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