?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
avcodec/jpegxl_parse{,r}: fix integer overflow for some malformed files
[FFMpeg-mirror.git] / libavfilter / vf_dctdnoiz.c
blob50cd18f7ce8fec382dcf6e304d815efc26da1c26
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
37 #include "filters.h"
38 #include "video.h"
39
40 static const char *const var_names[] = { "c", NULL };
41 enum { VAR_C, VAR_VARS_NB };
42
43 #define MAX_THREADS 8
44
45 typedef struct DCTdnoizContext {
46 const AVClass *class;
47
48 /* coefficient factor expression */
49 char *expr_str;
50 AVExpr *expr[MAX_THREADS];
51 double var_values[MAX_THREADS][VAR_VARS_NB];
52
53 int nb_threads;
54 int pr_width, pr_height; // width and height to process
55 float sigma; // used when no expression are st
56 float th; // threshold (3*sigma)
57 float *cbuf[2][3]; // two planar rgb color buffers
58 float *slices[MAX_THREADS]; // slices buffers (1 slice buffer per thread)
59 float *weights; // dct coeff are cumulated with overlapping; these values are used for averaging
60 int p_linesize; // line sizes for color and weights
61 int overlap; // number of block overlapping pixels
62 int step; // block step increment (blocksize - overlap)
63 int n; // 1<<n is the block size
64 int bsize; // block size, 1<<n
65 void (*filter_freq_func)(struct DCTdnoizContext *s,
66 const float *src, int src_linesize,
67 float *dst, int dst_linesize,
68 int thread_id);
69 void (*color_decorrelation)(float **dst, int dst_linesize,
70 const uint8_t **src, int src_linesize,
71 int w, int h);
72 void (*color_correlation)(uint8_t **dst, int dst_linesize,
73 float **src, int src_linesize,
74 int w, int h);
75 } DCTdnoizContext;
76
77 #define MIN_NBITS 3 /* blocksize = 1<<3 = 8 */
78 #define MAX_NBITS 4 /* blocksize = 1<<4 = 16 */
79 #define DEFAULT_NBITS 3
80
81 #define OFFSET(x) offsetof(DCTdnoizContext, x)
82 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
83 static const AVOption dctdnoiz_options[] = {
84 { "sigma", "set noise sigma constant", OFFSET(sigma), AV_OPT_TYPE_FLOAT, {.dbl=0}, 0, 999, .flags = FLAGS },
85 { "s", "set noise sigma constant", OFFSET(sigma), AV_OPT_TYPE_FLOAT, {.dbl=0}, 0, 999, .flags = FLAGS },
86 { "overlap", "set number of block overlapping pixels", OFFSET(overlap), AV_OPT_TYPE_INT, {.i64=-1}, -1, (1<<MAX_NBITS)-1, .flags = FLAGS },
87 { "expr", "set coefficient factor expression", OFFSET(expr_str), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
88 { "e", "set coefficient factor expression", OFFSET(expr_str), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
89 { "n", "set the block size, expressed in bits", OFFSET(n), AV_OPT_TYPE_INT, {.i64=DEFAULT_NBITS}, MIN_NBITS, MAX_NBITS, .flags = FLAGS },
90 { NULL }
91 };
92
93 AVFILTER_DEFINE_CLASS(dctdnoiz);
94
95 static void av_always_inline fdct8_1d(float *dst, const float *src,
96 int dst_stridea, int dst_strideb,
97 int src_stridea, int src_strideb)
98 {
99 int i;
100
101 for (i = 0; i < 8; i++) {
102 const float x00 = src[0*src_stridea] + src[7*src_stridea];
103 const float x01 = src[1*src_stridea] + src[6*src_stridea];
104 const float x02 = src[2*src_stridea] + src[5*src_stridea];
105 const float x03 = src[3*src_stridea] + src[4*src_stridea];
106 const float x04 = src[0*src_stridea] - src[7*src_stridea];
107 const float x05 = src[1*src_stridea] - src[6*src_stridea];
108 const float x06 = src[2*src_stridea] - src[5*src_stridea];
109 const float x07 = src[3*src_stridea] - src[4*src_stridea];
110 const float x08 = x00 + x03;
111 const float x09 = x01 + x02;
112 const float x0a = x00 - x03;
113 const float x0b = x01 - x02;
114 const float x0c = 1.38703984532215f*x04 + 0.275899379282943f*x07;
115 const float x0d = 1.17587560241936f*x05 + 0.785694958387102f*x06;
116 const float x0e = -0.785694958387102f*x05 + 1.17587560241936f*x06;
117 const float x0f = 0.275899379282943f*x04 - 1.38703984532215f*x07;
118 const float x10 = 0.353553390593274f * (x0c - x0d);
119 const float x11 = 0.353553390593274f * (x0e - x0f);
120 dst[0*dst_stridea] = 0.353553390593274f * (x08 + x09);
121 dst[1*dst_stridea] = 0.353553390593274f * (x0c + x0d);
122 dst[2*dst_stridea] = 0.461939766255643f*x0a + 0.191341716182545f*x0b;
123 dst[3*dst_stridea] = 0.707106781186547f * (x10 - x11);
124 dst[4*dst_stridea] = 0.353553390593274f * (x08 - x09);
125 dst[5*dst_stridea] = 0.707106781186547f * (x10 + x11);
126 dst[6*dst_stridea] = 0.191341716182545f*x0a - 0.461939766255643f*x0b;
127 dst[7*dst_stridea] = 0.353553390593274f * (x0e + x0f);
128 dst += dst_strideb;
129 src += src_strideb;
130 }
131 }
132
133 static void av_always_inline idct8_1d(float *dst, const float *src,
134 int dst_stridea, int dst_strideb,
135 int src_stridea, int src_strideb,
136 int add)
137 {
138 int i;
139
140 for (i = 0; i < 8; i++) {
141 const float x00 = 1.4142135623731f *src[0*src_stridea];
142 const float x01 = 1.38703984532215f *src[1*src_stridea] + 0.275899379282943f*src[7*src_stridea];
143 const float x02 = 1.30656296487638f *src[2*src_stridea] + 0.541196100146197f*src[6*src_stridea];
144 const float x03 = 1.17587560241936f *src[3*src_stridea] + 0.785694958387102f*src[5*src_stridea];
145 const float x04 = 1.4142135623731f *src[4*src_stridea];
146 const float x05 = -0.785694958387102f*src[3*src_stridea] + 1.17587560241936f*src[5*src_stridea];
147 const float x06 = 0.541196100146197f*src[2*src_stridea] - 1.30656296487638f*src[6*src_stridea];
148 const float x07 = -0.275899379282943f*src[1*src_stridea] + 1.38703984532215f*src[7*src_stridea];
149 const float x09 = x00 + x04;
150 const float x0a = x01 + x03;
151 const float x0b = 1.4142135623731f*x02;
152 const float x0c = x00 - x04;
153 const float x0d = x01 - x03;
154 const float x0e = 0.353553390593274f * (x09 - x0b);
155 const float x0f = 0.353553390593274f * (x0c + x0d);
156 const float x10 = 0.353553390593274f * (x0c - x0d);
157 const float x11 = 1.4142135623731f*x06;
158 const float x12 = x05 + x07;
159 const float x13 = x05 - x07;
160 const float x14 = 0.353553390593274f * (x11 + x12);
161 const float x15 = 0.353553390593274f * (x11 - x12);
162 const float x16 = 0.5f*x13;
163 dst[0*dst_stridea] = (add ? dst[ 0*dst_stridea] : 0) + 0.25f * (x09 + x0b) + 0.353553390593274f*x0a;
164 dst[1*dst_stridea] = (add ? dst[ 1*dst_stridea] : 0) + 0.707106781186547f * (x0f + x15);
165 dst[2*dst_stridea] = (add ? dst[ 2*dst_stridea] : 0) + 0.707106781186547f * (x0f - x15);
166 dst[3*dst_stridea] = (add ? dst[ 3*dst_stridea] : 0) + 0.707106781186547f * (x0e + x16);
167 dst[4*dst_stridea] = (add ? dst[ 4*dst_stridea] : 0) + 0.707106781186547f * (x0e - x16);
168 dst[5*dst_stridea] = (add ? dst[ 5*dst_stridea] : 0) + 0.707106781186547f * (x10 - x14);
169 dst[6*dst_stridea] = (add ? dst[ 6*dst_stridea] : 0) + 0.707106781186547f * (x10 + x14);
170 dst[7*dst_stridea] = (add ? dst[ 7*dst_stridea] : 0) + 0.25f * (x09 + x0b) - 0.353553390593274f*x0a;
171 dst += dst_strideb;
172 src += src_strideb;
173 }
174 }
175
176
177 static void av_always_inline fdct16_1d(float *dst, const float *src,
178 int dst_stridea, int dst_strideb,
179 int src_stridea, int src_strideb)
180 {
181 int i;
182
183 for (i = 0; i < 16; i++) {
184 const float x00 = src[ 0*src_stridea] + src[15*src_stridea];
185 const float x01 = src[ 1*src_stridea] + src[14*src_stridea];
186 const float x02 = src[ 2*src_stridea] + src[13*src_stridea];
187 const float x03 = src[ 3*src_stridea] + src[12*src_stridea];
188 const float x04 = src[ 4*src_stridea] + src[11*src_stridea];
189 const float x05 = src[ 5*src_stridea] + src[10*src_stridea];
190 const float x06 = src[ 6*src_stridea] + src[ 9*src_stridea];
191 const float x07 = src[ 7*src_stridea] + src[ 8*src_stridea];
192 const float x08 = src[ 0*src_stridea] - src[15*src_stridea];
193 const float x09 = src[ 1*src_stridea] - src[14*src_stridea];
194 const float x0a = src[ 2*src_stridea] - src[13*src_stridea];
195 const float x0b = src[ 3*src_stridea] - src[12*src_stridea];
196 const float x0c = src[ 4*src_stridea] - src[11*src_stridea];
197 const float x0d = src[ 5*src_stridea] - src[10*src_stridea];
198 const float x0e = src[ 6*src_stridea] - src[ 9*src_stridea];
199 const float x0f = src[ 7*src_stridea] - src[ 8*src_stridea];
200 const float x10 = x00 + x07;
201 const float x11 = x01 + x06;
202 const float x12 = x02 + x05;
203 const float x13 = x03 + x04;
204 const float x14 = x00 - x07;
205 const float x15 = x01 - x06;
206 const float x16 = x02 - x05;
207 const float x17 = x03 - x04;
208 const float x18 = x10 + x13;
209 const float x19 = x11 + x12;
210 const float x1a = x10 - x13;
211 const float x1b = x11 - x12;
212 const float x1c = 1.38703984532215f*x14 + 0.275899379282943f*x17;
213 const float x1d = 1.17587560241936f*x15 + 0.785694958387102f*x16;
214 const float x1e = -0.785694958387102f*x15 + 1.17587560241936f *x16;
215 const float x1f = 0.275899379282943f*x14 - 1.38703984532215f *x17;
216 const float x20 = 0.25f * (x1c - x1d);
217 const float x21 = 0.25f * (x1e - x1f);
218 const float x22 = 1.40740373752638f *x08 + 0.138617169199091f*x0f;
219 const float x23 = 1.35331800117435f *x09 + 0.410524527522357f*x0e;
220 const float x24 = 1.24722501298667f *x0a + 0.666655658477747f*x0d;
221 const float x25 = 1.09320186700176f *x0b + 0.897167586342636f*x0c;
222 const float x26 = -0.897167586342636f*x0b + 1.09320186700176f *x0c;
223 const float x27 = 0.666655658477747f*x0a - 1.24722501298667f *x0d;
224 const float x28 = -0.410524527522357f*x09 + 1.35331800117435f *x0e;
225 const float x29 = 0.138617169199091f*x08 - 1.40740373752638f *x0f;
226 const float x2a = x22 + x25;
227 const float x2b = x23 + x24;
228 const float x2c = x22 - x25;
229 const float x2d = x23 - x24;
230 const float x2e = 0.25f * (x2a - x2b);
231 const float x2f = 0.326640741219094f*x2c + 0.135299025036549f*x2d;
232 const float x30 = 0.135299025036549f*x2c - 0.326640741219094f*x2d;
233 const float x31 = x26 + x29;
234 const float x32 = x27 + x28;
235 const float x33 = x26 - x29;
236 const float x34 = x27 - x28;
237 const float x35 = 0.25f * (x31 - x32);
238 const float x36 = 0.326640741219094f*x33 + 0.135299025036549f*x34;
239 const float x37 = 0.135299025036549f*x33 - 0.326640741219094f*x34;
240 dst[ 0*dst_stridea] = 0.25f * (x18 + x19);
241 dst[ 1*dst_stridea] = 0.25f * (x2a + x2b);
242 dst[ 2*dst_stridea] = 0.25f * (x1c + x1d);
243 dst[ 3*dst_stridea] = 0.707106781186547f * (x2f - x37);
244 dst[ 4*dst_stridea] = 0.326640741219094f*x1a + 0.135299025036549f*x1b;
245 dst[ 5*dst_stridea] = 0.707106781186547f * (x2f + x37);
246 dst[ 6*dst_stridea] = 0.707106781186547f * (x20 - x21);
247 dst[ 7*dst_stridea] = 0.707106781186547f * (x2e + x35);
248 dst[ 8*dst_stridea] = 0.25f * (x18 - x19);
249 dst[ 9*dst_stridea] = 0.707106781186547f * (x2e - x35);
250 dst[10*dst_stridea] = 0.707106781186547f * (x20 + x21);
251 dst[11*dst_stridea] = 0.707106781186547f * (x30 - x36);
252 dst[12*dst_stridea] = 0.135299025036549f*x1a - 0.326640741219094f*x1b;
253 dst[13*dst_stridea] = 0.707106781186547f * (x30 + x36);
254 dst[14*dst_stridea] = 0.25f * (x1e + x1f);
255 dst[15*dst_stridea] = 0.25f * (x31 + x32);
256 dst += dst_strideb;
257 src += src_strideb;
258 }
259 }
260
261 static void av_always_inline idct16_1d(float *dst, const float *src,
262 int dst_stridea, int dst_strideb,
263 int src_stridea, int src_strideb,
264 int add)
265 {
266 int i;
267
268 for (i = 0; i < 16; i++) {
269 const float x00 = 1.4142135623731f *src[ 0*src_stridea];
270 const float x01 = 1.40740373752638f *src[ 1*src_stridea] + 0.138617169199091f*src[15*src_stridea];
271 const float x02 = 1.38703984532215f *src[ 2*src_stridea] + 0.275899379282943f*src[14*src_stridea];
272 const float x03 = 1.35331800117435f *src[ 3*src_stridea] + 0.410524527522357f*src[13*src_stridea];
273 const float x04 = 1.30656296487638f *src[ 4*src_stridea] + 0.541196100146197f*src[12*src_stridea];
274 const float x05 = 1.24722501298667f *src[ 5*src_stridea] + 0.666655658477747f*src[11*src_stridea];
275 const float x06 = 1.17587560241936f *src[ 6*src_stridea] + 0.785694958387102f*src[10*src_stridea];
276 const float x07 = 1.09320186700176f *src[ 7*src_stridea] + 0.897167586342636f*src[ 9*src_stridea];
277 const float x08 = 1.4142135623731f *src[ 8*src_stridea];
278 const float x09 = -0.897167586342636f*src[ 7*src_stridea] + 1.09320186700176f*src[ 9*src_stridea];
279 const float x0a = 0.785694958387102f*src[ 6*src_stridea] - 1.17587560241936f*src[10*src_stridea];
280 const float x0b = -0.666655658477747f*src[ 5*src_stridea] + 1.24722501298667f*src[11*src_stridea];
281 const float x0c = 0.541196100146197f*src[ 4*src_stridea] - 1.30656296487638f*src[12*src_stridea];
282 const float x0d = -0.410524527522357f*src[ 3*src_stridea] + 1.35331800117435f*src[13*src_stridea];
283 const float x0e = 0.275899379282943f*src[ 2*src_stridea] - 1.38703984532215f*src[14*src_stridea];
284 const float x0f = -0.138617169199091f*src[ 1*src_stridea] + 1.40740373752638f*src[15*src_stridea];
285 const float x12 = x00 + x08;
286 const float x13 = x01 + x07;
287 const float x14 = x02 + x06;
288 const float x15 = x03 + x05;
289 const float x16 = 1.4142135623731f*x04;
290 const float x17 = x00 - x08;
291 const float x18 = x01 - x07;
292 const float x19 = x02 - x06;
293 const float x1a = x03 - x05;
294 const float x1d = x12 + x16;
295 const float x1e = x13 + x15;
296 const float x1f = 1.4142135623731f*x14;
297 const float x20 = x12 - x16;
298 const float x21 = x13 - x15;
299 const float x22 = 0.25f * (x1d - x1f);
300 const float x23 = 0.25f * (x20 + x21);
301 const float x24 = 0.25f * (x20 - x21);
302 const float x25 = 1.4142135623731f*x17;
303 const float x26 = 1.30656296487638f*x18 + 0.541196100146197f*x1a;
304 const float x27 = 1.4142135623731f*x19;
305 const float x28 = -0.541196100146197f*x18 + 1.30656296487638f*x1a;
306 const float x29 = 0.176776695296637f * (x25 + x27) + 0.25f*x26;
307 const float x2a = 0.25f * (x25 - x27);
308 const float x2b = 0.176776695296637f * (x25 + x27) - 0.25f*x26;
309 const float x2c = 0.353553390593274f*x28;
310 const float x1b = 0.707106781186547f * (x2a - x2c);
311 const float x1c = 0.707106781186547f * (x2a + x2c);
312 const float x2d = 1.4142135623731f*x0c;
313 const float x2e = x0b + x0d;
314 const float x2f = x0a + x0e;
315 const float x30 = x09 + x0f;
316 const float x31 = x09 - x0f;
317 const float x32 = x0a - x0e;
318 const float x33 = x0b - x0d;
319 const float x37 = 1.4142135623731f*x2d;
320 const float x38 = 1.30656296487638f*x2e + 0.541196100146197f*x30;
321 const float x39 = 1.4142135623731f*x2f;
322 const float x3a = -0.541196100146197f*x2e + 1.30656296487638f*x30;
323 const float x3b = 0.176776695296637f * (x37 + x39) + 0.25f*x38;
324 const float x3c = 0.25f * (x37 - x39);
325 const float x3d = 0.176776695296637f * (x37 + x39) - 0.25f*x38;
326 const float x3e = 0.353553390593274f*x3a;
327 const float x34 = 0.707106781186547f * (x3c - x3e);
328 const float x35 = 0.707106781186547f * (x3c + x3e);
329 const float x3f = 1.4142135623731f*x32;
330 const float x40 = x31 + x33;
331 const float x41 = x31 - x33;
332 const float x42 = 0.25f * (x3f + x40);
333 const float x43 = 0.25f * (x3f - x40);
334 const float x44 = 0.353553390593274f*x41;
335 dst[ 0*dst_stridea] = (add ? dst[ 0*dst_stridea] : 0) + 0.176776695296637f * (x1d + x1f) + 0.25f*x1e;
336 dst[ 1*dst_stridea] = (add ? dst[ 1*dst_stridea] : 0) + 0.707106781186547f * (x29 + x3d);
337 dst[ 2*dst_stridea] = (add ? dst[ 2*dst_stridea] : 0) + 0.707106781186547f * (x29 - x3d);
338 dst[ 3*dst_stridea] = (add ? dst[ 3*dst_stridea] : 0) + 0.707106781186547f * (x23 - x43);
339 dst[ 4*dst_stridea] = (add ? dst[ 4*dst_stridea] : 0) + 0.707106781186547f * (x23 + x43);
340 dst[ 5*dst_stridea] = (add ? dst[ 5*dst_stridea] : 0) + 0.707106781186547f * (x1b - x35);
341 dst[ 6*dst_stridea] = (add ? dst[ 6*dst_stridea] : 0) + 0.707106781186547f * (x1b + x35);
342 dst[ 7*dst_stridea] = (add ? dst[ 7*dst_stridea] : 0) + 0.707106781186547f * (x22 + x44);
343 dst[ 8*dst_stridea] = (add ? dst[ 8*dst_stridea] : 0) + 0.707106781186547f * (x22 - x44);
344 dst[ 9*dst_stridea] = (add ? dst[ 9*dst_stridea] : 0) + 0.707106781186547f * (x1c + x34);
345 dst[10*dst_stridea] = (add ? dst[10*dst_stridea] : 0) + 0.707106781186547f * (x1c - x34);
346 dst[11*dst_stridea] = (add ? dst[11*dst_stridea] : 0) + 0.707106781186547f * (x24 + x42);
347 dst[12*dst_stridea] = (add ? dst[12*dst_stridea] : 0) + 0.707106781186547f * (x24 - x42);
348 dst[13*dst_stridea] = (add ? dst[13*dst_stridea] : 0) + 0.707106781186547f * (x2b - x3b);
349 dst[14*dst_stridea] = (add ? dst[14*dst_stridea] : 0) + 0.707106781186547f * (x2b + x3b);
350 dst[15*dst_stridea] = (add ? dst[15*dst_stridea] : 0) + 0.176776695296637f * (x1d + x1f) - 0.25f*x1e;
351 dst += dst_strideb;
352 src += src_strideb;
353 }
354 }
355
356 #define DEF_FILTER_FREQ_FUNCS(bsize) \
357 static av_always_inline void filter_freq_##bsize(const float *src, int src_linesize, \
358 float *dst, int dst_linesize, \
359 AVExpr *expr, double *var_values, \
360 int sigma_th) \
361 { \
362 unsigned i; \
363 DECLARE_ALIGNED(32, float, tmp_block1)[bsize * bsize]; \
364 DECLARE_ALIGNED(32, float, tmp_block2)[bsize * bsize]; \
365 \
366 /* forward DCT */ \
367 fdct##bsize##_1d(tmp_block1, src, 1, bsize, 1, src_linesize); \
368 fdct##bsize##_1d(tmp_block2, tmp_block1, bsize, 1, bsize, 1); \
369 \
370 for (i = 0; i < bsize*bsize; i++) { \
371 float *b = &tmp_block2[i]; \
372 /* frequency filtering */ \
373 if (expr) { \
374 var_values[VAR_C] = fabsf(*b); \
375 *b *= av_expr_eval(expr, var_values, NULL); \
376 } else { \
377 if (fabsf(*b) < sigma_th) \
378 *b = 0; \
379 } \
380 } \
381 \
382 /* inverse DCT */ \
383 idct##bsize##_1d(tmp_block1, tmp_block2, 1, bsize, 1, bsize, 0); \
384 idct##bsize##_1d(dst, tmp_block1, dst_linesize, 1, bsize, 1, 1); \
385 } \
386 \
387 static void filter_freq_sigma_##bsize(DCTdnoizContext *s, \
388 const float *src, int src_linesize, \
389 float *dst, int dst_linesize, int thread_id) \
390 { \
391 filter_freq_##bsize(src, src_linesize, dst, dst_linesize, NULL, NULL, s->th); \
392 } \
393 \
394 static void filter_freq_expr_##bsize(DCTdnoizContext *s, \
395 const float *src, int src_linesize, \
396 float *dst, int dst_linesize, int thread_id) \
397 { \
398 filter_freq_##bsize(src, src_linesize, dst, dst_linesize, \
399 s->expr[thread_id], s->var_values[thread_id], 0); \
400 }
401
402 DEF_FILTER_FREQ_FUNCS(8)
403 DEF_FILTER_FREQ_FUNCS(16)
404
405 #define DCT3X3_0_0 0.5773502691896258f /* 1/sqrt(3) */
406 #define DCT3X3_0_1 0.5773502691896258f /* 1/sqrt(3) */
407 #define DCT3X3_0_2 0.5773502691896258f /* 1/sqrt(3) */
408 #define DCT3X3_1_0 0.7071067811865475f /* 1/sqrt(2) */
409 #define DCT3X3_1_2 -0.7071067811865475f /* -1/sqrt(2) */
410 #define DCT3X3_2_0 0.4082482904638631f /* 1/sqrt(6) */
411 #define DCT3X3_2_1 -0.8164965809277261f /* -2/sqrt(6) */
412 #define DCT3X3_2_2 0.4082482904638631f /* 1/sqrt(6) */
413
414 static av_always_inline void color_decorrelation(float **dst, int dst_linesize,
415 const uint8_t **src, int src_linesize,
416 int w, int h,
417 int r, int g, int b)
418 {
419 int x, y;
420 float *dstp_r = dst[0];
421 float *dstp_g = dst[1];
422 float *dstp_b = dst[2];
423 const uint8_t *srcp = src[0];
424
425 for (y = 0; y < h; y++) {
426 for (x = 0; x < w; x++) {
427 dstp_r[x] = srcp[r] * DCT3X3_0_0 + srcp[g] * DCT3X3_0_1 + srcp[b] * DCT3X3_0_2;
428 dstp_g[x] = srcp[r] * DCT3X3_1_0 + srcp[b] * DCT3X3_1_2;
429 dstp_b[x] = srcp[r] * DCT3X3_2_0 + srcp[g] * DCT3X3_2_1 + srcp[b] * DCT3X3_2_2;
430 srcp += 3;
431 }
432 srcp += src_linesize - w * 3;
433 dstp_r += dst_linesize;
434 dstp_g += dst_linesize;
435 dstp_b += dst_linesize;
436 }
437 }
438
439 static av_always_inline void color_correlation(uint8_t **dst, int dst_linesize,
440 float **src, int src_linesize,
441 int w, int h,
442 int r, int g, int b)
443 {
444 int x, y;
445 const float *src_r = src[0];
446 const float *src_g = src[1];
447 const float *src_b = src[2];
448 uint8_t *dstp = dst[0];
449
450 for (y = 0; y < h; y++) {
451 for (x = 0; x < w; x++) {
452 dstp[r] = av_clip_uint8(src_r[x] * DCT3X3_0_0 + src_g[x] * DCT3X3_1_0 + src_b[x] * DCT3X3_2_0);
453 dstp[g] = av_clip_uint8(src_r[x] * DCT3X3_0_1 + src_b[x] * DCT3X3_2_1);
454 dstp[b] = av_clip_uint8(src_r[x] * DCT3X3_0_2 + src_g[x] * DCT3X3_1_2 + src_b[x] * DCT3X3_2_2);
455 dstp += 3;
456 }
457 dstp += dst_linesize - w * 3;
458 src_r += src_linesize;
459 src_g += src_linesize;
460 src_b += src_linesize;
461 }
462 }
463
464 #define DECLARE_COLOR_FUNCS(name, r, g, b) \
465 static void color_decorrelation_##name(float **dst, int dst_linesize, \
466 const uint8_t **src, int src_linesize, \
467 int w, int h) \
468 { \
469 color_decorrelation(dst, dst_linesize, src, src_linesize, w, h, r, g, b); \
470 } \
471 \
472 static void color_correlation_##name(uint8_t **dst, int dst_linesize, \
473 float **src, int src_linesize, \
474 int w, int h) \
475 { \
476 color_correlation(dst, dst_linesize, src, src_linesize, w, h, r, g, b); \
477 }
478
479 DECLARE_COLOR_FUNCS(rgb, 0, 1, 2)
480 DECLARE_COLOR_FUNCS(bgr, 2, 1, 0)
481
482 static av_always_inline void color_decorrelation_gbrp(float **dst, int dst_linesize,
483 const uint8_t **src, int src_linesize,
484 int w, int h)
485 {
486 int x, y;
487 float *dstp_r = dst[0];
488 float *dstp_g = dst[1];
489 float *dstp_b = dst[2];
490 const uint8_t *srcp_r = src[2];
491 const uint8_t *srcp_g = src[0];
492 const uint8_t *srcp_b = src[1];
493
494 for (y = 0; y < h; y++) {
495 for (x = 0; x < w; x++) {
496 dstp_r[x] = srcp_r[x] * DCT3X3_0_0 + srcp_g[x] * DCT3X3_0_1 + srcp_b[x] * DCT3X3_0_2;
497 dstp_g[x] = srcp_r[x] * DCT3X3_1_0 + srcp_b[x] * DCT3X3_1_2;
498 dstp_b[x] = srcp_r[x] * DCT3X3_2_0 + srcp_g[x] * DCT3X3_2_1 + srcp_b[x] * DCT3X3_2_2;
499 }
500 srcp_r += src_linesize;
501 srcp_g += src_linesize;
502 srcp_b += src_linesize;
503 dstp_r += dst_linesize;
504 dstp_g += dst_linesize;
505 dstp_b += dst_linesize;
506 }
507 }
508
509 static av_always_inline void color_correlation_gbrp(uint8_t **dst, int dst_linesize,
510 float **src, int src_linesize,
511 int w, int h)
512 {
513 int x, y;
514 const float *src_r = src[0];
515 const float *src_g = src[1];
516 const float *src_b = src[2];
517 uint8_t *dstp_r = dst[2];
518 uint8_t *dstp_g = dst[0];
519 uint8_t *dstp_b = dst[1];
520
521 for (y = 0; y < h; y++) {
522 for (x = 0; x < w; x++) {
523 dstp_r[x] = av_clip_uint8(src_r[x] * DCT3X3_0_0 + src_g[x] * DCT3X3_1_0 + src_b[x] * DCT3X3_2_0);
524 dstp_g[x] = av_clip_uint8(src_r[x] * DCT3X3_0_1 + src_b[x] * DCT3X3_2_1);
525 dstp_b[x] = av_clip_uint8(src_r[x] * DCT3X3_0_2 + src_g[x] * DCT3X3_1_2 + src_b[x] * DCT3X3_2_2);
526 }
527 dstp_r += dst_linesize;
528 dstp_g += dst_linesize;
529 dstp_b += dst_linesize;
530 src_r += src_linesize;
531 src_g += src_linesize;
532 src_b += src_linesize;
533 }
534 }
535
536 static int config_input(AVFilterLink *inlink)
537 {
538 AVFilterContext *ctx = inlink->dst;
539 DCTdnoizContext *s = ctx->priv;
540 int i, x, y, bx, by, linesize, *iweights, max_slice_h, slice_h;
541 const int bsize = 1 << s->n;
542
543 switch (inlink->format) {
544 case AV_PIX_FMT_BGR24:
545 s->color_decorrelation = color_decorrelation_bgr;
546 s->color_correlation = color_correlation_bgr;
547 break;
548 case AV_PIX_FMT_RGB24:
549 s->color_decorrelation = color_decorrelation_rgb;
550 s->color_correlation = color_correlation_rgb;
551 break;
552 case AV_PIX_FMT_GBRP:
553 s->color_decorrelation = color_decorrelation_gbrp;
554 s->color_correlation = color_correlation_gbrp;
555 break;
556 default:
557 av_assert0(0);
558 }
559
560 s->pr_width = inlink->w - (inlink->w - bsize) % s->step;
561 s->pr_height = inlink->h - (inlink->h - bsize) % s->step;
562 if (s->pr_width != inlink->w)
563 av_log(ctx, AV_LOG_WARNING, "The last %d horizontal pixels won't be denoised\n",
564 inlink->w - s->pr_width);
565 if (s->pr_height != inlink->h)
566 av_log(ctx, AV_LOG_WARNING, "The last %d vertical pixels won't be denoised\n",
567 inlink->h - s->pr_height);
568
569 max_slice_h = s->pr_height / ((s->bsize - 1) * 2);
570 if (max_slice_h == 0)
571 return AVERROR(EINVAL);
572
573 s->nb_threads = FFMIN3(MAX_THREADS, ff_filter_get_nb_threads(ctx), max_slice_h);
574 av_log(ctx, AV_LOG_DEBUG, "threads: [max=%d hmax=%d user=%d] => %d\n",
575 MAX_THREADS, max_slice_h, ff_filter_get_nb_threads(ctx), s->nb_threads);
576
577 s->p_linesize = linesize = FFALIGN(s->pr_width, 32);
578 for (i = 0; i < 2; i++) {
579 s->cbuf[i][0] = av_malloc_array(linesize * s->pr_height, sizeof(*s->cbuf[i][0]));
580 s->cbuf[i][1] = av_malloc_array(linesize * s->pr_height, sizeof(*s->cbuf[i][1]));
581 s->cbuf[i][2] = av_malloc_array(linesize * s->pr_height, sizeof(*s->cbuf[i][2]));
582 if (!s->cbuf[i][0] || !s->cbuf[i][1] || !s->cbuf[i][2])
583 return AVERROR(ENOMEM);
584 }
585
586 /* eval expressions are probably not thread safe when the eval internal
587 * state can be changed (typically through load & store operations) */
588 if (s->expr_str) {
589 for (i = 0; i < s->nb_threads; i++) {
590 int ret = av_expr_parse(&s->expr[i], s->expr_str, var_names,
591 NULL, NULL, NULL, NULL, 0, ctx);
592 if (ret < 0)
593 return ret;
594 }
595 }
596
597 /* each slice will need to (pre & re)process the top and bottom block of
598 * the previous one in in addition to its processing area. This is because
599 * each pixel is averaged by all the surrounding blocks */
600 slice_h = (int)ceilf(s->pr_height / (float)s->nb_threads) + (s->bsize - 1) * 2;
601 for (i = 0; i < s->nb_threads; i++) {
602 s->slices[i] = av_malloc_array(linesize, slice_h * sizeof(*s->slices[i]));
603 if (!s->slices[i])
604 return AVERROR(ENOMEM);
605 }
606
607 s->weights = av_malloc(s->pr_height * linesize * sizeof(*s->weights));
608 if (!s->weights)
609 return AVERROR(ENOMEM);
610 iweights = av_calloc(s->pr_height, linesize * sizeof(*iweights));
611 if (!iweights)
612 return AVERROR(ENOMEM);
613 for (y = 0; y < s->pr_height - bsize + 1; y += s->step)
614 for (x = 0; x < s->pr_width - bsize + 1; x += s->step)
615 for (by = 0; by < bsize; by++)
616 for (bx = 0; bx < bsize; bx++)
617 iweights[(y + by)*linesize + x + bx]++;
618 for (y = 0; y < s->pr_height; y++)
619 for (x = 0; x < s->pr_width; x++)
620 s->weights[y*linesize + x] = 1. / iweights[y*linesize + x];
621 av_free(iweights);
622
623 return 0;
624 }
625
626 static av_cold int init(AVFilterContext *ctx)
627 {
628 DCTdnoizContext *s = ctx->priv;
629
630 s->bsize = 1 << s->n;
631 if (s->overlap == -1)
632 s->overlap = s->bsize - 1;
633
634 if (s->overlap > s->bsize - 1) {
635 av_log(s, AV_LOG_ERROR, "Overlap value can not except %d "
636 "with a block size of %dx%d\n",
637 s->bsize - 1, s->bsize, s->bsize);
638 return AVERROR(EINVAL);
639 }
640
641 if (s->expr_str) {
642 switch (s->n) {
643 case 3: s->filter_freq_func = filter_freq_expr_8; break;
644 case 4: s->filter_freq_func = filter_freq_expr_16; break;
645 default: av_assert0(0);
646 }
647 } else {
648 switch (s->n) {
649 case 3: s->filter_freq_func = filter_freq_sigma_8; break;
650 case 4: s->filter_freq_func = filter_freq_sigma_16; break;
651 default: av_assert0(0);
652 }
653 }
654
655 s->th = s->sigma * 3.;
656 s->step = s->bsize - s->overlap;
657 return 0;
658 }
659
660 static const enum AVPixelFormat pix_fmts[] = {
661 AV_PIX_FMT_BGR24, AV_PIX_FMT_RGB24,
662 AV_PIX_FMT_GBRP,
663 AV_PIX_FMT_NONE
664 };
665
666 typedef struct ThreadData {
667 float *src, *dst;
668 } ThreadData;
669
670 static int filter_slice(AVFilterContext *ctx,
671 void *arg, int jobnr, int nb_jobs)
672 {
673 int x, y;
674 DCTdnoizContext *s = ctx->priv;
675 const ThreadData *td = arg;
676 const int w = s->pr_width;
677 const int h = s->pr_height;
678 const int slice_start = (h * jobnr ) / nb_jobs;
679 const int slice_end = (h * (jobnr+1)) / nb_jobs;
680 const int slice_start_ctx = FFMAX(slice_start - s->bsize + 1, 0);
681 const int slice_end_ctx = FFMIN(slice_end, h - s->bsize + 1);
682 const int slice_h = slice_end_ctx - slice_start_ctx;
683 const int src_linesize = s->p_linesize;
684 const int dst_linesize = s->p_linesize;
685 const int slice_linesize = s->p_linesize;
686 float *dst;
687 const float *src = td->src + slice_start_ctx * src_linesize;
688 const float *weights = s->weights + slice_start * dst_linesize;
689 float *slice = s->slices[jobnr];
690
691 // reset block sums
692 memset(slice, 0, (slice_h + s->bsize - 1) * dst_linesize * sizeof(*slice));
693
694 // block dct sums
695 for (y = 0; y < slice_h; y += s->step) {
696 for (x = 0; x < w - s->bsize + 1; x += s->step)
697 s->filter_freq_func(s, src + x, src_linesize,
698 slice + x, slice_linesize,
699 jobnr);
700 src += s->step * src_linesize;
701 slice += s->step * slice_linesize;
702 }
703
704 // average blocks
705 slice = s->slices[jobnr] + (slice_start - slice_start_ctx) * slice_linesize;
706 dst = td->dst + slice_start * dst_linesize;
707 for (y = slice_start; y < slice_end; y++) {
708 for (x = 0; x < w; x++)
709 dst[x] = slice[x] * weights[x];
710 slice += slice_linesize;
711 dst += dst_linesize;
712 weights += dst_linesize;
713 }
714
715 return 0;
716 }
717
718 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
719 {
720 AVFilterContext *ctx = inlink->dst;
721 DCTdnoizContext *s = ctx->priv;
722 AVFilterLink *outlink = inlink->dst->outputs[0];
723 int direct, plane;
724 AVFrame *out;
725
726 if (av_frame_is_writable(in)) {
727 direct = 1;
728 out = in;
729 } else {
730 direct = 0;
731 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
732 if (!out) {
733 av_frame_free(&in);
734 return AVERROR(ENOMEM);
735 }
736 av_frame_copy_props(out, in);
737 }
738
739 s->color_decorrelation(s->cbuf[0], s->p_linesize,
740 (const uint8_t **)in->data, in->linesize[0],
741 s->pr_width, s->pr_height);
742 for (plane = 0; plane < 3; plane++) {
743 ThreadData td = {
744 .src = s->cbuf[0][plane],
745 .dst = s->cbuf[1][plane],
746 };
747 ff_filter_execute(ctx, filter_slice, &td, NULL, s->nb_threads);
748 }
749 s->color_correlation(out->data, out->linesize[0],
750 s->cbuf[1], s->p_linesize,
751 s->pr_width, s->pr_height);
752
753 if (!direct) {
754 int y;
755 uint8_t *dst = out->data[0];
756 const uint8_t *src = in->data[0];
757 const int dst_linesize = out->linesize[0];
758 const int src_linesize = in->linesize[0];
759 const int hpad = (inlink->w - s->pr_width) * 3;
760 const int vpad = (inlink->h - s->pr_height);
761
762 if (hpad) {
763 uint8_t *dstp = dst + s->pr_width * 3;
764 const uint8_t *srcp = src + s->pr_width * 3;
765
766 for (y = 0; y < s->pr_height; y++) {
767 memcpy(dstp, srcp, hpad);
768 dstp += dst_linesize;
769 srcp += src_linesize;
770 }
771 }
772 if (vpad) {
773 uint8_t *dstp = dst + s->pr_height * dst_linesize;
774 const uint8_t *srcp = src + s->pr_height * src_linesize;
775
776 for (y = 0; y < vpad; y++) {
777 memcpy(dstp, srcp, inlink->w * 3);
778 dstp += dst_linesize;
779 srcp += src_linesize;
780 }
781 }
782
783 av_frame_free(&in);
784 }
785
786 return ff_filter_frame(outlink, out);
787 }
788
789 static av_cold void uninit(AVFilterContext *ctx)
790 {
791 int i;
792 DCTdnoizContext *s = ctx->priv;
793
794 av_freep(&s->weights);
795 for (i = 0; i < 2; i++) {
796 av_freep(&s->cbuf[i][0]);
797 av_freep(&s->cbuf[i][1]);
798 av_freep(&s->cbuf[i][2]);
799 }
800 for (i = 0; i < s->nb_threads; i++) {
801 av_freep(&s->slices[i]);
802 av_expr_free(s->expr[i]);
803 }
804 }
805
806 static const AVFilterPad dctdnoiz_inputs[] = {
807 {
808 .name = "default",
809 .type = AVMEDIA_TYPE_VIDEO,
810 .filter_frame = filter_frame,
811 .config_props = config_input,
812 },
813 };
814
815 const FFFilter ff_vf_dctdnoiz = {
816 .p.name = "dctdnoiz",
817 .p.description = NULL_IF_CONFIG_SMALL("Denoise frames using 2D DCT."),
818 .p.priv_class = &dctdnoiz_class,
819 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
820 .priv_size = sizeof(DCTdnoizContext),
821 .init = init,
822 .uninit = uninit,
823 FILTER_INPUTS(dctdnoiz_inputs),
824 FILTER_OUTPUTS(ff_video_default_filterpad),
825 FILTER_PIXFMTS_ARRAY(pix_fmts),
826 };
Mirror of the official FFMpeg Git repo