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_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
89 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
90 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12,
91 AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
92 AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14,
93 AV_PIX_FMT_YUVA444P9, AV_PIX_FMT_YUVA444P10, AV_PIX_FMT_YUVA444P12, AV_PIX_FMT_YUVA444P16,
94 AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA422P12, AV_PIX_FMT_YUVA422P16,
95 AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA420P16,
96 AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
100 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
102 return AVERROR(ENOMEM);
103 return ff_set_common_formats(ctx, fmts_list);
106 static void filter_simple_low(int32_t *work_line,
107 uint8_t *in_lines_cur[2],
108 const int16_t *coef, int linesize)
112 for (i = 0; i < linesize; i++) {
113 *work_line = *in_lines_cur[0]++ * coef[0];
114 *work_line++ += *in_lines_cur[1]++ * coef[1];
118 static void filter_complex_low(int32_t *work_line,
119 uint8_t *in_lines_cur[4],
120 const int16_t *coef, int linesize)
124 for (i = 0; i < linesize; i++) {
125 *work_line = *in_lines_cur[0]++ * coef[0];
126 *work_line += *in_lines_cur[1]++ * coef[1];
127 *work_line += *in_lines_cur[2]++ * coef[2];
128 *work_line++ += *in_lines_cur[3]++ * coef[3];
132 static void filter_simple_high(int32_t *work_line,
133 uint8_t *in_lines_cur[3],
134 uint8_t *in_lines_adj[3],
135 const int16_t *coef, int linesize)
139 for (i = 0; i < linesize; i++) {
140 *work_line += *in_lines_cur[0]++ * coef[0];
141 *work_line += *in_lines_adj[0]++ * coef[0];
142 *work_line += *in_lines_cur[1]++ * coef[1];
143 *work_line += *in_lines_adj[1]++ * coef[1];
144 *work_line += *in_lines_cur[2]++ * coef[2];
145 *work_line++ += *in_lines_adj[2]++ * coef[2];
149 static void filter_complex_high(int32_t *work_line,
150 uint8_t *in_lines_cur[5],
151 uint8_t *in_lines_adj[5],
152 const int16_t *coef, int linesize)
156 for (i = 0; i < linesize; i++) {
157 *work_line += *in_lines_cur[0]++ * coef[0];
158 *work_line += *in_lines_adj[0]++ * coef[0];
159 *work_line += *in_lines_cur[1]++ * coef[1];
160 *work_line += *in_lines_adj[1]++ * coef[1];
161 *work_line += *in_lines_cur[2]++ * coef[2];
162 *work_line += *in_lines_adj[2]++ * coef[2];
163 *work_line += *in_lines_cur[3]++ * coef[3];
164 *work_line += *in_lines_adj[3]++ * coef[3];
165 *work_line += *in_lines_cur[4]++ * coef[4];
166 *work_line++ += *in_lines_adj[4]++ * coef[4];
170 static void filter_scale(uint8_t *out_pixel, const int32_t *work_pixel, int linesize, int max)
174 for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
175 *out_pixel = av_clip(*work_pixel, 0, 255 * 256 * 128) >> 15;
178 static void filter16_simple_low(int32_t *work_line,
179 uint8_t *in_lines_cur8[2],
180 const int16_t *coef, int linesize)
182 uint16_t *in_lines_cur[2] = { (uint16_t *)in_lines_cur8[0], (uint16_t *)in_lines_cur8[1] };
186 for (i = 0; i < linesize; i++) {
187 *work_line = *in_lines_cur[0]++ * coef[0];
188 *work_line++ += *in_lines_cur[1]++ * coef[1];
192 static void filter16_complex_low(int32_t *work_line,
193 uint8_t *in_lines_cur8[4],
194 const int16_t *coef, int linesize)
196 uint16_t *in_lines_cur[4] = { (uint16_t *)in_lines_cur8[0],
197 (uint16_t *)in_lines_cur8[1],
198 (uint16_t *)in_lines_cur8[2],
199 (uint16_t *)in_lines_cur8[3] };
203 for (i = 0; i < linesize; i++) {
204 *work_line = *in_lines_cur[0]++ * coef[0];
205 *work_line += *in_lines_cur[1]++ * coef[1];
206 *work_line += *in_lines_cur[2]++ * coef[2];
207 *work_line++ += *in_lines_cur[3]++ * coef[3];
211 static void filter16_simple_high(int32_t *work_line,
212 uint8_t *in_lines_cur8[3],
213 uint8_t *in_lines_adj8[3],
214 const int16_t *coef, int linesize)
216 uint16_t *in_lines_cur[3] = { (uint16_t *)in_lines_cur8[0],
217 (uint16_t *)in_lines_cur8[1],
218 (uint16_t *)in_lines_cur8[2] };
219 uint16_t *in_lines_adj[3] = { (uint16_t *)in_lines_adj8[0],
220 (uint16_t *)in_lines_adj8[1],
221 (uint16_t *)in_lines_adj8[2] };
225 for (i = 0; i < linesize; i++) {
226 *work_line += *in_lines_cur[0]++ * coef[0];
227 *work_line += *in_lines_adj[0]++ * coef[0];
228 *work_line += *in_lines_cur[1]++ * coef[1];
229 *work_line += *in_lines_adj[1]++ * coef[1];
230 *work_line += *in_lines_cur[2]++ * coef[2];
231 *work_line++ += *in_lines_adj[2]++ * coef[2];
235 static void filter16_complex_high(int32_t *work_line,
236 uint8_t *in_lines_cur8[5],
237 uint8_t *in_lines_adj8[5],
238 const int16_t *coef, int linesize)
240 uint16_t *in_lines_cur[5] = { (uint16_t *)in_lines_cur8[0],
241 (uint16_t *)in_lines_cur8[1],
242 (uint16_t *)in_lines_cur8[2],
243 (uint16_t *)in_lines_cur8[3],
244 (uint16_t *)in_lines_cur8[4] };
245 uint16_t *in_lines_adj[5] = { (uint16_t *)in_lines_adj8[0],
246 (uint16_t *)in_lines_adj8[1],
247 (uint16_t *)in_lines_adj8[2],
248 (uint16_t *)in_lines_adj8[3],
249 (uint16_t *)in_lines_adj8[4] };
253 for (i = 0; i < linesize; i++) {
254 *work_line += *in_lines_cur[0]++ * coef[0];
255 *work_line += *in_lines_adj[0]++ * coef[0];
256 *work_line += *in_lines_cur[1]++ * coef[1];
257 *work_line += *in_lines_adj[1]++ * coef[1];
258 *work_line += *in_lines_cur[2]++ * coef[2];
259 *work_line += *in_lines_adj[2]++ * coef[2];
260 *work_line += *in_lines_cur[3]++ * coef[3];
261 *work_line += *in_lines_adj[3]++ * coef[3];
262 *work_line += *in_lines_cur[4]++ * coef[4];
263 *work_line++ += *in_lines_adj[4]++ * coef[4];
267 static void filter16_scale(uint8_t *out_pixel8, const int32_t *work_pixel, int linesize, int max)
269 uint16_t *out_pixel = (uint16_t *)out_pixel8;
273 for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
274 *out_pixel = av_clip(*work_pixel, 0, max) >> 15;
277 static int config_input(AVFilterLink *inlink)
279 AVFilterContext *ctx = inlink->dst;
280 W3FDIFContext *s = ctx->priv;
281 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
284 if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
287 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
288 s->planeheight[0] = s->planeheight[3] = inlink->h;
291 av_log(ctx, AV_LOG_ERROR, "Video of less than 3 lines is not supported\n");
292 return AVERROR(EINVAL);
295 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
296 s->nb_threads = ff_filter_get_nb_threads(ctx);
297 s->work_line = av_calloc(s->nb_threads, sizeof(*s->work_line));
299 return AVERROR(ENOMEM);
301 for (i = 0; i < s->nb_threads; i++) {
302 s->work_line[i] = av_calloc(FFALIGN(s->linesize[0], 32), sizeof(*s->work_line[0]));
303 if (!s->work_line[i])
304 return AVERROR(ENOMEM);
307 depth = desc->comp[0].depth;
308 s->max = ((1 << depth) - 1) * 256 * 128;
310 s->dsp.filter_simple_low = filter_simple_low;
311 s->dsp.filter_complex_low = filter_complex_low;
312 s->dsp.filter_simple_high = filter_simple_high;
313 s->dsp.filter_complex_high = filter_complex_high;
314 s->dsp.filter_scale = filter_scale;
316 s->dsp.filter_simple_low = filter16_simple_low;
317 s->dsp.filter_complex_low = filter16_complex_low;
318 s->dsp.filter_simple_high = filter16_simple_high;
319 s->dsp.filter_complex_high = filter16_complex_high;
320 s->dsp.filter_scale = filter16_scale;
324 ff_w3fdif_init_x86(&s->dsp, depth);
329 static int config_output(AVFilterLink *outlink)
331 AVFilterLink *inlink = outlink->src->inputs[0];
333 outlink->time_base.num = inlink->time_base.num;
334 outlink->time_base.den = inlink->time_base.den * 2;
335 outlink->frame_rate.num = inlink->frame_rate.num * 2;
336 outlink->frame_rate.den = inlink->frame_rate.den;
342 * Filter coefficients from PH-2071, scaled by 256 * 128.
343 * Each set of coefficients has a set for low-frequencies and high-frequencies.
344 * n_coef_lf[] and n_coef_hf[] are the number of coefs for simple and more-complex.
345 * It is important for later that n_coef_lf[] is even and n_coef_hf[] is odd.
346 * coef_lf[][] and coef_hf[][] are the coefficients for low-frequencies
347 * and high-frequencies for simple and more-complex mode.
349 static const int8_t n_coef_lf[2] = { 2, 4 };
350 static const int16_t coef_lf[2][4] = {{ 16384, 16384, 0, 0},
351 { -852, 17236, 17236, -852}};
352 static const int8_t n_coef_hf[2] = { 3, 5 };
353 static const int16_t coef_hf[2][5] = {{ -2048, 4096, -2048, 0, 0},
354 { 1016, -3801, 5570, -3801, 1016}};
356 typedef struct ThreadData {
357 AVFrame *out, *cur, *adj;
361 static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
363 W3FDIFContext *s = ctx->priv;
364 ThreadData *td = arg;
365 AVFrame *out = td->out;
366 AVFrame *cur = td->cur;
367 AVFrame *adj = td->adj;
368 const int plane = td->plane;
369 const int filter = s->filter;
370 uint8_t *in_line, *in_lines_cur[5], *in_lines_adj[5];
371 uint8_t *out_line, *out_pixel;
372 int32_t *work_line, *work_pixel;
373 uint8_t *cur_data = cur->data[plane];
374 uint8_t *adj_data = adj->data[plane];
375 uint8_t *dst_data = out->data[plane];
376 const int linesize = s->linesize[plane];
377 const int height = s->planeheight[plane];
378 const int cur_line_stride = cur->linesize[plane];
379 const int adj_line_stride = adj->linesize[plane];
380 const int dst_line_stride = out->linesize[plane];
381 const int start = (height * jobnr) / nb_jobs;
382 const int end = (height * (jobnr+1)) / nb_jobs;
383 const int max = s->max;
384 const int tff = (s->field == (s->parity == -1 ? cur->top_field_first == cur->interlaced_frame : s->parity == 0 ? !cur->interlaced_frame : cur->interlaced_frame));
387 /* copy unchanged the lines of the field */
388 y_out = start + (tff ^ (start & 1));
390 in_line = cur_data + (y_out * cur_line_stride);
391 out_line = dst_data + (y_out * dst_line_stride);
393 while (y_out < end) {
394 memcpy(out_line, in_line, linesize);
396 in_line += cur_line_stride * 2;
397 out_line += dst_line_stride * 2;
400 /* interpolate other lines of the field */
401 y_out = start + ((!tff) ^ (start & 1));
403 out_line = dst_data + (y_out * dst_line_stride);
405 while (y_out < end) {
406 /* get low vertical frequencies from current field */
407 for (j = 0; j < n_coef_lf[filter]; j++) {
408 y_in = (y_out + 1) + (j * 2) - n_coef_lf[filter];
412 while (y_in >= height)
415 in_lines_cur[j] = cur_data + (y_in * cur_line_stride);
418 work_line = s->work_line[jobnr];
419 switch (n_coef_lf[filter]) {
421 s->dsp.filter_simple_low(work_line, in_lines_cur,
422 coef_lf[filter], linesize);
425 s->dsp.filter_complex_low(work_line, in_lines_cur,
426 coef_lf[filter], linesize);
429 /* get high vertical frequencies from adjacent fields */
430 for (j = 0; j < n_coef_hf[filter]; j++) {
431 y_in = (y_out + 1) + (j * 2) - n_coef_hf[filter];
435 while (y_in >= height)
438 in_lines_cur[j] = cur_data + (y_in * cur_line_stride);
439 in_lines_adj[j] = adj_data + (y_in * adj_line_stride);
442 work_line = s->work_line[jobnr];
443 switch (n_coef_hf[filter]) {
445 s->dsp.filter_simple_high(work_line, in_lines_cur, in_lines_adj,
446 coef_hf[filter], linesize);
449 s->dsp.filter_complex_high(work_line, in_lines_cur, in_lines_adj,
450 coef_hf[filter], linesize);
453 /* save scaled result to the output frame, scaling down by 256 * 128 */
454 work_pixel = s->work_line[jobnr];
455 out_pixel = out_line;
457 s->dsp.filter_scale(out_pixel, work_pixel, linesize, max);
459 /* move on to next line */
461 out_line += dst_line_stride * 2;
467 static int filter(AVFilterContext *ctx, int is_second)
469 W3FDIFContext *s = ctx->priv;
470 AVFilterLink *outlink = ctx->outputs[0];
475 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
477 return AVERROR(ENOMEM);
478 av_frame_copy_props(out, s->cur);
479 out->interlaced_frame = 0;
482 if (out->pts != AV_NOPTS_VALUE)
485 int64_t cur_pts = s->cur->pts;
486 int64_t next_pts = s->next->pts;
488 if (next_pts != AV_NOPTS_VALUE && cur_pts != AV_NOPTS_VALUE) {
489 out->pts = cur_pts + next_pts;
491 out->pts = AV_NOPTS_VALUE;
495 adj = s->field ? s->next : s->prev;
496 td.out = out; td.cur = s->cur; td.adj = adj;
497 for (plane = 0; plane < s->nb_planes; plane++) {
499 ctx->internal->execute(ctx, deinterlace_slice, &td, NULL, FFMIN(s->planeheight[plane], s->nb_threads));
503 s->field = !s->field;
505 return ff_filter_frame(outlink, out);
508 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
510 AVFilterContext *ctx = inlink->dst;
511 W3FDIFContext *s = ctx->priv;
514 av_frame_free(&s->prev);
520 s->cur = av_frame_clone(s->next);
522 return AVERROR(ENOMEM);
525 if ((s->deint && !s->cur->interlaced_frame) || ctx->is_disabled) {
526 AVFrame *out = av_frame_clone(s->cur);
528 return AVERROR(ENOMEM);
530 av_frame_free(&s->prev);
531 if (out->pts != AV_NOPTS_VALUE)
533 return ff_filter_frame(ctx->outputs[0], out);
539 ret = filter(ctx, 0);
540 if (ret < 0 || s->mode == 0)
543 return filter(ctx, 1);
546 static int request_frame(AVFilterLink *outlink)
548 AVFilterContext *ctx = outlink->src;
549 W3FDIFContext *s = ctx->priv;
555 ret = ff_request_frame(ctx->inputs[0]);
557 if (ret == AVERROR_EOF && s->cur) {
558 AVFrame *next = av_frame_clone(s->next);
560 return AVERROR(ENOMEM);
561 next->pts = s->next->pts * 2 - s->cur->pts;
562 filter_frame(ctx->inputs[0], next);
564 } else if (ret < 0) {
571 static av_cold void uninit(AVFilterContext *ctx)
573 W3FDIFContext *s = ctx->priv;
576 av_frame_free(&s->prev);
577 av_frame_free(&s->cur );
578 av_frame_free(&s->next);
580 for (i = 0; i < s->nb_threads; i++)
581 av_freep(&s->work_line[i]);
583 av_freep(&s->work_line);
586 static const AVFilterPad w3fdif_inputs[] = {
589 .type = AVMEDIA_TYPE_VIDEO,
590 .filter_frame = filter_frame,
591 .config_props = config_input,
596 static const AVFilterPad w3fdif_outputs[] = {
599 .type = AVMEDIA_TYPE_VIDEO,
600 .config_props = config_output,
601 .request_frame = request_frame,
606 AVFilter ff_vf_w3fdif = {
608 .description = NULL_IF_CONFIG_SMALL("Apply Martin Weston three field deinterlace."),
609 .priv_size = sizeof(W3FDIFContext),
610 .priv_class = &w3fdif_class,
612 .query_formats = query_formats,
613 .inputs = w3fdif_inputs,
614 .outputs = w3fdif_outputs,
615 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
616 .process_command = ff_filter_process_command,