FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libavfilter/vf_dctdnoiz.c
Date: 2024-04-25 15:36:26
Exec Total Coverage
Lines: 0 478 0.0%
Functions: 0 23 0.0%
Branches: 0 174 0.0%
Line Branch Exec Source
1 /*
2 * Copyright (c) 2013-2014 Clément Bœsch
3 *
4 * This file is part of FFmpeg.
5 *
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.
10 *
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.
15 *
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
19 */
20
21 /**
22 * A simple, relatively efficient and slow DCT image denoiser.
23 *
24 * @see http://www.ipol.im/pub/art/2011/ys-dct/
25 *
26 * The DCT factorization used is based on "Fast and numerically stable
27 * algorithms for discrete cosine transforms" from Gerlind Plonkaa & Manfred
28 * Tasche (DOI: 10.1016/j.laa.2004.07.015).
29 */
30
31 #include "libavutil/avassert.h"
32 #include "libavutil/eval.h"
33 #include "libavutil/mem.h"
34 #include "libavutil/mem_internal.h"
35 #include "libavutil/opt.h"
36 #include "internal.h"
37 #include "video.h"
38
39 static const char *const var_names[] = { "c", NULL };
40 enum { VAR_C, VAR_VARS_NB };
41
42 #define MAX_THREADS 8
43
44 typedef struct DCTdnoizContext {
45 const AVClass *class;
46
47 /* coefficient factor expression */
48 char *expr_str;
49 AVExpr *expr[MAX_THREADS];
50 double var_values[MAX_THREADS][VAR_VARS_NB];
51
52 int nb_threads;
53 int pr_width, pr_height; // width and height to process
54 float sigma; // used when no expression are st
55 float th; // threshold (3*sigma)
56 float *cbuf[2][3]; // two planar rgb color buffers
57 float *slices[MAX_THREADS]; // slices buffers (1 slice buffer per thread)
58 float *weights; // dct coeff are cumulated with overlapping; these values are used for averaging
59 int p_linesize; // line sizes for color and weights
60 int overlap; // number of block overlapping pixels
61 int step; // block step increment (blocksize - overlap)
62 int n; // 1<<n is the block size
63 int bsize; // block size, 1<<n
64 void (*filter_freq_func)(struct DCTdnoizContext *s,
65 const float *src, int src_linesize,
66 float *dst, int dst_linesize,
67 int thread_id);
68 void (*color_decorrelation)(float **dst, int dst_linesize,
69 const uint8_t **src, int src_linesize,
70 int w, int h);
71 void (*color_correlation)(uint8_t **dst, int dst_linesize,
72 float **src, int src_linesize,
73 int w, int h);
74 } DCTdnoizContext;
75
76 #define MIN_NBITS 3 /* blocksize = 1<<3 = 8 */
77 #define MAX_NBITS 4 /* blocksize = 1<<4 = 16 */
78 #define DEFAULT_NBITS 3
79
80 #define OFFSET(x) offsetof(DCTdnoizContext, x)
81 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
82 static const AVOption dctdnoiz_options[] = {
83 { "sigma", "set noise sigma constant", OFFSET(sigma), AV_OPT_TYPE_FLOAT, {.dbl=0}, 0, 999, .flags = FLAGS },
84 { "s", "set noise sigma constant", OFFSET(sigma), AV_OPT_TYPE_FLOAT, {.dbl=0}, 0, 999, .flags = FLAGS },
85 { "overlap", "set number of block overlapping pixels", OFFSET(overlap), AV_OPT_TYPE_INT, {.i64=-1}, -1, (1<<MAX_NBITS)-1, .flags = FLAGS },
86 { "expr", "set coefficient factor expression", OFFSET(expr_str), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
87 { "e", "set coefficient factor expression", OFFSET(expr_str), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
88 { "n", "set the block size, expressed in bits", OFFSET(n), AV_OPT_TYPE_INT, {.i64=DEFAULT_NBITS}, MIN_NBITS, MAX_NBITS, .flags = FLAGS },
89 { NULL }
90 };
91
92 AVFILTER_DEFINE_CLASS(dctdnoiz);
93
94 ✗ static void av_always_inline fdct8_1d(float *dst, const float *src,
95 int dst_stridea, int dst_strideb,
96 int src_stridea, int src_strideb)
97 {
98 int i;
99
100 ✗ for (i = 0; i < 8; i++) {
101 ✗ const float x00 = src[0*src_stridea] + src[7*src_stridea];
102 ✗ const float x01 = src[1*src_stridea] + src[6*src_stridea];
103 ✗ const float x02 = src[2*src_stridea] + src[5*src_stridea];
104 ✗ const float x03 = src[3*src_stridea] + src[4*src_stridea];
105 ✗ const float x04 = src[0*src_stridea] - src[7*src_stridea];
106 ✗ const float x05 = src[1*src_stridea] - src[6*src_stridea];
107 ✗ const float x06 = src[2*src_stridea] - src[5*src_stridea];
108 ✗ const float x07 = src[3*src_stridea] - src[4*src_stridea];
109 ✗ const float x08 = x00 + x03;
110 ✗ const float x09 = x01 + x02;
111 ✗ const float x0a = x00 - x03;
112 ✗ const float x0b = x01 - x02;
113 ✗ const float x0c = 1.38703984532215f*x04 + 0.275899379282943f*x07;
114 ✗ const float x0d = 1.17587560241936f*x05 + 0.785694958387102f*x06;
115 ✗ const float x0e = -0.785694958387102f*x05 + 1.17587560241936f*x06;
116 ✗ const float x0f = 0.275899379282943f*x04 - 1.38703984532215f*x07;
117 ✗ const float x10 = 0.353553390593274f * (x0c - x0d);
118 ✗ const float x11 = 0.353553390593274f * (x0e - x0f);
119 ✗ dst[0*dst_stridea] = 0.353553390593274f * (x08 + x09);
120 ✗ dst[1*dst_stridea] = 0.353553390593274f * (x0c + x0d);
121 ✗ dst[2*dst_stridea] = 0.461939766255643f*x0a + 0.191341716182545f*x0b;
122 ✗ dst[3*dst_stridea] = 0.707106781186547f * (x10 - x11);
123 ✗ dst[4*dst_stridea] = 0.353553390593274f * (x08 - x09);
124 ✗ dst[5*dst_stridea] = 0.707106781186547f * (x10 + x11);
125 ✗ dst[6*dst_stridea] = 0.191341716182545f*x0a - 0.461939766255643f*x0b;
126 ✗ dst[7*dst_stridea] = 0.353553390593274f * (x0e + x0f);
127 ✗ dst += dst_strideb;
128 ✗ src += src_strideb;
129 }
130 ✗ }
131
132 ✗ static void av_always_inline idct8_1d(float *dst, const float *src,
133 int dst_stridea, int dst_strideb,
134 int src_stridea, int src_strideb,
135 int add)
136 {
137 int i;
138
139 ✗ for (i = 0; i < 8; i++) {
140 ✗ const float x00 = 1.4142135623731f *src[0*src_stridea];
141 ✗ const float x01 = 1.38703984532215f *src[1*src_stridea] + 0.275899379282943f*src[7*src_stridea];
142 ✗ const float x02 = 1.30656296487638f *src[2*src_stridea] + 0.541196100146197f*src[6*src_stridea];
143 ✗ const float x03 = 1.17587560241936f *src[3*src_stridea] + 0.785694958387102f*src[5*src_stridea];
144 ✗ const float x04 = 1.4142135623731f *src[4*src_stridea];
145 ✗ const float x05 = -0.785694958387102f*src[3*src_stridea] + 1.17587560241936f*src[5*src_stridea];
146 ✗ const float x06 = 0.541196100146197f*src[2*src_stridea] - 1.30656296487638f*src[6*src_stridea];
147 ✗ const float x07 = -0.275899379282943f*src[1*src_stridea] + 1.38703984532215f*src[7*src_stridea];
148 ✗ const float x09 = x00 + x04;
149 ✗ const float x0a = x01 + x03;
150 ✗ const float x0b = 1.4142135623731f*x02;
151 ✗ const float x0c = x00 - x04;
152 ✗ const float x0d = x01 - x03;
153 ✗ const float x0e = 0.353553390593274f * (x09 - x0b);
154 ✗ const float x0f = 0.353553390593274f * (x0c + x0d);
155 ✗ const float x10 = 0.353553390593274f * (x0c - x0d);
156 ✗ const float x11 = 1.4142135623731f*x06;
157 ✗ const float x12 = x05 + x07;
158 ✗ const float x13 = x05 - x07;
159 ✗ const float x14 = 0.353553390593274f * (x11 + x12);
160 ✗ const float x15 = 0.353553390593274f * (x11 - x12);
161 ✗ const float x16 = 0.5f*x13;
162 ✗ dst[0*dst_stridea] = (add ? dst[ 0*dst_stridea] : 0) + 0.25f * (x09 + x0b) + 0.353553390593274f*x0a;
163 ✗ dst[1*dst_stridea] = (add ? dst[ 1*dst_stridea] : 0) + 0.707106781186547f * (x0f + x15);
164 ✗ dst[2*dst_stridea] = (add ? dst[ 2*dst_stridea] : 0) + 0.707106781186547f * (x0f - x15);
165 ✗ dst[3*dst_stridea] = (add ? dst[ 3*dst_stridea] : 0) + 0.707106781186547f * (x0e + x16);
166 ✗ dst[4*dst_stridea] = (add ? dst[ 4*dst_stridea] : 0) + 0.707106781186547f * (x0e - x16);
167 ✗ dst[5*dst_stridea] = (add ? dst[ 5*dst_stridea] : 0) + 0.707106781186547f * (x10 - x14);
168 ✗ dst[6*dst_stridea] = (add ? dst[ 6*dst_stridea] : 0) + 0.707106781186547f * (x10 + x14);
169 ✗ dst[7*dst_stridea] = (add ? dst[ 7*dst_stridea] : 0) + 0.25f * (x09 + x0b) - 0.353553390593274f*x0a;
170 ✗ dst += dst_strideb;
171 ✗ src += src_strideb;
172 }
173 ✗ }
174
175
176 ✗ static void av_always_inline fdct16_1d(float *dst, const float *src,
177 int dst_stridea, int dst_strideb,
178 int src_stridea, int src_strideb)
179 {
180 int i;
181
182 ✗ for (i = 0; i < 16; i++) {
183 ✗ const float x00 = src[ 0*src_stridea] + src[15*src_stridea];
184 ✗ const float x01 = src[ 1*src_stridea] + src[14*src_stridea];
185 ✗ const float x02 = src[ 2*src_stridea] + src[13*src_stridea];
186 ✗ const float x03 = src[ 3*src_stridea] + src[12*src_stridea];
187 ✗ const float x04 = src[ 4*src_stridea] + src[11*src_stridea];
188 ✗ const float x05 = src[ 5*src_stridea] + src[10*src_stridea];
189 ✗ const float x06 = src[ 6*src_stridea] + src[ 9*src_stridea];
190 ✗ const float x07 = src[ 7*src_stridea] + src[ 8*src_stridea];
191 ✗ const float x08 = src[ 0*src_stridea] - src[15*src_stridea];
192 ✗ const float x09 = src[ 1*src_stridea] - src[14*src_stridea];
193 ✗ const float x0a = src[ 2*src_stridea] - src[13*src_stridea];
194 ✗ const float x0b = src[ 3*src_stridea] - src[12*src_stridea];
195 ✗ const float x0c = src[ 4*src_stridea] - src[11*src_stridea];
196 ✗ const float x0d = src[ 5*src_stridea] - src[10*src_stridea];
197 ✗ const float x0e = src[ 6*src_stridea] - src[ 9*src_stridea];
198 ✗ const float x0f = src[ 7*src_stridea] - src[ 8*src_stridea];
199 ✗ const float x10 = x00 + x07;
200 ✗ const float x11 = x01 + x06;
201 ✗ const float x12 = x02 + x05;
202 ✗ const float x13 = x03 + x04;
203 ✗ const float x14 = x00 - x07;
204 ✗ const float x15 = x01 - x06;
205 ✗ const float x16 = x02 - x05;
206 ✗ const float x17 = x03 - x04;
207 ✗ const float x18 = x10 + x13;
208 ✗ const float x19 = x11 + x12;
209 ✗ const float x1a = x10 - x13;
210 ✗ const float x1b = x11 - x12;
211 ✗ const float x1c = 1.38703984532215f*x14 + 0.275899379282943f*x17;
212 ✗ const float x1d = 1.17587560241936f*x15 + 0.785694958387102f*x16;
213 ✗ const float x1e = -0.785694958387102f*x15 + 1.17587560241936f *x16;
214 ✗ const float x1f = 0.275899379282943f*x14 - 1.38703984532215f *x17;
215 ✗ const float x20 = 0.25f * (x1c - x1d);
216 ✗ const float x21 = 0.25f * (x1e - x1f);
217 ✗ const float x22 = 1.40740373752638f *x08 + 0.138617169199091f*x0f;
218 ✗ const float x23 = 1.35331800117435f *x09 + 0.410524527522357f*x0e;
219 ✗ const float x24 = 1.24722501298667f *x0a + 0.666655658477747f*x0d;
220 ✗ const float x25 = 1.09320186700176f *x0b + 0.897167586342636f*x0c;
221 ✗ const float x26 = -0.897167586342636f*x0b + 1.09320186700176f *x0c;
222 ✗ const float x27 = 0.666655658477747f*x0a - 1.24722501298667f *x0d;
223 ✗ const float x28 = -0.410524527522357f*x09 + 1.35331800117435f *x0e;
224 ✗ const float x29 = 0.138617169199091f*x08 - 1.40740373752638f *x0f;
225 ✗ const float x2a = x22 + x25;
226 ✗ const float x2b = x23 + x24;
227 ✗ const float x2c = x22 - x25;
228 ✗ const float x2d = x23 - x24;
229 ✗ const float x2e = 0.25f * (x2a - x2b);
230 ✗ const float x2f = 0.326640741219094f*x2c + 0.135299025036549f*x2d;
231 ✗ const float x30 = 0.135299025036549f*x2c - 0.326640741219094f*x2d;
232 ✗ const float x31 = x26 + x29;
233 ✗ const float x32 = x27 + x28;
234 ✗ const float x33 = x26 - x29;
235 ✗ const float x34 = x27 - x28;
236 ✗ const float x35 = 0.25f * (x31 - x32);
237 ✗ const float x36 = 0.326640741219094f*x33 + 0.135299025036549f*x34;
238 ✗ const float x37 = 0.135299025036549f*x33 - 0.326640741219094f*x34;
239 ✗ dst[ 0*dst_stridea] = 0.25f * (x18 + x19);
240 ✗ dst[ 1*dst_stridea] = 0.25f * (x2a + x2b);
241 ✗ dst[ 2*dst_stridea] = 0.25f * (x1c + x1d);
242 ✗ dst[ 3*dst_stridea] = 0.707106781186547f * (x2f - x37);
243 ✗ dst[ 4*dst_stridea] = 0.326640741219094f*x1a + 0.135299025036549f*x1b;
244 ✗ dst[ 5*dst_stridea] = 0.707106781186547f * (x2f + x37);
245 ✗ dst[ 6*dst_stridea] = 0.707106781186547f * (x20 - x21);
246 ✗ dst[ 7*dst_stridea] = 0.707106781186547f * (x2e + x35);
247 ✗ dst[ 8*dst_stridea] = 0.25f * (x18 - x19);
248 ✗ dst[ 9*dst_stridea] = 0.707106781186547f * (x2e - x35);
249 ✗ dst[10*dst_stridea] = 0.707106781186547f * (x20 + x21);
250 ✗ dst[11*dst_stridea] = 0.707106781186547f * (x30 - x36);
251 ✗ dst[12*dst_stridea] = 0.135299025036549f*x1a - 0.326640741219094f*x1b;
252 ✗ dst[13*dst_stridea] = 0.707106781186547f * (x30 + x36);
253 ✗ dst[14*dst_stridea] = 0.25f * (x1e + x1f);
254 ✗ dst[15*dst_stridea] = 0.25f * (x31 + x32);
255 ✗ dst += dst_strideb;
256 ✗ src += src_strideb;
257 }
258 ✗ }
259
260 ✗ static void av_always_inline idct16_1d(float *dst, const float *src,
261 int dst_stridea, int dst_strideb,
262 int src_stridea, int src_strideb,
263 int add)
264 {
265 int i;
266
267 ✗ for (i = 0; i < 16; i++) {
268 ✗ const float x00 = 1.4142135623731f *src[ 0*src_stridea];
269 ✗ const float x01 = 1.40740373752638f *src[ 1*src_stridea] + 0.138617169199091f*src[15*src_stridea];
270 ✗ const float x02 = 1.38703984532215f *src[ 2*src_stridea] + 0.275899379282943f*src[14*src_stridea];
271 ✗ const float x03 = 1.35331800117435f *src[ 3*src_stridea] + 0.410524527522357f*src[13*src_stridea];
272 ✗ const float x04 = 1.30656296487638f *src[ 4*src_stridea] + 0.541196100146197f*src[12*src_stridea];
273 ✗ const float x05 = 1.24722501298667f *src[ 5*src_stridea] + 0.666655658477747f*src[11*src_stridea];
274 ✗ const float x06 = 1.17587560241936f *src[ 6*src_stridea] + 0.785694958387102f*src[10*src_stridea];
275 ✗ const float x07 = 1.09320186700176f *src[ 7*src_stridea] + 0.897167586342636f*src[ 9*src_stridea];
276 ✗ const float x08 = 1.4142135623731f *src[ 8*src_stridea];
277 ✗ const float x09 = -0.897167586342636f*src[ 7*src_stridea] + 1.09320186700176f*src[ 9*src_stridea];
278 ✗ const float x0a = 0.785694958387102f*src[ 6*src_stridea] - 1.17587560241936f*src[10*src_stridea];
279 ✗ const float x0b = -0.666655658477747f*src[ 5*src_stridea] + 1.24722501298667f*src[11*src_stridea];
280 ✗ const float x0c = 0.541196100146197f*src[ 4*src_stridea] - 1.30656296487638f*src[12*src_stridea];
281 ✗ const float x0d = -0.410524527522357f*src[ 3*src_stridea] + 1.35331800117435f*src[13*src_stridea];
282 ✗ const float x0e = 0.275899379282943f*src[ 2*src_stridea] - 1.38703984532215f*src[14*src_stridea];
283 ✗ const float x0f = -0.138617169199091f*src[ 1*src_stridea] + 1.40740373752638f*src[15*src_stridea];
284 ✗ const float x12 = x00 + x08;
285 ✗ const float x13 = x01 + x07;
286 ✗ const float x14 = x02 + x06;
287 ✗ const float x15 = x03 + x05;
288 ✗ const float x16 = 1.4142135623731f*x04;
289 ✗ const float x17 = x00 - x08;
290 ✗ const float x18 = x01 - x07;
291 ✗ const float x19 = x02 - x06;
292 ✗ const float x1a = x03 - x05;
293 ✗ const float x1d = x12 + x16;
294 ✗ const float x1e = x13 + x15;
295 ✗ const float x1f = 1.4142135623731f*x14;
296 ✗ const float x20 = x12 - x16;
297 ✗ const float x21 = x13 - x15;
298 ✗ const float x22 = 0.25f * (x1d - x1f);
299 ✗ const float x23 = 0.25f * (x20 + x21);
300 ✗ const float x24 = 0.25f * (x20 - x21);
301 ✗ const float x25 = 1.4142135623731f*x17;
302 ✗ const float x26 = 1.30656296487638f*x18 + 0.541196100146197f*x1a;
303 ✗ const float x27 = 1.4142135623731f*x19;
304 ✗ const float x28 = -0.541196100146197f*x18 + 1.30656296487638f*x1a;
305 ✗ const float x29 = 0.176776695296637f * (x25 + x27) + 0.25f*x26;
306 ✗ const float x2a = 0.25f * (x25 - x27);
307 ✗ const float x2b = 0.176776695296637f * (x25 + x27) - 0.25f*x26;
308 ✗ const float x2c = 0.353553390593274f*x28;
309 ✗ const float x1b = 0.707106781186547f * (x2a - x2c);
310 ✗ const float x1c = 0.707106781186547f * (x2a + x2c);
311 ✗ const float x2d = 1.4142135623731f*x0c;
312 ✗ const float x2e = x0b + x0d;
313 ✗ const float x2f = x0a + x0e;
314 ✗ const float x30 = x09 + x0f;
315 ✗ const float x31 = x09 - x0f;
316 ✗ const float x32 = x0a - x0e;
317 ✗ const float x33 = x0b - x0d;
318 ✗ const float x37 = 1.4142135623731f*x2d;
319 ✗ const float x38 = 1.30656296487638f*x2e + 0.541196100146197f*x30;
320 ✗ const float x39 = 1.4142135623731f*x2f;
321 ✗ const float x3a = -0.541196100146197f*x2e + 1.30656296487638f*x30;
322 ✗ const float x3b = 0.176776695296637f * (x37 + x39) + 0.25f*x38;
323 ✗ const float x3c = 0.25f * (x37 - x39);
324 ✗ const float x3d = 0.176776695296637f * (x37 + x39) - 0.25f*x38;
325 ✗ const float x3e = 0.353553390593274f*x3a;
326 ✗ const float x34 = 0.707106781186547f * (x3c - x3e);
327 ✗ const float x35 = 0.707106781186547f * (x3c + x3e);
328 ✗ const float x3f = 1.4142135623731f*x32;
329 ✗ const float x40 = x31 + x33;
330 ✗ const float x41 = x31 - x33;
331 ✗ const float x42 = 0.25f * (x3f + x40);
332 ✗ const float x43 = 0.25f * (x3f - x40);
333 ✗ const float x44 = 0.353553390593274f*x41;
334 ✗ dst[ 0*dst_stridea] = (add ? dst[ 0*dst_stridea] : 0) + 0.176776695296637f * (x1d + x1f) + 0.25f*x1e;
335 ✗ dst[ 1*dst_stridea] = (add ? dst[ 1*dst_stridea] : 0) + 0.707106781186547f * (x29 + x3d);
336 ✗ dst[ 2*dst_stridea] = (add ? dst[ 2*dst_stridea] : 0) + 0.707106781186547f * (x29 - x3d);
337 ✗ dst[ 3*dst_stridea] = (add ? dst[ 3*dst_stridea] : 0) + 0.707106781186547f * (x23 - x43);
338 ✗ dst[ 4*dst_stridea] = (add ? dst[ 4*dst_stridea] : 0) + 0.707106781186547f * (x23 + x43);
339 ✗ dst[ 5*dst_stridea] = (add ? dst[ 5*dst_stridea] : 0) + 0.707106781186547f * (x1b - x35);
340 ✗ dst[ 6*dst_stridea] = (add ? dst[ 6*dst_stridea] : 0) + 0.707106781186547f * (x1b + x35);
341 ✗ dst[ 7*dst_stridea] = (add ? dst[ 7*dst_stridea] : 0) + 0.707106781186547f * (x22 + x44);
342 ✗ dst[ 8*dst_stridea] = (add ? dst[ 8*dst_stridea] : 0) + 0.707106781186547f * (x22 - x44);
343 ✗ dst[ 9*dst_stridea] = (add ? dst[ 9*dst_stridea] : 0) + 0.707106781186547f * (x1c + x34);
344 ✗ dst[10*dst_stridea] = (add ? dst[10*dst_stridea] : 0) + 0.707106781186547f * (x1c - x34);
345 ✗ dst[11*dst_stridea] = (add ? dst[11*dst_stridea] : 0) + 0.707106781186547f * (x24 + x42);
346 ✗ dst[12*dst_stridea] = (add ? dst[12*dst_stridea] : 0) + 0.707106781186547f * (x24 - x42);
347 ✗ dst[13*dst_stridea] = (add ? dst[13*dst_stridea] : 0) + 0.707106781186547f * (x2b - x3b);
348 ✗ dst[14*dst_stridea] = (add ? dst[14*dst_stridea] : 0) + 0.707106781186547f * (x2b + x3b);
349 ✗ dst[15*dst_stridea] = (add ? dst[15*dst_stridea] : 0) + 0.176776695296637f * (x1d + x1f) - 0.25f*x1e;
350 ✗ dst += dst_strideb;
351 ✗ src += src_strideb;
352 }
353 ✗ }
354
355 #define DEF_FILTER_FREQ_FUNCS(bsize) \
356 static av_always_inline void filter_freq_##bsize(const float *src, int src_linesize, \
357 float *dst, int dst_linesize, \
358 AVExpr *expr, double *var_values, \
359 int sigma_th) \
360 { \
361 unsigned i; \
362 DECLARE_ALIGNED(32, float, tmp_block1)[bsize * bsize]; \
363 DECLARE_ALIGNED(32, float, tmp_block2)[bsize * bsize]; \
364 \
365 /* forward DCT */ \
366 fdct##bsize##_1d(tmp_block1, src, 1, bsize, 1, src_linesize); \
367 fdct##bsize##_1d(tmp_block2, tmp_block1, bsize, 1, bsize, 1); \
368 \
369 for (i = 0; i < bsize*bsize; i++) { \
370 float *b = &tmp_block2[i]; \
371 /* frequency filtering */ \
372 if (expr) { \
373 var_values[VAR_C] = fabsf(*b); \
374 *b *= av_expr_eval(expr, var_values, NULL); \
375 } else { \
376 if (fabsf(*b) < sigma_th) \
377 *b = 0; \
378 } \
379 } \
380 \
381 /* inverse DCT */ \
382 idct##bsize##_1d(tmp_block1, tmp_block2, 1, bsize, 1, bsize, 0); \
383 idct##bsize##_1d(dst, tmp_block1, dst_linesize, 1, bsize, 1, 1); \
384 } \
385 \
386 static void filter_freq_sigma_##bsize(DCTdnoizContext *s, \
387 const float *src, int src_linesize, \
388 float *dst, int dst_linesize, int thread_id) \
389 { \
390 filter_freq_##bsize(src, src_linesize, dst, dst_linesize, NULL, NULL, s->th); \
391 } \
392 \
393 static void filter_freq_expr_##bsize(DCTdnoizContext *s, \
394 const float *src, int src_linesize, \
395 float *dst, int dst_linesize, int thread_id) \
396 { \
397 filter_freq_##bsize(src, src_linesize, dst, dst_linesize, \
398 s->expr[thread_id], s->var_values[thread_id], 0); \
399 }
400
401 ✗ DEF_FILTER_FREQ_FUNCS(8)
402 ✗ DEF_FILTER_FREQ_FUNCS(16)
403
404 #define DCT3X3_0_0 0.5773502691896258f /* 1/sqrt(3) */
405 #define DCT3X3_0_1 0.5773502691896258f /* 1/sqrt(3) */
406 #define DCT3X3_0_2 0.5773502691896258f /* 1/sqrt(3) */
407 #define DCT3X3_1_0 0.7071067811865475f /* 1/sqrt(2) */
408 #define DCT3X3_1_2 -0.7071067811865475f /* -1/sqrt(2) */
409 #define DCT3X3_2_0 0.4082482904638631f /* 1/sqrt(6) */
410 #define DCT3X3_2_1 -0.8164965809277261f /* -2/sqrt(6) */
411 #define DCT3X3_2_2 0.4082482904638631f /* 1/sqrt(6) */
412
413 ✗ static av_always_inline void color_decorrelation(float **dst, int dst_linesize,
414 const uint8_t **src, int src_linesize,
415 int w, int h,
416 int r, int g, int b)
417 {
418 int x, y;
419 ✗ float *dstp_r = dst[0];
420 ✗ float *dstp_g = dst[1];
421 ✗ float *dstp_b = dst[2];
422 ✗ const uint8_t *srcp = src[0];
423
424 ✗ for (y = 0; y < h; y++) {
425 ✗ for (x = 0; x < w; x++) {
426 ✗ dstp_r[x] = srcp[r] * DCT3X3_0_0 + srcp[g] * DCT3X3_0_1 + srcp[b] * DCT3X3_0_2;
427 ✗ dstp_g[x] = srcp[r] * DCT3X3_1_0 + srcp[b] * DCT3X3_1_2;
428 ✗ dstp_b[x] = srcp[r] * DCT3X3_2_0 + srcp[g] * DCT3X3_2_1 + srcp[b] * DCT3X3_2_2;
429 ✗ srcp += 3;
430 }
431 ✗ srcp += src_linesize - w * 3;
432 ✗ dstp_r += dst_linesize;
433 ✗ dstp_g += dst_linesize;
434 ✗ dstp_b += dst_linesize;
435 }
436 ✗ }
437
438 ✗ static av_always_inline void color_correlation(uint8_t **dst, int dst_linesize,
439 float **src, int src_linesize,
440 int w, int h,
441 int r, int g, int b)
442 {
443 int x, y;
444 ✗ const float *src_r = src[0];
445 ✗ const float *src_g = src[1];
446 ✗ const float *src_b = src[2];
447 ✗ uint8_t *dstp = dst[0];
448
449 ✗ for (y = 0; y < h; y++) {
450 ✗ for (x = 0; x < w; x++) {
451 ✗ dstp[r] = av_clip_uint8(src_r[x] * DCT3X3_0_0 + src_g[x] * DCT3X3_1_0 + src_b[x] * DCT3X3_2_0);
452 ✗ dstp[g] = av_clip_uint8(src_r[x] * DCT3X3_0_1 + src_b[x] * DCT3X3_2_1);
453 ✗ dstp[b] = av_clip_uint8(src_r[x] * DCT3X3_0_2 + src_g[x] * DCT3X3_1_2 + src_b[x] * DCT3X3_2_2);
454 ✗ dstp += 3;
455 }
456 ✗ dstp += dst_linesize - w * 3;
457 ✗ src_r += src_linesize;
458 ✗ src_g += src_linesize;
459 ✗ src_b += src_linesize;
460 }
461 ✗ }
462
463 #define DECLARE_COLOR_FUNCS(name, r, g, b) \
464 static void color_decorrelation_##name(float **dst, int dst_linesize, \
465 const uint8_t **src, int src_linesize, \
466 int w, int h) \
467 { \
468 color_decorrelation(dst, dst_linesize, src, src_linesize, w, h, r, g, b); \
469 } \
470 \
471 static void color_correlation_##name(uint8_t **dst, int dst_linesize, \
472 float **src, int src_linesize, \
473 int w, int h) \
474 { \
475 color_correlation(dst, dst_linesize, src, src_linesize, w, h, r, g, b); \
476 }
477
478 ✗ DECLARE_COLOR_FUNCS(rgb, 0, 1, 2)
479 ✗ DECLARE_COLOR_FUNCS(bgr, 2, 1, 0)
480
481 ✗ static av_always_inline void color_decorrelation_gbrp(float **dst, int dst_linesize,
482 const uint8_t **src, int src_linesize,
483 int w, int h)
484 {
485 int x, y;
486 ✗ float *dstp_r = dst[0];
487 ✗ float *dstp_g = dst[1];
488 ✗ float *dstp_b = dst[2];
489 ✗ const uint8_t *srcp_r = src[2];
490 ✗ const uint8_t *srcp_g = src[0];
491 ✗ const uint8_t *srcp_b = src[1];
492
493 ✗ for (y = 0; y < h; y++) {
494 ✗ for (x = 0; x < w; x++) {
495 ✗ dstp_r[x] = srcp_r[x] * DCT3X3_0_0 + srcp_g[x] * DCT3X3_0_1 + srcp_b[x] * DCT3X3_0_2;
496 ✗ dstp_g[x] = srcp_r[x] * DCT3X3_1_0 + srcp_b[x] * DCT3X3_1_2;
497 ✗ dstp_b[x] = srcp_r[x] * DCT3X3_2_0 + srcp_g[x] * DCT3X3_2_1 + srcp_b[x] * DCT3X3_2_2;
498 }
499 ✗ srcp_r += src_linesize;
500 ✗ srcp_g += src_linesize;
501 ✗ srcp_b += src_linesize;
502 ✗ dstp_r += dst_linesize;
503 ✗ dstp_g += dst_linesize;
504 ✗ dstp_b += dst_linesize;
505 }
506 ✗ }
507
508 ✗ static av_always_inline void color_correlation_gbrp(uint8_t **dst, int dst_linesize,
509 float **src, int src_linesize,
510 int w, int h)
511 {
512 int x, y;
513 ✗ const float *src_r = src[0];
514 ✗ const float *src_g = src[1];
515 ✗ const float *src_b = src[2];
516 ✗ uint8_t *dstp_r = dst[2];
517 ✗ uint8_t *dstp_g = dst[0];
518 ✗ uint8_t *dstp_b = dst[1];
519
520 ✗ for (y = 0; y < h; y++) {
521 ✗ for (x = 0; x < w; x++) {
522 ✗ dstp_r[x] = av_clip_uint8(src_r[x] * DCT3X3_0_0 + src_g[x] * DCT3X3_1_0 + src_b[x] * DCT3X3_2_0);
523 ✗ dstp_g[x] = av_clip_uint8(src_r[x] * DCT3X3_0_1 + src_b[x] * DCT3X3_2_1);
524 ✗ dstp_b[x] = av_clip_uint8(src_r[x] * DCT3X3_0_2 + src_g[x] * DCT3X3_1_2 + src_b[x] * DCT3X3_2_2);
525 }
526 ✗ dstp_r += dst_linesize;
527 ✗ dstp_g += dst_linesize;
528 ✗ dstp_b += dst_linesize;
529 ✗ src_r += src_linesize;
530 ✗ src_g += src_linesize;
531 ✗ src_b += src_linesize;
532 }
533 ✗ }
534
535 ✗ static int config_input(AVFilterLink *inlink)
536 {
537 ✗ AVFilterContext *ctx = inlink->dst;
538 ✗ DCTdnoizContext *s = ctx->priv;
539 int i, x, y, bx, by, linesize, *iweights, max_slice_h, slice_h;
540 ✗ const int bsize = 1 << s->n;
541
542 ✗ switch (inlink->format) {
543 ✗ case AV_PIX_FMT_BGR24:
544 ✗ s->color_decorrelation = color_decorrelation_bgr;
545 ✗ s->color_correlation = color_correlation_bgr;
546 ✗ break;
547 ✗ case AV_PIX_FMT_RGB24:
548 ✗ s->color_decorrelation = color_decorrelation_rgb;
549 ✗ s->color_correlation = color_correlation_rgb;
550 ✗ break;
551 ✗ case AV_PIX_FMT_GBRP:
552 ✗ s->color_decorrelation = color_decorrelation_gbrp;
553 ✗ s->color_correlation = color_correlation_gbrp;
554 ✗ break;
555 ✗ default:
556 ✗ av_assert0(0);
557 }
558
559 ✗ s->pr_width = inlink->w - (inlink->w - bsize) % s->step;
560 ✗ s->pr_height = inlink->h - (inlink->h - bsize) % s->step;
561 ✗ if (s->pr_width != inlink->w)
562 ✗ av_log(ctx, AV_LOG_WARNING, "The last %d horizontal pixels won't be denoised\n",
563 ✗ inlink->w - s->pr_width);
564 ✗ if (s->pr_height != inlink->h)
565 ✗ av_log(ctx, AV_LOG_WARNING, "The last %d vertical pixels won't be denoised\n",
566 ✗ inlink->h - s->pr_height);
567
568 ✗ max_slice_h = s->pr_height / ((s->bsize - 1) * 2);
569 ✗ if (max_slice_h == 0)
570 ✗ return AVERROR(EINVAL);
571
572 ✗ s->nb_threads = FFMIN3(MAX_THREADS, ff_filter_get_nb_threads(ctx), max_slice_h);
573 ✗ av_log(ctx, AV_LOG_DEBUG, "threads: [max=%d hmax=%d user=%d] => %d\n",
574 MAX_THREADS, max_slice_h, ff_filter_get_nb_threads(ctx), s->nb_threads);
575
576 ✗ s->p_linesize = linesize = FFALIGN(s->pr_width, 32);
577 ✗ for (i = 0; i < 2; i++) {
578 ✗ s->cbuf[i][0] = av_malloc_array(linesize * s->pr_height, sizeof(*s->cbuf[i][0]));
579 ✗ s->cbuf[i][1] = av_malloc_array(linesize * s->pr_height, sizeof(*s->cbuf[i][1]));
580 ✗ s->cbuf[i][2] = av_malloc_array(linesize * s->pr_height, sizeof(*s->cbuf[i][2]));
581 ✗ if (!s->cbuf[i][0] || !s->cbuf[i][1] || !s->cbuf[i][2])
582 ✗ return AVERROR(ENOMEM);
583 }
584
585 /* eval expressions are probably not thread safe when the eval internal
586 * state can be changed (typically through load & store operations) */
587 ✗ if (s->expr_str) {
588 ✗ for (i = 0; i < s->nb_threads; i++) {
589 ✗ int ret = av_expr_parse(&s->expr[i], s->expr_str, var_names,
590 NULL, NULL, NULL, NULL, 0, ctx);
591 ✗ if (ret < 0)
592 ✗ return ret;
593 }
594 }
595
596 /* each slice will need to (pre & re)process the top and bottom block of
597 * the previous one in in addition to its processing area. This is because
598 * each pixel is averaged by all the surrounding blocks */
599 ✗ slice_h = (int)ceilf(s->pr_height / (float)s->nb_threads) + (s->bsize - 1) * 2;
600 ✗ for (i = 0; i < s->nb_threads; i++) {
601 ✗ s->slices[i] = av_malloc_array(linesize, slice_h * sizeof(*s->slices[i]));
602 ✗ if (!s->slices[i])
603 ✗ return AVERROR(ENOMEM);
604 }
605
606 ✗ s->weights = av_malloc(s->pr_height * linesize * sizeof(*s->weights));
607 ✗ if (!s->weights)
608 ✗ return AVERROR(ENOMEM);
609 ✗ iweights = av_calloc(s->pr_height, linesize * sizeof(*iweights));
610 ✗ if (!iweights)
611 ✗ return AVERROR(ENOMEM);
612 ✗ for (y = 0; y < s->pr_height - bsize + 1; y += s->step)
613 ✗ for (x = 0; x < s->pr_width - bsize + 1; x += s->step)
614 ✗ for (by = 0; by < bsize; by++)
615 ✗ for (bx = 0; bx < bsize; bx++)
616 ✗ iweights[(y + by)*linesize + x + bx]++;
617 ✗ for (y = 0; y < s->pr_height; y++)
618 ✗ for (x = 0; x < s->pr_width; x++)
619 ✗ s->weights[y*linesize + x] = 1. / iweights[y*linesize + x];
620 ✗ av_free(iweights);
621
622 ✗ return 0;
623 }
624
625 ✗ static av_cold int init(AVFilterContext *ctx)
626 {
627 ✗ DCTdnoizContext *s = ctx->priv;
628
629 ✗ s->bsize = 1 << s->n;
630 ✗ if (s->overlap == -1)
631 ✗ s->overlap = s->bsize - 1;
632
633 ✗ if (s->overlap > s->bsize - 1) {
634 ✗ av_log(s, AV_LOG_ERROR, "Overlap value can not except %d "
635 "with a block size of %dx%d\n",
636 ✗ s->bsize - 1, s->bsize, s->bsize);
637 ✗ return AVERROR(EINVAL);
638 }
639
640 ✗ if (s->expr_str) {
641 ✗ switch (s->n) {
642 ✗ case 3: s->filter_freq_func = filter_freq_expr_8; break;
643 ✗ case 4: s->filter_freq_func = filter_freq_expr_16; break;
644 ✗ default: av_assert0(0);
645 }
646 } else {
647 ✗ switch (s->n) {
648 ✗ case 3: s->filter_freq_func = filter_freq_sigma_8; break;
649 ✗ case 4: s->filter_freq_func = filter_freq_sigma_16; break;
650 ✗ default: av_assert0(0);
651 }
652 }
653
654 ✗ s->th = s->sigma * 3.;
655 ✗ s->step = s->bsize - s->overlap;
656 ✗ return 0;
657 }
658
659 static const enum AVPixelFormat pix_fmts[] = {
660 AV_PIX_FMT_BGR24, AV_PIX_FMT_RGB24,
661 AV_PIX_FMT_GBRP,
662 AV_PIX_FMT_NONE
663 };
664
665 typedef struct ThreadData {
666 float *src, *dst;
667 } ThreadData;
668
669 ✗ static int filter_slice(AVFilterContext *ctx,
670 void *arg, int jobnr, int nb_jobs)
671 {
672 int x, y;
673 ✗ DCTdnoizContext *s = ctx->priv;
674 ✗ const ThreadData *td = arg;
675 ✗ const int w = s->pr_width;
676 ✗ const int h = s->pr_height;
677 ✗ const int slice_start = (h * jobnr ) / nb_jobs;
678 ✗ const int slice_end = (h * (jobnr+1)) / nb_jobs;
679 ✗ const int slice_start_ctx = FFMAX(slice_start - s->bsize + 1, 0);
680 ✗ const int slice_end_ctx = FFMIN(slice_end, h - s->bsize + 1);
681 ✗ const int slice_h = slice_end_ctx - slice_start_ctx;
682 ✗ const int src_linesize = s->p_linesize;
683 ✗ const int dst_linesize = s->p_linesize;
684 ✗ const int slice_linesize = s->p_linesize;
685 float *dst;
686 ✗ const float *src = td->src + slice_start_ctx * src_linesize;
687 ✗ const float *weights = s->weights + slice_start * dst_linesize;
688 ✗ float *slice = s->slices[jobnr];
689
690 // reset block sums
691 ✗ memset(slice, 0, (slice_h + s->bsize - 1) * dst_linesize * sizeof(*slice));
692
693 // block dct sums
694 ✗ for (y = 0; y < slice_h; y += s->step) {
695 ✗ for (x = 0; x < w - s->bsize + 1; x += s->step)
696 ✗ s->filter_freq_func(s, src + x, src_linesize,
697 ✗ slice + x, slice_linesize,
698 jobnr);
699 ✗ src += s->step * src_linesize;
700 ✗ slice += s->step * slice_linesize;
701 }
702
703 // average blocks
704 ✗ slice = s->slices[jobnr] + (slice_start - slice_start_ctx) * slice_linesize;
705 ✗ dst = td->dst + slice_start * dst_linesize;
706 ✗ for (y = slice_start; y < slice_end; y++) {
707 ✗ for (x = 0; x < w; x++)
708 ✗ dst[x] = slice[x] * weights[x];
709 ✗ slice += slice_linesize;
710 ✗ dst += dst_linesize;
711 ✗ weights += dst_linesize;
712 }
713
714 ✗ return 0;
715 }
716
717 ✗ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
718 {
719 ✗ AVFilterContext *ctx = inlink->dst;
720 ✗ DCTdnoizContext *s = ctx->priv;
721 ✗ AVFilterLink *outlink = inlink->dst->outputs[0];
722 int direct, plane;
723 AVFrame *out;
724
725 ✗ if (av_frame_is_writable(in)) {
726 ✗ direct = 1;
727 ✗ out = in;
728 } else {
729 ✗ direct = 0;
730 ✗ out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
731 ✗ if (!out) {
732 ✗ av_frame_free(&in);
733 ✗ return AVERROR(ENOMEM);
734 }
735 ✗ av_frame_copy_props(out, in);
736 }
737
738 ✗ s->color_decorrelation(s->cbuf[0], s->p_linesize,
739 ✗ (const uint8_t **)in->data, in->linesize[0],
740 s->pr_width, s->pr_height);
741 ✗ for (plane = 0; plane < 3; plane++) {
742 ✗ ThreadData td = {
743 ✗ .src = s->cbuf[0][plane],
744 ✗ .dst = s->cbuf[1][plane],
745 };
746 ✗ ff_filter_execute(ctx, filter_slice, &td, NULL, s->nb_threads);
747 }
748 ✗ s->color_correlation(out->data, out->linesize[0],
749 ✗ s->cbuf[1], s->p_linesize,
750 s->pr_width, s->pr_height);
751
752 ✗ if (!direct) {
753 int y;
754 ✗ uint8_t *dst = out->data[0];
755 ✗ const uint8_t *src = in->data[0];
756 ✗ const int dst_linesize = out->linesize[0];
757 ✗ const int src_linesize = in->linesize[0];
758 ✗ const int hpad = (inlink->w - s->pr_width) * 3;
759 ✗ const int vpad = (inlink->h - s->pr_height);
760
761 ✗ if (hpad) {
762 ✗ uint8_t *dstp = dst + s->pr_width * 3;
763 ✗ const uint8_t *srcp = src + s->pr_width * 3;
764
765 ✗ for (y = 0; y < s->pr_height; y++) {
766 ✗ memcpy(dstp, srcp, hpad);
767 ✗ dstp += dst_linesize;
768 ✗ srcp += src_linesize;
769 }
770 }
771 ✗ if (vpad) {
772 ✗ uint8_t *dstp = dst + s->pr_height * dst_linesize;
773 ✗ const uint8_t *srcp = src + s->pr_height * src_linesize;
774
775 ✗ for (y = 0; y < vpad; y++) {
776 ✗ memcpy(dstp, srcp, inlink->w * 3);
777 ✗ dstp += dst_linesize;
778 ✗ srcp += src_linesize;
779 }
780 }
781
782 ✗ av_frame_free(&in);
783 }
784
785 ✗ return ff_filter_frame(outlink, out);
786 }
787
788 ✗ static av_cold void uninit(AVFilterContext *ctx)
789 {
790 int i;
791 ✗ DCTdnoizContext *s = ctx->priv;
792
793 ✗ av_freep(&s->weights);
794 ✗ for (i = 0; i < 2; i++) {
795 ✗ av_freep(&s->cbuf[i][0]);
796 ✗ av_freep(&s->cbuf[i][1]);
797 ✗ av_freep(&s->cbuf[i][2]);
798 }
799 ✗ for (i = 0; i < s->nb_threads; i++) {
800 ✗ av_freep(&s->slices[i]);
801 ✗ av_expr_free(s->expr[i]);
802 }
803 ✗ }
804
805 static const AVFilterPad dctdnoiz_inputs[] = {
806 {
807 .name = "default",
808 .type = AVMEDIA_TYPE_VIDEO,
809 .filter_frame = filter_frame,
810 .config_props = config_input,
811 },
812 };
813
814 const AVFilter ff_vf_dctdnoiz = {
815 .name = "dctdnoiz",
816 .description = NULL_IF_CONFIG_SMALL("Denoise frames using 2D DCT."),
817 .priv_size = sizeof(DCTdnoizContext),
818 .init = init,
819 .uninit = uninit,
820 FILTER_INPUTS(dctdnoiz_inputs),
821 FILTER_OUTPUTS(ff_video_default_filterpad),
822 FILTER_PIXFMTS_ARRAY(pix_fmts),
823 .priv_class = &dctdnoiz_class,
824 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
825 };
826