy_plane(out->data[3], out->linesize[3],
292 in ->data[3], in ->linesize[3],
293 inlink->w, inlink->h);
294 av_frame_free(&in);
295 }
296
297 return ff_filter_frame(outlink, out);
298 }
299
300 static const enum AVPixelFormat pix_fmts[] = {
301 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P,
302 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P,
303 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV440P,
304 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUVA422P,
305 AV_PIX_FMT_YUVA420P,
306 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
307 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
308 AV_PIX_FMT_YUV440P10,
309 AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV420P12,
310 AV_PIX_FMT_YUV440P12,
311 AV_PIX_FMT_YUV444P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV420P14,
312 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
313 AV_PIX_FMT_NONE
314 };
315
config_input(AVFilterLink * inlink)316 static int config_input(AVFilterLink *inlink)
317 {
318 int i, j;
319 OWDenoiseContext *s = inlink->dst->priv;
320 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
321 const int h = FFALIGN(inlink->h, 16);
322
323 s->hsub = desc->log2_chroma_w;
324 s->vsub = desc->log2_chroma_h;
325 s->pixel_depth = desc->comp[0].depth;
326
327 s->linesize = FFALIGN(inlink->w, 16);
328 for (j = 0; j < 4; j++) {
329 for (i = 0; i <= s->depth; i++) {
330 s->plane[i][j] = av_malloc_array(s->linesize, h * sizeof(s->plane[0][0][0]));
331 if (!s->plane[i][j])
332 return AVERROR(ENOMEM);
333 }
334 }
335 return 0;
336 }
337
uninit(AVFilterContext * ctx)338 static av_cold void uninit(AVFilterContext *ctx)
339 {
340 int i, j;
341 OWDenoiseContext *s = ctx->priv;
342
343 for (j = 0; j < 4; j++)
344 for (i = 0; i <= s->depth; i++)
345 av_freep(&s->plane[i][j]);
346 }
347
348 static const AVFilterPad owdenoise_inputs[] = {
349 {
350 .name = "default",
351 .type = AVMEDIA_TYPE_VIDEO,
352 .filter_frame = filter_frame,
353 .config_props = config_input,
354 },
355 };
356
357 static const AVFilterPad owdenoise_outputs[] = {
358 {
359 .name = "default",
360 .type = AVMEDIA_TYPE_VIDEO,
361 },
362 };
363
364 const AVFilter ff_vf_owdenoise = {
365 .name = "owdenoise",
366 .description = NULL_IF_CONFIG_SMALL("Denoise using wavelets."),
367 .priv_size = sizeof(OWDenoiseContext),
368 .uninit = uninit,
369 FILTER_INPUTS(owdenoise_inputs),
370 FILTER_OUTPUTS(owdenoise_outputs),
371 FILTER_PIXFMTS_ARRAY(pix_fmts),
372 .priv_class = &owdenoise_class,
373 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
374 };
375