You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

460 lines
19KB

  1. /*
  2. * Copyright (c) 2015 Paul B Mahol
  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. * @file
  22. * Adaptive Temporal Averaging Denoiser,
  23. * based on paper "Video Denoising Based on Adaptive Temporal Averaging" by
  24. * David Bartovčak and Miroslav Vrankić
  25. */
  26. #include "libavutil/imgutils.h"
  27. #include "libavutil/opt.h"
  28. #include "libavutil/pixdesc.h"
  29. #include "avfilter.h"
  30. #define FF_BUFQUEUE_SIZE 129
  31. #include "bufferqueue.h"
  32. #include "atadenoise.h"
  33. #include "formats.h"
  34. #include "internal.h"
  35. #include "video.h"
  36. #define SIZE FF_BUFQUEUE_SIZE
  37. typedef struct ATADenoiseContext {
  38. const AVClass *class;
  39. float fthra[4], fthrb[4];
  40. int thra[4], thrb[4];
  41. int algorithm;
  42. int planes;
  43. int nb_planes;
  44. int planewidth[4];
  45. int planeheight[4];
  46. struct FFBufQueue q;
  47. void *data[4][SIZE];
  48. int linesize[4][SIZE];
  49. int size, mid;
  50. int available;
  51. int (*filter_slice)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
  52. ATADenoiseDSPContext dsp;
  53. } ATADenoiseContext;
  54. #define OFFSET(x) offsetof(ATADenoiseContext, x)
  55. #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
  56. #define VF AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
  57. static const AVOption atadenoise_options[] = {
  58. { "0a", "set threshold A for 1st plane", OFFSET(fthra[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.02}, 0, 0.3, FLAGS },
  59. { "0b", "set threshold B for 1st plane", OFFSET(fthrb[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.04}, 0, 5.0, FLAGS },
  60. { "1a", "set threshold A for 2nd plane", OFFSET(fthra[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.02}, 0, 0.3, FLAGS },
  61. { "1b", "set threshold B for 2nd plane", OFFSET(fthrb[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.04}, 0, 5.0, FLAGS },
  62. { "2a", "set threshold A for 3rd plane", OFFSET(fthra[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.02}, 0, 0.3, FLAGS },
  63. { "2b", "set threshold B for 3rd plane", OFFSET(fthrb[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.04}, 0, 5.0, FLAGS },
  64. { "s", "set how many frames to use", OFFSET(size), AV_OPT_TYPE_INT, {.i64=9}, 5, SIZE, VF },
  65. { "p", "set what planes to filter", OFFSET(planes), AV_OPT_TYPE_FLAGS, {.i64=7}, 0, 15, FLAGS },
  66. { "a", "set variant of algorithm", OFFSET(algorithm),AV_OPT_TYPE_INT, {.i64=PARALLEL}, 0, NB_ATAA-1, FLAGS, "a" },
  67. { "p", "parallel", 0, AV_OPT_TYPE_CONST, {.i64=PARALLEL}, 0, 0, FLAGS, "a" },
  68. { "s", "serial", 0, AV_OPT_TYPE_CONST, {.i64=SERIAL}, 0, 0, FLAGS, "a" },
  69. { NULL }
  70. };
  71. AVFILTER_DEFINE_CLASS(atadenoise);
  72. static int query_formats(AVFilterContext *ctx)
  73. {
  74. static const enum AVPixelFormat pixel_fmts[] = {
  75. AV_PIX_FMT_GRAY8,
  76. AV_PIX_FMT_GRAY9,
  77. AV_PIX_FMT_GRAY10,
  78. AV_PIX_FMT_GRAY12,
  79. AV_PIX_FMT_GRAY14,
  80. AV_PIX_FMT_GRAY16,
  81. AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P,
  82. AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P,
  83. AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV444P,
  84. AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
  85. AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ444P,
  86. AV_PIX_FMT_YUVJ411P,
  87. AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
  88. AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
  89. AV_PIX_FMT_YUV440P10,
  90. AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV420P12,
  91. AV_PIX_FMT_YUV440P12,
  92. AV_PIX_FMT_YUV444P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV420P14,
  93. AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
  94. AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
  95. AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
  96. AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA444P,
  97. AV_PIX_FMT_YUVA444P9, AV_PIX_FMT_YUVA444P10, AV_PIX_FMT_YUVA444P12, AV_PIX_FMT_YUVA444P16,
  98. AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA422P12, AV_PIX_FMT_YUVA422P16,
  99. AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA420P16,
  100. AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
  101. AV_PIX_FMT_NONE
  102. };
  103. AVFilterFormats *formats = ff_make_format_list(pixel_fmts);
  104. if (!formats)
  105. return AVERROR(ENOMEM);
  106. return ff_set_common_formats(ctx, formats);
  107. }
  108. static av_cold int init(AVFilterContext *ctx)
  109. {
  110. ATADenoiseContext *s = ctx->priv;
  111. if (!(s->size & 1)) {
  112. av_log(ctx, AV_LOG_WARNING, "size %d is invalid. Must be an odd value, setting it to %d.\n", s->size, s->size|1);
  113. s->size |= 1;
  114. }
  115. s->mid = s->size / 2 + 1;
  116. return 0;
  117. }
  118. typedef struct ThreadData {
  119. AVFrame *in, *out;
  120. } ThreadData;
  121. #define FILTER_ROW(type, name) \
  122. static void filter_row##name(const uint8_t *ssrc, uint8_t *ddst, \
  123. const uint8_t *ssrcf[SIZE], \
  124. int w, int mid, int size, \
  125. int thra, int thrb) \
  126. { \
  127. const type *src = (const type *)ssrc; \
  128. const type **srcf = (const type **)ssrcf; \
  129. type *dst = (type *)ddst; \
  130. \
  131. for (int x = 0; x < w; x++) { \
  132. const int srcx = src[x]; \
  133. unsigned lsumdiff = 0, rsumdiff = 0; \
  134. unsigned ldiff, rdiff; \
  135. unsigned sum = srcx; \
  136. int l = 0, r = 0; \
  137. int srcjx, srcix; \
  138. \
  139. for (int j = mid - 1, i = mid + 1; j >= 0 && i < size; j--, i++) { \
  140. srcjx = srcf[j][x]; \
  141. \
  142. ldiff = FFABS(srcx - srcjx); \
  143. lsumdiff += ldiff; \
  144. if (ldiff > thra || \
  145. lsumdiff > thrb) \
  146. break; \
  147. l++; \
  148. sum += srcjx; \
  149. \
  150. srcix = srcf[i][x]; \
  151. \
  152. rdiff = FFABS(srcx - srcix); \
  153. rsumdiff += rdiff; \
  154. if (rdiff > thra || \
  155. rsumdiff > thrb) \
  156. break; \
  157. r++; \
  158. sum += srcix; \
  159. } \
  160. \
  161. dst[x] = (sum + ((r + l + 1) >> 1)) / (r + l + 1); \
  162. } \
  163. }
  164. FILTER_ROW(uint8_t, 8)
  165. FILTER_ROW(uint16_t, 16)
  166. #define FILTER_ROW_SERIAL(type, name) \
  167. static void filter_row##name##_serial(const uint8_t *ssrc, uint8_t *ddst, \
  168. const uint8_t *ssrcf[SIZE], \
  169. int w, int mid, int size, \
  170. int thra, int thrb) \
  171. { \
  172. const type *src = (const type *)ssrc; \
  173. const type **srcf = (const type **)ssrcf; \
  174. type *dst = (type *)ddst; \
  175. \
  176. for (int x = 0; x < w; x++) { \
  177. const int srcx = src[x]; \
  178. unsigned lsumdiff = 0, rsumdiff = 0; \
  179. unsigned ldiff, rdiff; \
  180. unsigned sum = srcx; \
  181. int l = 0, r = 0; \
  182. int srcjx, srcix; \
  183. \
  184. for (int j = mid - 1; j >= 0; j--) { \
  185. srcjx = srcf[j][x]; \
  186. \
  187. ldiff = FFABS(srcx - srcjx); \
  188. lsumdiff += ldiff; \
  189. if (ldiff > thra || \
  190. lsumdiff > thrb) \
  191. break; \
  192. l++; \
  193. sum += srcjx; \
  194. } \
  195. \
  196. for (int i = mid + 1; i < size; i++) { \
  197. srcix = srcf[i][x]; \
  198. \
  199. rdiff = FFABS(srcx - srcix); \
  200. rsumdiff += rdiff; \
  201. if (rdiff > thra || \
  202. rsumdiff > thrb) \
  203. break; \
  204. r++; \
  205. sum += srcix; \
  206. } \
  207. \
  208. dst[x] = (sum + ((r + l + 1) >> 1)) / (r + l + 1); \
  209. } \
  210. }
  211. FILTER_ROW_SERIAL(uint8_t, 8)
  212. FILTER_ROW_SERIAL(uint16_t, 16)
  213. static int filter_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
  214. {
  215. ATADenoiseContext *s = ctx->priv;
  216. ThreadData *td = arg;
  217. AVFrame *in = td->in;
  218. AVFrame *out = td->out;
  219. const int size = s->size;
  220. const int mid = s->mid;
  221. int p, y, i;
  222. for (p = 0; p < s->nb_planes; p++) {
  223. const int h = s->planeheight[p];
  224. const int w = s->planewidth[p];
  225. const int slice_start = (h * jobnr) / nb_jobs;
  226. const int slice_end = (h * (jobnr+1)) / nb_jobs;
  227. const uint8_t *src = in->data[p] + slice_start * in->linesize[p];
  228. uint8_t *dst = out->data[p] + slice_start * out->linesize[p];
  229. const int thra = s->thra[p];
  230. const int thrb = s->thrb[p];
  231. const uint8_t **data = (const uint8_t **)s->data[p];
  232. const int *linesize = (const int *)s->linesize[p];
  233. const uint8_t *srcf[SIZE];
  234. if (!((1 << p) & s->planes)) {
  235. av_image_copy_plane(dst, out->linesize[p], src, in->linesize[p],
  236. w, slice_end - slice_start);
  237. continue;
  238. }
  239. for (i = 0; i < size; i++)
  240. srcf[i] = data[i] + slice_start * linesize[i];
  241. for (y = slice_start; y < slice_end; y++) {
  242. s->dsp.filter_row(src, dst, srcf, w, mid, size, thra, thrb);
  243. dst += out->linesize[p];
  244. src += in->linesize[p];
  245. for (i = 0; i < size; i++)
  246. srcf[i] += linesize[i];
  247. }
  248. }
  249. return 0;
  250. }
  251. static int config_input(AVFilterLink *inlink)
  252. {
  253. const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
  254. AVFilterContext *ctx = inlink->dst;
  255. ATADenoiseContext *s = ctx->priv;
  256. int depth;
  257. s->nb_planes = desc->nb_components;
  258. s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
  259. s->planeheight[0] = s->planeheight[3] = inlink->h;
  260. s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
  261. s->planewidth[0] = s->planewidth[3] = inlink->w;
  262. depth = desc->comp[0].depth;
  263. s->filter_slice = filter_slice;
  264. if (depth == 8)
  265. s->dsp.filter_row = s->algorithm == PARALLEL ? filter_row8 : filter_row8_serial;
  266. else
  267. s->dsp.filter_row = s->algorithm == PARALLEL ? filter_row16 : filter_row16_serial;
  268. s->thra[0] = s->fthra[0] * (1 << depth) - 1;
  269. s->thra[1] = s->fthra[1] * (1 << depth) - 1;
  270. s->thra[2] = s->fthra[2] * (1 << depth) - 1;
  271. s->thrb[0] = s->fthrb[0] * (1 << depth) - 1;
  272. s->thrb[1] = s->fthrb[1] * (1 << depth) - 1;
  273. s->thrb[2] = s->fthrb[2] * (1 << depth) - 1;
  274. if (ARCH_X86)
  275. ff_atadenoise_init_x86(&s->dsp, depth, s->algorithm);
  276. return 0;
  277. }
  278. static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
  279. {
  280. AVFilterContext *ctx = inlink->dst;
  281. AVFilterLink *outlink = ctx->outputs[0];
  282. ATADenoiseContext *s = ctx->priv;
  283. AVFrame *out, *in;
  284. int i;
  285. if (s->q.available != s->size) {
  286. if (s->q.available < s->mid) {
  287. for (i = 0; i < s->mid; i++) {
  288. out = av_frame_clone(buf);
  289. if (!out) {
  290. av_frame_free(&buf);
  291. return AVERROR(ENOMEM);
  292. }
  293. ff_bufqueue_add(ctx, &s->q, out);
  294. }
  295. }
  296. if (s->q.available < s->size) {
  297. ff_bufqueue_add(ctx, &s->q, buf);
  298. s->available++;
  299. }
  300. return 0;
  301. }
  302. in = ff_bufqueue_peek(&s->q, s->mid);
  303. if (!ctx->is_disabled) {
  304. ThreadData td;
  305. out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
  306. if (!out) {
  307. av_frame_free(&buf);
  308. return AVERROR(ENOMEM);
  309. }
  310. for (i = 0; i < s->size; i++) {
  311. AVFrame *frame = ff_bufqueue_peek(&s->q, i);
  312. s->data[0][i] = frame->data[0];
  313. s->data[1][i] = frame->data[1];
  314. s->data[2][i] = frame->data[2];
  315. s->linesize[0][i] = frame->linesize[0];
  316. s->linesize[1][i] = frame->linesize[1];
  317. s->linesize[2][i] = frame->linesize[2];
  318. }
  319. td.in = in; td.out = out;
  320. ctx->internal->execute(ctx, s->filter_slice, &td, NULL,
  321. FFMIN3(s->planeheight[1],
  322. s->planeheight[2],
  323. ff_filter_get_nb_threads(ctx)));
  324. av_frame_copy_props(out, in);
  325. } else {
  326. out = av_frame_clone(in);
  327. if (!out) {
  328. av_frame_free(&buf);
  329. return AVERROR(ENOMEM);
  330. }
  331. }
  332. in = ff_bufqueue_get(&s->q);
  333. av_frame_free(&in);
  334. ff_bufqueue_add(ctx, &s->q, buf);
  335. return ff_filter_frame(outlink, out);
  336. }
  337. static int request_frame(AVFilterLink *outlink)
  338. {
  339. AVFilterContext *ctx = outlink->src;
  340. ATADenoiseContext *s = ctx->priv;
  341. int ret = 0;
  342. ret = ff_request_frame(ctx->inputs[0]);
  343. if (ret == AVERROR_EOF && !ctx->is_disabled && s->available) {
  344. AVFrame *buf = av_frame_clone(ff_bufqueue_peek(&s->q, s->available));
  345. if (!buf)
  346. return AVERROR(ENOMEM);
  347. ret = filter_frame(ctx->inputs[0], buf);
  348. s->available--;
  349. }
  350. return ret;
  351. }
  352. static av_cold void uninit(AVFilterContext *ctx)
  353. {
  354. ATADenoiseContext *s = ctx->priv;
  355. ff_bufqueue_discard_all(&s->q);
  356. }
  357. static int process_command(AVFilterContext *ctx,
  358. const char *cmd,
  359. const char *arg,
  360. char *res,
  361. int res_len,
  362. int flags)
  363. {
  364. int ret = ff_filter_process_command(ctx, cmd, arg, res, res_len, flags);
  365. if (ret < 0)
  366. return ret;
  367. return config_input(ctx->inputs[0]);
  368. }
  369. static const AVFilterPad inputs[] = {
  370. {
  371. .name = "default",
  372. .type = AVMEDIA_TYPE_VIDEO,
  373. .filter_frame = filter_frame,
  374. .config_props = config_input,
  375. },
  376. { NULL }
  377. };
  378. static const AVFilterPad outputs[] = {
  379. {
  380. .name = "default",
  381. .type = AVMEDIA_TYPE_VIDEO,
  382. .request_frame = request_frame,
  383. },
  384. { NULL }
  385. };
  386. AVFilter ff_vf_atadenoise = {
  387. .name = "atadenoise",
  388. .description = NULL_IF_CONFIG_SMALL("Apply an Adaptive Temporal Averaging Denoiser."),
  389. .priv_size = sizeof(ATADenoiseContext),
  390. .priv_class = &atadenoise_class,
  391. .init = init,
  392. .uninit = uninit,
  393. .query_formats = query_formats,
  394. .inputs = inputs,
  395. .outputs = outputs,
  396. .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
  397. .process_command = process_command,
  398. };