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 deint; ///< which frames to deinterlace
38 int linesize[4]; ///< bytes of pixel data per line for each plane
39 int planeheight[4]; ///< height of each plane
40 int field; ///< which field are we on, 0 or 1
43 AVFrame *prev, *cur, *next; ///< previous, current, next frames
44 int32_t **work_line; ///< lines we are calculating
51 #define OFFSET(x) offsetof(W3FDIFContext, x)
52 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
53 #define CONST(name, help, val, unit) { name, help, 0, AV_OPT_TYPE_CONST, {.i64=val}, 0, 0, FLAGS, unit }
55 static const AVOption w3fdif_options[] = {
56 { "filter", "specify the filter", OFFSET(filter), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, FLAGS, "filter" },
57 CONST("simple", NULL, 0, "filter"),
58 CONST("complex", NULL, 1, "filter"),
59 { "deint", "specify which frames to deinterlace", OFFSET(deint), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, FLAGS, "deint" },
60 CONST("all", "deinterlace all frames", 0, "deint"),
61 CONST("interlaced", "only deinterlace frames marked as interlaced", 1, "deint"),
65 AVFILTER_DEFINE_CLASS(w3fdif);
67 static int query_formats(AVFilterContext *ctx)
69 static const enum AVPixelFormat pix_fmts[] = {
70 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P,
71 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P,
72 AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV444P,
73 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
74 AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
76 AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA444P,
77 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP,
79 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
80 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
81 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12,
82 AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
83 AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14,
87 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
89 return AVERROR(ENOMEM);
90 return ff_set_common_formats(ctx, fmts_list);
93 static void filter_simple_low(int32_t *work_line,
94 uint8_t *in_lines_cur[2],
95 const int16_t *coef, int linesize)
99 for (i = 0; i < linesize; i++) {
100 *work_line = *in_lines_cur[0]++ * coef[0];
101 *work_line++ += *in_lines_cur[1]++ * coef[1];
105 static void filter_complex_low(int32_t *work_line,
106 uint8_t *in_lines_cur[4],
107 const int16_t *coef, int linesize)
111 for (i = 0; i < linesize; i++) {
112 *work_line = *in_lines_cur[0]++ * coef[0];
113 *work_line += *in_lines_cur[1]++ * coef[1];
114 *work_line += *in_lines_cur[2]++ * coef[2];
115 *work_line++ += *in_lines_cur[3]++ * coef[3];
119 static void filter_simple_high(int32_t *work_line,
120 uint8_t *in_lines_cur[3],
121 uint8_t *in_lines_adj[3],
122 const int16_t *coef, int linesize)
126 for (i = 0; i < linesize; i++) {
127 *work_line += *in_lines_cur[0]++ * coef[0];
128 *work_line += *in_lines_adj[0]++ * coef[0];
129 *work_line += *in_lines_cur[1]++ * coef[1];
130 *work_line += *in_lines_adj[1]++ * coef[1];
131 *work_line += *in_lines_cur[2]++ * coef[2];
132 *work_line++ += *in_lines_adj[2]++ * coef[2];
136 static void filter_complex_high(int32_t *work_line,
137 uint8_t *in_lines_cur[5],
138 uint8_t *in_lines_adj[5],
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];
150 *work_line += *in_lines_cur[3]++ * coef[3];
151 *work_line += *in_lines_adj[3]++ * coef[3];
152 *work_line += *in_lines_cur[4]++ * coef[4];
153 *work_line++ += *in_lines_adj[4]++ * coef[4];
157 static void filter_scale(uint8_t *out_pixel, const int32_t *work_pixel, int linesize, int max)
161 for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
162 *out_pixel = av_clip(*work_pixel, 0, 255 * 256 * 128) >> 15;
165 static void filter16_simple_low(int32_t *work_line,
166 uint8_t *in_lines_cur8[2],
167 const int16_t *coef, int linesize)
169 uint16_t *in_lines_cur[2] = { (uint16_t *)in_lines_cur8[0], (uint16_t *)in_lines_cur8[1] };
173 for (i = 0; i < linesize; i++) {
174 *work_line = *in_lines_cur[0]++ * coef[0];
175 *work_line++ += *in_lines_cur[1]++ * coef[1];
179 static void filter16_complex_low(int32_t *work_line,
180 uint8_t *in_lines_cur8[4],
181 const int16_t *coef, int linesize)
183 uint16_t *in_lines_cur[4] = { (uint16_t *)in_lines_cur8[0],
184 (uint16_t *)in_lines_cur8[1],
185 (uint16_t *)in_lines_cur8[2],
186 (uint16_t *)in_lines_cur8[3] };
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];
193 *work_line += *in_lines_cur[2]++ * coef[2];
194 *work_line++ += *in_lines_cur[3]++ * coef[3];
198 static void filter16_simple_high(int32_t *work_line,
199 uint8_t *in_lines_cur8[3],
200 uint8_t *in_lines_adj8[3],
201 const int16_t *coef, int linesize)
203 uint16_t *in_lines_cur[3] = { (uint16_t *)in_lines_cur8[0],
204 (uint16_t *)in_lines_cur8[1],
205 (uint16_t *)in_lines_cur8[2] };
206 uint16_t *in_lines_adj[3] = { (uint16_t *)in_lines_adj8[0],
207 (uint16_t *)in_lines_adj8[1],
208 (uint16_t *)in_lines_adj8[2] };
212 for (i = 0; i < linesize; i++) {
213 *work_line += *in_lines_cur[0]++ * coef[0];
214 *work_line += *in_lines_adj[0]++ * coef[0];
215 *work_line += *in_lines_cur[1]++ * coef[1];
216 *work_line += *in_lines_adj[1]++ * coef[1];
217 *work_line += *in_lines_cur[2]++ * coef[2];
218 *work_line++ += *in_lines_adj[2]++ * coef[2];
222 static void filter16_complex_high(int32_t *work_line,
223 uint8_t *in_lines_cur8[5],
224 uint8_t *in_lines_adj8[5],
225 const int16_t *coef, int linesize)
227 uint16_t *in_lines_cur[5] = { (uint16_t *)in_lines_cur8[0],
228 (uint16_t *)in_lines_cur8[1],
229 (uint16_t *)in_lines_cur8[2],
230 (uint16_t *)in_lines_cur8[3],
231 (uint16_t *)in_lines_cur8[4] };
232 uint16_t *in_lines_adj[5] = { (uint16_t *)in_lines_adj8[0],
233 (uint16_t *)in_lines_adj8[1],
234 (uint16_t *)in_lines_adj8[2],
235 (uint16_t *)in_lines_adj8[3],
236 (uint16_t *)in_lines_adj8[4] };
240 for (i = 0; i < linesize; i++) {
241 *work_line += *in_lines_cur[0]++ * coef[0];
242 *work_line += *in_lines_adj[0]++ * coef[0];
243 *work_line += *in_lines_cur[1]++ * coef[1];
244 *work_line += *in_lines_adj[1]++ * coef[1];
245 *work_line += *in_lines_cur[2]++ * coef[2];
246 *work_line += *in_lines_adj[2]++ * coef[2];
247 *work_line += *in_lines_cur[3]++ * coef[3];
248 *work_line += *in_lines_adj[3]++ * coef[3];
249 *work_line += *in_lines_cur[4]++ * coef[4];
250 *work_line++ += *in_lines_adj[4]++ * coef[4];
254 static void filter16_scale(uint8_t *out_pixel8, const int32_t *work_pixel, int linesize, int max)
256 uint16_t *out_pixel = (uint16_t *)out_pixel8;
260 for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
261 *out_pixel = av_clip(*work_pixel, 0, max) >> 15;
264 static int config_input(AVFilterLink *inlink)
266 AVFilterContext *ctx = inlink->dst;
267 W3FDIFContext *s = ctx->priv;
268 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
271 if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
274 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
275 s->planeheight[0] = s->planeheight[3] = inlink->h;
278 av_log(ctx, AV_LOG_ERROR, "Video of less than 3 lines is not supported\n");
279 return AVERROR(EINVAL);
282 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
283 s->nb_threads = ff_filter_get_nb_threads(ctx);
284 s->work_line = av_calloc(s->nb_threads, sizeof(*s->work_line));
286 return AVERROR(ENOMEM);
288 for (i = 0; i < s->nb_threads; i++) {
289 s->work_line[i] = av_calloc(FFALIGN(s->linesize[0], 32), sizeof(*s->work_line[0]));
290 if (!s->work_line[i])
291 return AVERROR(ENOMEM);
294 depth = desc->comp[0].depth;
295 s->max = ((1 << depth) - 1) * 256 * 128;
297 s->dsp.filter_simple_low = filter_simple_low;
298 s->dsp.filter_complex_low = filter_complex_low;
299 s->dsp.filter_simple_high = filter_simple_high;
300 s->dsp.filter_complex_high = filter_complex_high;
301 s->dsp.filter_scale = filter_scale;
303 s->dsp.filter_simple_low = filter16_simple_low;
304 s->dsp.filter_complex_low = filter16_complex_low;
305 s->dsp.filter_simple_high = filter16_simple_high;
306 s->dsp.filter_complex_high = filter16_complex_high;
307 s->dsp.filter_scale = filter16_scale;
311 ff_w3fdif_init_x86(&s->dsp, depth);
316 static int config_output(AVFilterLink *outlink)
318 AVFilterLink *inlink = outlink->src->inputs[0];
320 outlink->time_base.num = inlink->time_base.num;
321 outlink->time_base.den = inlink->time_base.den * 2;
322 outlink->frame_rate.num = inlink->frame_rate.num * 2;
323 outlink->frame_rate.den = inlink->frame_rate.den;
329 * Filter coefficients from PH-2071, scaled by 256 * 128.
330 * Each set of coefficients has a set for low-frequencies and high-frequencies.
331 * n_coef_lf[] and n_coef_hf[] are the number of coefs for simple and more-complex.
332 * It is important for later that n_coef_lf[] is even and n_coef_hf[] is odd.
333 * coef_lf[][] and coef_hf[][] are the coefficients for low-frequencies
334 * and high-frequencies for simple and more-complex mode.
336 static const int8_t n_coef_lf[2] = { 2, 4 };
337 static const int16_t coef_lf[2][4] = {{ 16384, 16384, 0, 0},
338 { -852, 17236, 17236, -852}};
339 static const int8_t n_coef_hf[2] = { 3, 5 };
340 static const int16_t coef_hf[2][5] = {{ -2048, 4096, -2048, 0, 0},
341 { 1016, -3801, 5570, -3801, 1016}};
343 typedef struct ThreadData {
344 AVFrame *out, *cur, *adj;
348 static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
350 W3FDIFContext *s = ctx->priv;
351 ThreadData *td = arg;
352 AVFrame *out = td->out;
353 AVFrame *cur = td->cur;
354 AVFrame *adj = td->adj;
355 const int plane = td->plane;
356 const int filter = s->filter;
357 uint8_t *in_line, *in_lines_cur[5], *in_lines_adj[5];
358 uint8_t *out_line, *out_pixel;
359 int32_t *work_line, *work_pixel;
360 uint8_t *cur_data = cur->data[plane];
361 uint8_t *adj_data = adj->data[plane];
362 uint8_t *dst_data = out->data[plane];
363 const int linesize = s->linesize[plane];
364 const int height = s->planeheight[plane];
365 const int cur_line_stride = cur->linesize[plane];
366 const int adj_line_stride = adj->linesize[plane];
367 const int dst_line_stride = out->linesize[plane];
368 const int start = (height * jobnr) / nb_jobs;
369 const int end = (height * (jobnr+1)) / nb_jobs;
370 const int max = s->max;
373 /* copy unchanged the lines of the field */
374 y_out = start + ((s->field == cur->top_field_first) ^ (start & 1));
376 in_line = cur_data + (y_out * cur_line_stride);
377 out_line = dst_data + (y_out * dst_line_stride);
379 while (y_out < end) {
380 memcpy(out_line, in_line, linesize);
382 in_line += cur_line_stride * 2;
383 out_line += dst_line_stride * 2;
386 /* interpolate other lines of the field */
387 y_out = start + ((s->field != cur->top_field_first) ^ (start & 1));
389 out_line = dst_data + (y_out * dst_line_stride);
391 while (y_out < end) {
392 /* get low vertical frequencies from current field */
393 for (j = 0; j < n_coef_lf[filter]; j++) {
394 y_in = (y_out + 1) + (j * 2) - n_coef_lf[filter];
398 while (y_in >= height)
401 in_lines_cur[j] = cur_data + (y_in * cur_line_stride);
404 work_line = s->work_line[jobnr];
405 switch (n_coef_lf[filter]) {
407 s->dsp.filter_simple_low(work_line, in_lines_cur,
408 coef_lf[filter], linesize);
411 s->dsp.filter_complex_low(work_line, in_lines_cur,
412 coef_lf[filter], linesize);
415 /* get high vertical frequencies from adjacent fields */
416 for (j = 0; j < n_coef_hf[filter]; j++) {
417 y_in = (y_out + 1) + (j * 2) - n_coef_hf[filter];
421 while (y_in >= height)
424 in_lines_cur[j] = cur_data + (y_in * cur_line_stride);
425 in_lines_adj[j] = adj_data + (y_in * adj_line_stride);
428 work_line = s->work_line[jobnr];
429 switch (n_coef_hf[filter]) {
431 s->dsp.filter_simple_high(work_line, in_lines_cur, in_lines_adj,
432 coef_hf[filter], linesize);
435 s->dsp.filter_complex_high(work_line, in_lines_cur, in_lines_adj,
436 coef_hf[filter], linesize);
439 /* save scaled result to the output frame, scaling down by 256 * 128 */
440 work_pixel = s->work_line[jobnr];
441 out_pixel = out_line;
443 s->dsp.filter_scale(out_pixel, work_pixel, linesize, max);
445 /* move on to next line */
447 out_line += dst_line_stride * 2;
453 static int filter(AVFilterContext *ctx, int is_second)
455 W3FDIFContext *s = ctx->priv;
456 AVFilterLink *outlink = ctx->outputs[0];
461 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
463 return AVERROR(ENOMEM);
464 av_frame_copy_props(out, s->cur);
465 out->interlaced_frame = 0;
468 if (out->pts != AV_NOPTS_VALUE)
471 int64_t cur_pts = s->cur->pts;
472 int64_t next_pts = s->next->pts;
474 if (next_pts != AV_NOPTS_VALUE && cur_pts != AV_NOPTS_VALUE) {
475 out->pts = cur_pts + next_pts;
477 out->pts = AV_NOPTS_VALUE;
481 adj = s->field ? s->next : s->prev;
482 td.out = out; td.cur = s->cur; td.adj = adj;
483 for (plane = 0; plane < s->nb_planes; plane++) {
485 ctx->internal->execute(ctx, deinterlace_slice, &td, NULL, FFMIN(s->planeheight[plane], s->nb_threads));
488 s->field = !s->field;
490 return ff_filter_frame(outlink, out);
493 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
495 AVFilterContext *ctx = inlink->dst;
496 W3FDIFContext *s = ctx->priv;
499 av_frame_free(&s->prev);
505 s->cur = av_frame_clone(s->next);
507 return AVERROR(ENOMEM);
510 if ((s->deint && !s->cur->interlaced_frame) || ctx->is_disabled) {
511 AVFrame *out = av_frame_clone(s->cur);
513 return AVERROR(ENOMEM);
515 av_frame_free(&s->prev);
516 if (out->pts != AV_NOPTS_VALUE)
518 return ff_filter_frame(ctx->outputs[0], out);
524 ret = filter(ctx, 0);
528 return filter(ctx, 1);
531 static int request_frame(AVFilterLink *outlink)
533 AVFilterContext *ctx = outlink->src;
534 W3FDIFContext *s = ctx->priv;
540 ret = ff_request_frame(ctx->inputs[0]);
542 if (ret == AVERROR_EOF && s->cur) {
543 AVFrame *next = av_frame_clone(s->next);
545 return AVERROR(ENOMEM);
546 next->pts = s->next->pts * 2 - s->cur->pts;
547 filter_frame(ctx->inputs[0], next);
549 } else if (ret < 0) {
556 static av_cold void uninit(AVFilterContext *ctx)
558 W3FDIFContext *s = ctx->priv;
561 av_frame_free(&s->prev);
562 av_frame_free(&s->cur );
563 av_frame_free(&s->next);
565 for (i = 0; i < s->nb_threads; i++)
566 av_freep(&s->work_line[i]);
568 av_freep(&s->work_line);
571 static const AVFilterPad w3fdif_inputs[] = {
574 .type = AVMEDIA_TYPE_VIDEO,
575 .filter_frame = filter_frame,
576 .config_props = config_input,
581 static const AVFilterPad w3fdif_outputs[] = {
584 .type = AVMEDIA_TYPE_VIDEO,
585 .config_props = config_output,
586 .request_frame = request_frame,
591 AVFilter ff_vf_w3fdif = {
593 .description = NULL_IF_CONFIG_SMALL("Apply Martin Weston three field deinterlace."),
594 .priv_size = sizeof(W3FDIFContext),
595 .priv_class = &w3fdif_class,
597 .query_formats = query_formats,
598 .inputs = w3fdif_inputs,
599 .outputs = w3fdif_outputs,
600 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,