2 * Copyright (C) 2012 British Broadcasting Corporation, All Rights Reserved
3 * Author of de-interlace algorithm: Jim Easterbrook for BBC R&D
4 * Based on the process described by Martin Weston for BBC R&D
5 * Author of FFmpeg filter: Mark Himsley for BBC Broadcast Systems Development
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 #include "libavutil/common.h"
25 #include "libavutil/imgutils.h"
26 #include "libavutil/opt.h"
27 #include "libavutil/pixdesc.h"
34 typedef struct W3FDIFContext {
36 int filter; ///< 0 is simple, 1 is more complex
37 int mode; ///< 0 is frame, 1 is field
38 int parity; ///< frame field parity
39 int deint; ///< which frames to deinterlace
40 int linesize[4]; ///< bytes of pixel data per line for each plane
41 int planeheight[4]; ///< height of each plane
42 int field; ///< which field are we on, 0 or 1
45 AVFrame *prev, *cur, *next; ///< previous, current, next frames
46 int32_t **work_line; ///< lines we are calculating
53 #define OFFSET(x) offsetof(W3FDIFContext, x)
54 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
55 #define CONST(name, help, val, unit) { name, help, 0, AV_OPT_TYPE_CONST, {.i64=val}, 0, 0, FLAGS, unit }
57 static const AVOption w3fdif_options[] = {
58 { "filter", "specify the filter", OFFSET(filter), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, FLAGS, "filter" },
59 CONST("simple", NULL, 0, "filter"),
60 CONST("complex", NULL, 1, "filter"),
61 { "mode", "specify the interlacing mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, FLAGS, "mode"},
62 CONST("frame", "send one frame for each frame", 0, "mode"),
63 CONST("field", "send one frame for each field", 1, "mode"),
64 { "parity", "specify the assumed picture field parity", OFFSET(parity), AV_OPT_TYPE_INT, {.i64=-1}, -1, 1, FLAGS, "parity" },
65 CONST("tff", "assume top field first", 0, "parity"),
66 CONST("bff", "assume bottom field first", 1, "parity"),
67 CONST("auto", "auto detect parity", -1, "parity"),
68 { "deint", "specify which frames to deinterlace", OFFSET(deint), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, FLAGS, "deint" },
69 CONST("all", "deinterlace all frames", 0, "deint"),
70 CONST("interlaced", "only deinterlace frames marked as interlaced", 1, "deint"),
74 AVFILTER_DEFINE_CLASS(w3fdif);
76 static int query_formats(AVFilterContext *ctx)
78 static const enum AVPixelFormat pix_fmts[] = {
79 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P,
80 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P,
81 AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV444P,
82 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
83 AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
85 AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA444P,
86 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP,
88 AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16,
89 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
90 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
92 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12,
94 AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
95 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
96 AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
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_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
104 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
106 return AVERROR(ENOMEM);
107 return ff_set_common_formats(ctx, fmts_list);
110 static void filter_simple_low(int32_t *work_line,
111 uint8_t *in_lines_cur[2],
112 const int16_t *coef, int linesize)
116 for (i = 0; i < linesize; i++) {
117 *work_line = *in_lines_cur[0]++ * coef[0];
118 *work_line++ += *in_lines_cur[1]++ * coef[1];
122 static void filter_complex_low(int32_t *work_line,
123 uint8_t *in_lines_cur[4],
124 const int16_t *coef, int linesize)
128 for (i = 0; i < linesize; i++) {
129 *work_line = *in_lines_cur[0]++ * coef[0];
130 *work_line += *in_lines_cur[1]++ * coef[1];
131 *work_line += *in_lines_cur[2]++ * coef[2];
132 *work_line++ += *in_lines_cur[3]++ * coef[3];
136 static void filter_simple_high(int32_t *work_line,
137 uint8_t *in_lines_cur[3],
138 uint8_t *in_lines_adj[3],
139 const int16_t *coef, int linesize)
143 for (i = 0; i < linesize; i++) {
144 *work_line += *in_lines_cur[0]++ * coef[0];
145 *work_line += *in_lines_adj[0]++ * coef[0];
146 *work_line += *in_lines_cur[1]++ * coef[1];
147 *work_line += *in_lines_adj[1]++ * coef[1];
148 *work_line += *in_lines_cur[2]++ * coef[2];
149 *work_line++ += *in_lines_adj[2]++ * coef[2];
153 static void filter_complex_high(int32_t *work_line,
154 uint8_t *in_lines_cur[5],
155 uint8_t *in_lines_adj[5],
156 const int16_t *coef, int linesize)
160 for (i = 0; i < linesize; i++) {
161 *work_line += *in_lines_cur[0]++ * coef[0];
162 *work_line += *in_lines_adj[0]++ * coef[0];
163 *work_line += *in_lines_cur[1]++ * coef[1];
164 *work_line += *in_lines_adj[1]++ * coef[1];
165 *work_line += *in_lines_cur[2]++ * coef[2];
166 *work_line += *in_lines_adj[2]++ * coef[2];
167 *work_line += *in_lines_cur[3]++ * coef[3];
168 *work_line += *in_lines_adj[3]++ * coef[3];
169 *work_line += *in_lines_cur[4]++ * coef[4];
170 *work_line++ += *in_lines_adj[4]++ * coef[4];
174 static void filter_scale(uint8_t *out_pixel, const int32_t *work_pixel, int linesize, int max)
178 for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
179 *out_pixel = av_clip(*work_pixel, 0, 255 * 256 * 128) >> 15;
182 static void filter16_simple_low(int32_t *work_line,
183 uint8_t *in_lines_cur8[2],
184 const int16_t *coef, int linesize)
186 uint16_t *in_lines_cur[2] = { (uint16_t *)in_lines_cur8[0], (uint16_t *)in_lines_cur8[1] };
190 for (i = 0; i < linesize; i++) {
191 *work_line = *in_lines_cur[0]++ * coef[0];
192 *work_line++ += *in_lines_cur[1]++ * coef[1];
196 static void filter16_complex_low(int32_t *work_line,
197 uint8_t *in_lines_cur8[4],
198 const int16_t *coef, int linesize)
200 uint16_t *in_lines_cur[4] = { (uint16_t *)in_lines_cur8[0],
201 (uint16_t *)in_lines_cur8[1],
202 (uint16_t *)in_lines_cur8[2],
203 (uint16_t *)in_lines_cur8[3] };
207 for (i = 0; i < linesize; i++) {
208 *work_line = *in_lines_cur[0]++ * coef[0];
209 *work_line += *in_lines_cur[1]++ * coef[1];
210 *work_line += *in_lines_cur[2]++ * coef[2];
211 *work_line++ += *in_lines_cur[3]++ * coef[3];
215 static void filter16_simple_high(int32_t *work_line,
216 uint8_t *in_lines_cur8[3],
217 uint8_t *in_lines_adj8[3],
218 const int16_t *coef, int linesize)
220 uint16_t *in_lines_cur[3] = { (uint16_t *)in_lines_cur8[0],
221 (uint16_t *)in_lines_cur8[1],
222 (uint16_t *)in_lines_cur8[2] };
223 uint16_t *in_lines_adj[3] = { (uint16_t *)in_lines_adj8[0],
224 (uint16_t *)in_lines_adj8[1],
225 (uint16_t *)in_lines_adj8[2] };
229 for (i = 0; i < linesize; i++) {
230 *work_line += *in_lines_cur[0]++ * coef[0];
231 *work_line += *in_lines_adj[0]++ * coef[0];
232 *work_line += *in_lines_cur[1]++ * coef[1];
233 *work_line += *in_lines_adj[1]++ * coef[1];
234 *work_line += *in_lines_cur[2]++ * coef[2];
235 *work_line++ += *in_lines_adj[2]++ * coef[2];
239 static void filter16_complex_high(int32_t *work_line,
240 uint8_t *in_lines_cur8[5],
241 uint8_t *in_lines_adj8[5],
242 const int16_t *coef, int linesize)
244 uint16_t *in_lines_cur[5] = { (uint16_t *)in_lines_cur8[0],
245 (uint16_t *)in_lines_cur8[1],
246 (uint16_t *)in_lines_cur8[2],
247 (uint16_t *)in_lines_cur8[3],
248 (uint16_t *)in_lines_cur8[4] };
249 uint16_t *in_lines_adj[5] = { (uint16_t *)in_lines_adj8[0],
250 (uint16_t *)in_lines_adj8[1],
251 (uint16_t *)in_lines_adj8[2],
252 (uint16_t *)in_lines_adj8[3],
253 (uint16_t *)in_lines_adj8[4] };
257 for (i = 0; i < linesize; i++) {
258 *work_line += *in_lines_cur[0]++ * coef[0];
259 *work_line += *in_lines_adj[0]++ * coef[0];
260 *work_line += *in_lines_cur[1]++ * coef[1];
261 *work_line += *in_lines_adj[1]++ * coef[1];
262 *work_line += *in_lines_cur[2]++ * coef[2];
263 *work_line += *in_lines_adj[2]++ * coef[2];
264 *work_line += *in_lines_cur[3]++ * coef[3];
265 *work_line += *in_lines_adj[3]++ * coef[3];
266 *work_line += *in_lines_cur[4]++ * coef[4];
267 *work_line++ += *in_lines_adj[4]++ * coef[4];
271 static void filter16_scale(uint8_t *out_pixel8, const int32_t *work_pixel, int linesize, int max)
273 uint16_t *out_pixel = (uint16_t *)out_pixel8;
277 for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
278 *out_pixel = av_clip(*work_pixel, 0, max) >> 15;
281 static int config_input(AVFilterLink *inlink)
283 AVFilterContext *ctx = inlink->dst;
284 W3FDIFContext *s = ctx->priv;
285 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
288 if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
291 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
292 s->planeheight[0] = s->planeheight[3] = inlink->h;
295 av_log(ctx, AV_LOG_ERROR, "Video of less than 3 lines is not supported\n");
296 return AVERROR(EINVAL);
299 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
300 s->nb_threads = ff_filter_get_nb_threads(ctx);
301 s->work_line = av_calloc(s->nb_threads, sizeof(*s->work_line));
303 return AVERROR(ENOMEM);
305 for (i = 0; i < s->nb_threads; i++) {
306 s->work_line[i] = av_calloc(FFALIGN(s->linesize[0], 32), sizeof(*s->work_line[0]));
307 if (!s->work_line[i])
308 return AVERROR(ENOMEM);
311 depth = desc->comp[0].depth;
312 s->max = ((1 << depth) - 1) * 256 * 128;
314 s->dsp.filter_simple_low = filter_simple_low;
315 s->dsp.filter_complex_low = filter_complex_low;
316 s->dsp.filter_simple_high = filter_simple_high;
317 s->dsp.filter_complex_high = filter_complex_high;
318 s->dsp.filter_scale = filter_scale;
320 s->dsp.filter_simple_low = filter16_simple_low;
321 s->dsp.filter_complex_low = filter16_complex_low;
322 s->dsp.filter_simple_high = filter16_simple_high;
323 s->dsp.filter_complex_high = filter16_complex_high;
324 s->dsp.filter_scale = filter16_scale;
328 ff_w3fdif_init_x86(&s->dsp, depth);
333 static int config_output(AVFilterLink *outlink)
335 AVFilterLink *inlink = outlink->src->inputs[0];
337 outlink->time_base.num = inlink->time_base.num;
338 outlink->time_base.den = inlink->time_base.den * 2;
339 outlink->frame_rate.num = inlink->frame_rate.num * 2;
340 outlink->frame_rate.den = inlink->frame_rate.den;
346 * Filter coefficients from PH-2071, scaled by 256 * 128.
347 * Each set of coefficients has a set for low-frequencies and high-frequencies.
348 * n_coef_lf[] and n_coef_hf[] are the number of coefs for simple and more-complex.
349 * It is important for later that n_coef_lf[] is even and n_coef_hf[] is odd.
350 * coef_lf[][] and coef_hf[][] are the coefficients for low-frequencies
351 * and high-frequencies for simple and more-complex mode.
353 static const int8_t n_coef_lf[2] = { 2, 4 };
354 static const int16_t coef_lf[2][4] = {{ 16384, 16384, 0, 0},
355 { -852, 17236, 17236, -852}};
356 static const int8_t n_coef_hf[2] = { 3, 5 };
357 static const int16_t coef_hf[2][5] = {{ -2048, 4096, -2048, 0, 0},
358 { 1016, -3801, 5570, -3801, 1016}};
360 typedef struct ThreadData {
361 AVFrame *out, *cur, *adj;
364 static int deinterlace_plane_slice(AVFilterContext *ctx, void *arg,
365 int jobnr, int nb_jobs, int plane)
367 W3FDIFContext *s = ctx->priv;
368 ThreadData *td = arg;
369 AVFrame *out = td->out;
370 AVFrame *cur = td->cur;
371 AVFrame *adj = td->adj;
372 const int filter = s->filter;
373 uint8_t *in_line, *in_lines_cur[5], *in_lines_adj[5];
374 uint8_t *out_line, *out_pixel;
375 int32_t *work_line, *work_pixel;
376 uint8_t *cur_data = cur->data[plane];
377 uint8_t *adj_data = adj->data[plane];
378 uint8_t *dst_data = out->data[plane];
379 const int linesize = s->linesize[plane];
380 const int height = s->planeheight[plane];
381 const int cur_line_stride = cur->linesize[plane];
382 const int adj_line_stride = adj->linesize[plane];
383 const int dst_line_stride = out->linesize[plane];
384 const int start = (height * jobnr) / nb_jobs;
385 const int end = (height * (jobnr+1)) / nb_jobs;
386 const int max = s->max;
387 const int interlaced = cur->interlaced_frame;
388 const int tff = s->field == (s->parity == -1 ? interlaced ? cur->top_field_first : 1 :
392 /* copy unchanged the lines of the field */
393 y_out = start + (tff ^ (start & 1));
395 in_line = cur_data + (y_out * cur_line_stride);
396 out_line = dst_data + (y_out * dst_line_stride);
398 while (y_out < end) {
399 memcpy(out_line, in_line, linesize);
401 in_line += cur_line_stride * 2;
402 out_line += dst_line_stride * 2;
405 /* interpolate other lines of the field */
406 y_out = start + ((!tff) ^ (start & 1));
408 out_line = dst_data + (y_out * dst_line_stride);
410 while (y_out < end) {
411 /* get low vertical frequencies from current field */
412 for (j = 0; j < n_coef_lf[filter]; j++) {
413 y_in = (y_out + 1) + (j * 2) - n_coef_lf[filter];
417 while (y_in >= height)
420 in_lines_cur[j] = cur_data + (y_in * cur_line_stride);
423 work_line = s->work_line[jobnr];
424 switch (n_coef_lf[filter]) {
426 s->dsp.filter_simple_low(work_line, in_lines_cur,
427 coef_lf[filter], linesize);
430 s->dsp.filter_complex_low(work_line, in_lines_cur,
431 coef_lf[filter], linesize);
434 /* get high vertical frequencies from adjacent fields */
435 for (j = 0; j < n_coef_hf[filter]; j++) {
436 y_in = (y_out + 1) + (j * 2) - n_coef_hf[filter];
440 while (y_in >= height)
443 in_lines_cur[j] = cur_data + (y_in * cur_line_stride);
444 in_lines_adj[j] = adj_data + (y_in * adj_line_stride);
447 work_line = s->work_line[jobnr];
448 switch (n_coef_hf[filter]) {
450 s->dsp.filter_simple_high(work_line, in_lines_cur, in_lines_adj,
451 coef_hf[filter], linesize);
454 s->dsp.filter_complex_high(work_line, in_lines_cur, in_lines_adj,
455 coef_hf[filter], linesize);
458 /* save scaled result to the output frame, scaling down by 256 * 128 */
459 work_pixel = s->work_line[jobnr];
460 out_pixel = out_line;
462 s->dsp.filter_scale(out_pixel, work_pixel, linesize, max);
464 /* move on to next line */
466 out_line += dst_line_stride * 2;
472 static int deinterlace_slice(AVFilterContext *ctx, void *arg,
473 int jobnr, int nb_jobs)
475 W3FDIFContext *s = ctx->priv;
477 for (int p = 0; p < s->nb_planes; p++)
478 deinterlace_plane_slice(ctx, arg, jobnr, nb_jobs, p);
483 static int filter(AVFilterContext *ctx, int is_second)
485 W3FDIFContext *s = ctx->priv;
486 AVFilterLink *outlink = ctx->outputs[0];
490 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
492 return AVERROR(ENOMEM);
493 av_frame_copy_props(out, s->cur);
494 out->interlaced_frame = 0;
497 if (out->pts != AV_NOPTS_VALUE)
500 int64_t cur_pts = s->cur->pts;
501 int64_t next_pts = s->next->pts;
503 if (next_pts != AV_NOPTS_VALUE && cur_pts != AV_NOPTS_VALUE) {
504 out->pts = cur_pts + next_pts;
506 out->pts = AV_NOPTS_VALUE;
510 adj = s->field ? s->next : s->prev;
511 td.out = out; td.cur = s->cur; td.adj = adj;
512 ctx->internal->execute(ctx, deinterlace_slice, &td, NULL, FFMIN(s->planeheight[1], s->nb_threads));
515 s->field = !s->field;
517 return ff_filter_frame(outlink, out);
520 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
522 AVFilterContext *ctx = inlink->dst;
523 W3FDIFContext *s = ctx->priv;
526 av_frame_free(&s->prev);
532 s->cur = av_frame_clone(s->next);
534 return AVERROR(ENOMEM);
537 if ((s->deint && !s->cur->interlaced_frame) || ctx->is_disabled) {
538 AVFrame *out = av_frame_clone(s->cur);
540 return AVERROR(ENOMEM);
542 av_frame_free(&s->prev);
543 if (out->pts != AV_NOPTS_VALUE)
545 return ff_filter_frame(ctx->outputs[0], out);
551 ret = filter(ctx, 0);
552 if (ret < 0 || s->mode == 0)
555 return filter(ctx, 1);
558 static int request_frame(AVFilterLink *outlink)
560 AVFilterContext *ctx = outlink->src;
561 W3FDIFContext *s = ctx->priv;
567 ret = ff_request_frame(ctx->inputs[0]);
569 if (ret == AVERROR_EOF && s->cur) {
570 AVFrame *next = av_frame_clone(s->next);
572 return AVERROR(ENOMEM);
573 next->pts = s->next->pts * 2 - s->cur->pts;
574 filter_frame(ctx->inputs[0], next);
576 } else if (ret < 0) {
583 static av_cold void uninit(AVFilterContext *ctx)
585 W3FDIFContext *s = ctx->priv;
588 av_frame_free(&s->prev);
589 av_frame_free(&s->cur );
590 av_frame_free(&s->next);
592 for (i = 0; i < s->nb_threads; i++)
593 av_freep(&s->work_line[i]);
595 av_freep(&s->work_line);
598 static const AVFilterPad w3fdif_inputs[] = {
601 .type = AVMEDIA_TYPE_VIDEO,
602 .filter_frame = filter_frame,
603 .config_props = config_input,
608 static const AVFilterPad w3fdif_outputs[] = {
611 .type = AVMEDIA_TYPE_VIDEO,
612 .config_props = config_output,
613 .request_frame = request_frame,
618 const AVFilter ff_vf_w3fdif = {
620 .description = NULL_IF_CONFIG_SMALL("Apply Martin Weston three field deinterlace."),
621 .priv_size = sizeof(W3FDIFContext),
622 .priv_class = &w3fdif_class,
624 .query_formats = query_formats,
625 .inputs = w3fdif_inputs,
626 .outputs = w3fdif_outputs,
627 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
628 .process_command = ff_filter_process_command,