blob: 87d4144e9e18f7326fdd3a13b39aa9805cdfc8e3
1 | /* |
2 | * Copyright (C) 2012 Michael Niedermayer <michaelni@gmx.at> |
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 | #include <float.h> /* FLT_MAX */ |
22 | |
23 | #include "libavutil/cpu.h" |
24 | #include "libavutil/common.h" |
25 | #include "libavutil/opt.h" |
26 | #include "internal.h" |
27 | #include "vf_idet.h" |
28 | |
29 | #define OFFSET(x) offsetof(IDETContext, x) |
30 | #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM |
31 | |
32 | static const AVOption idet_options[] = { |
33 | { "intl_thres", "set interlacing threshold", OFFSET(interlace_threshold), AV_OPT_TYPE_FLOAT, {.dbl = 1.04}, -1, FLT_MAX, FLAGS }, |
34 | { "prog_thres", "set progressive threshold", OFFSET(progressive_threshold), AV_OPT_TYPE_FLOAT, {.dbl = 1.5}, -1, FLT_MAX, FLAGS }, |
35 | { "rep_thres", "set repeat threshold", OFFSET(repeat_threshold), AV_OPT_TYPE_FLOAT, {.dbl = 3.0}, -1, FLT_MAX, FLAGS }, |
36 | { "half_life", "half life of cumulative statistics", OFFSET(half_life), AV_OPT_TYPE_FLOAT, {.dbl = 0.0}, -1, INT_MAX, FLAGS }, |
37 | { "analyze_interlaced_flag", "set number of frames to use to determine if the interlace flag is accurate", OFFSET(analyze_interlaced_flag), AV_OPT_TYPE_INT, {.i64 = 0 }, 0, INT_MAX, FLAGS }, |
38 | { NULL } |
39 | }; |
40 | |
41 | AVFILTER_DEFINE_CLASS(idet); |
42 | |
43 | static const char *type2str(Type type) |
44 | { |
45 | switch(type) { |
46 | case TFF : return "tff"; |
47 | case BFF : return "bff"; |
48 | case PROGRESSIVE : return "progressive"; |
49 | case UNDETERMINED : return "undetermined"; |
50 | } |
51 | return NULL; |
52 | } |
53 | |
54 | #define PRECISION 1048576 |
55 | |
56 | static uint64_t uintpow(uint64_t b,unsigned int e) |
57 | { |
58 | uint64_t r=1; |
59 | while(e--) r*=b; |
60 | return r; |
61 | } |
62 | |
63 | static int av_dict_set_fxp(AVDictionary **pm, const char *key, uint64_t value, unsigned int digits, |
64 | int flags) |
65 | { |
66 | char valuestr[44]; |
67 | uint64_t print_precision = uintpow(10, digits); |
68 | |
69 | value = av_rescale(value, print_precision, PRECISION); |
70 | |
71 | snprintf(valuestr, sizeof(valuestr), "%"PRId64".%0*"PRId64, |
72 | value / print_precision, digits, value % print_precision); |
73 | |
74 | return av_dict_set(pm, key, valuestr, flags); |
75 | } |
76 | |
77 | static const char *rep2str(RepeatedField repeated_field) |
78 | { |
79 | switch(repeated_field) { |
80 | case REPEAT_NONE : return "neither"; |
81 | case REPEAT_TOP : return "top"; |
82 | case REPEAT_BOTTOM : return "bottom"; |
83 | } |
84 | return NULL; |
85 | } |
86 | |
87 | int ff_idet_filter_line_c(const uint8_t *a, const uint8_t *b, const uint8_t *c, int w) |
88 | { |
89 | int x; |
90 | int ret=0; |
91 | |
92 | for(x=0; x<w; x++){ |
93 | int v = (*a++ + *c++) - 2 * *b++; |
94 | ret += FFABS(v); |
95 | } |
96 | |
97 | return ret; |
98 | } |
99 | |
100 | int ff_idet_filter_line_c_16bit(const uint16_t *a, const uint16_t *b, const uint16_t *c, int w) |
101 | { |
102 | int x; |
103 | int ret=0; |
104 | |
105 | for(x=0; x<w; x++){ |
106 | int v = (*a++ + *c++) - 2 * *b++; |
107 | ret += FFABS(v); |
108 | } |
109 | |
110 | return ret; |
111 | } |
112 | |
113 | static void filter(AVFilterContext *ctx) |
114 | { |
115 | IDETContext *idet = ctx->priv; |
116 | int y, i; |
117 | int64_t alpha[2]={0}; |
118 | int64_t delta=0; |
119 | int64_t gamma[2]={0}; |
120 | Type type, best_type; |
121 | RepeatedField repeat; |
122 | int match = 0; |
123 | AVDictionary **metadata = avpriv_frame_get_metadatap(idet->cur); |
124 | |
125 | for (i = 0; i < idet->csp->nb_components; i++) { |
126 | int w = idet->cur->width; |
127 | int h = idet->cur->height; |
128 | int refs = idet->cur->linesize[i]; |
129 | |
130 | if (i && i<3) { |
131 | w = AV_CEIL_RSHIFT(w, idet->csp->log2_chroma_w); |
132 | h = AV_CEIL_RSHIFT(h, idet->csp->log2_chroma_h); |
133 | } |
134 | |
135 | for (y = 2; y < h - 2; y++) { |
136 | uint8_t *prev = &idet->prev->data[i][y*refs]; |
137 | uint8_t *cur = &idet->cur ->data[i][y*refs]; |
138 | uint8_t *next = &idet->next->data[i][y*refs]; |
139 | alpha[ y &1] += idet->filter_line(cur-refs, prev, cur+refs, w); |
140 | alpha[(y^1)&1] += idet->filter_line(cur-refs, next, cur+refs, w); |
141 | delta += idet->filter_line(cur-refs, cur, cur+refs, w); |
142 | gamma[(y^1)&1] += idet->filter_line(cur , prev, cur , w); |
143 | } |
144 | } |
145 | |
146 | if (alpha[0] > idet->interlace_threshold * alpha[1]){ |
147 | type = TFF; |
148 | }else if(alpha[1] > idet->interlace_threshold * alpha[0]){ |
149 | type = BFF; |
150 | }else if(alpha[1] > idet->progressive_threshold * delta){ |
151 | type = PROGRESSIVE; |
152 | }else{ |
153 | type = UNDETERMINED; |
154 | } |
155 | |
156 | if ( gamma[0] > idet->repeat_threshold * gamma[1] ){ |
157 | repeat = REPEAT_TOP; |
158 | } else if ( gamma[1] > idet->repeat_threshold * gamma[0] ){ |
159 | repeat = REPEAT_BOTTOM; |
160 | } else { |
161 | repeat = REPEAT_NONE; |
162 | } |
163 | |
164 | memmove(idet->history+1, idet->history, HIST_SIZE-1); |
165 | idet->history[0] = type; |
166 | best_type = UNDETERMINED; |
167 | for(i=0; i<HIST_SIZE; i++){ |
168 | if(idet->history[i] != UNDETERMINED){ |
169 | if(best_type == UNDETERMINED) |
170 | best_type = idet->history[i]; |
171 | |
172 | if(idet->history[i] == best_type) { |
173 | match++; |
174 | }else{ |
175 | match=0; |
176 | break; |
177 | } |
178 | } |
179 | } |
180 | if(idet->last_type == UNDETERMINED){ |
181 | if(match ) idet->last_type = best_type; |
182 | }else{ |
183 | if(match>2) idet->last_type = best_type; |
184 | } |
185 | |
186 | if (idet->last_type == TFF){ |
187 | idet->cur->top_field_first = 1; |
188 | idet->cur->interlaced_frame = 1; |
189 | }else if(idet->last_type == BFF){ |
190 | idet->cur->top_field_first = 0; |
191 | idet->cur->interlaced_frame = 1; |
192 | }else if(idet->last_type == PROGRESSIVE){ |
193 | idet->cur->interlaced_frame = 0; |
194 | } |
195 | |
196 | for(i=0; i<3; i++) |
197 | idet->repeats[i] = av_rescale(idet->repeats [i], idet->decay_coefficient, PRECISION); |
198 | |
199 | for(i=0; i<4; i++){ |
200 | idet->prestat [i] = av_rescale(idet->prestat [i], idet->decay_coefficient, PRECISION); |
201 | idet->poststat[i] = av_rescale(idet->poststat[i], idet->decay_coefficient, PRECISION); |
202 | } |
203 | |
204 | idet->total_repeats [ repeat] ++; |
205 | idet->repeats [ repeat] += PRECISION; |
206 | |
207 | idet->total_prestat [ type] ++; |
208 | idet->prestat [ type] += PRECISION; |
209 | |
210 | idet->total_poststat[idet->last_type] ++; |
211 | idet->poststat [idet->last_type] += PRECISION; |
212 | |
213 | av_log(ctx, AV_LOG_DEBUG, "Repeated Field:%12s, Single frame:%12s, Multi frame:%12s\n", |
214 | rep2str(repeat), type2str(type), type2str(idet->last_type)); |
215 | |
216 | av_dict_set (metadata, "lavfi.idet.repeated.current_frame", rep2str(repeat), 0); |
217 | av_dict_set_fxp(metadata, "lavfi.idet.repeated.neither", idet->repeats[REPEAT_NONE], 2, 0); |
218 | av_dict_set_fxp(metadata, "lavfi.idet.repeated.top", idet->repeats[REPEAT_TOP], 2, 0); |
219 | av_dict_set_fxp(metadata, "lavfi.idet.repeated.bottom", idet->repeats[REPEAT_BOTTOM], 2, 0); |
220 | |
221 | av_dict_set (metadata, "lavfi.idet.single.current_frame", type2str(type), 0); |
222 | av_dict_set_fxp(metadata, "lavfi.idet.single.tff", idet->prestat[TFF], 2 , 0); |
223 | av_dict_set_fxp(metadata, "lavfi.idet.single.bff", idet->prestat[BFF], 2, 0); |
224 | av_dict_set_fxp(metadata, "lavfi.idet.single.progressive", idet->prestat[PROGRESSIVE], 2, 0); |
225 | av_dict_set_fxp(metadata, "lavfi.idet.single.undetermined", idet->prestat[UNDETERMINED], 2, 0); |
226 | |
227 | av_dict_set (metadata, "lavfi.idet.multiple.current_frame", type2str(idet->last_type), 0); |
228 | av_dict_set_fxp(metadata, "lavfi.idet.multiple.tff", idet->poststat[TFF], 2, 0); |
229 | av_dict_set_fxp(metadata, "lavfi.idet.multiple.bff", idet->poststat[BFF], 2, 0); |
230 | av_dict_set_fxp(metadata, "lavfi.idet.multiple.progressive", idet->poststat[PROGRESSIVE], 2, 0); |
231 | av_dict_set_fxp(metadata, "lavfi.idet.multiple.undetermined", idet->poststat[UNDETERMINED], 2, 0); |
232 | } |
233 | |
234 | static int filter_frame(AVFilterLink *link, AVFrame *picref) |
235 | { |
236 | AVFilterContext *ctx = link->dst; |
237 | IDETContext *idet = ctx->priv; |
238 | |
239 | // initial frame(s) and not interlaced, just pass through for |
240 | // the analyze_interlaced_flag mode |
241 | if (idet->analyze_interlaced_flag && |
242 | !picref->interlaced_frame && |
243 | !idet->next) { |
244 | return ff_filter_frame(ctx->outputs[0], picref); |
245 | } |
246 | if (idet->analyze_interlaced_flag_done) { |
247 | if (picref->interlaced_frame && idet->interlaced_flag_accuracy < 0) |
248 | picref->interlaced_frame = 0; |
249 | return ff_filter_frame(ctx->outputs[0], picref); |
250 | } |
251 | |
252 | av_frame_free(&idet->prev); |
253 | |
254 | if( picref->width != link->w |
255 | || picref->height != link->h |
256 | || picref->format != link->format) { |
257 | link->dst->inputs[0]->format = picref->format; |
258 | link->dst->inputs[0]->w = picref->width; |
259 | link->dst->inputs[0]->h = picref->height; |
260 | |
261 | av_frame_free(&idet->cur ); |
262 | av_frame_free(&idet->next); |
263 | } |
264 | |
265 | idet->prev = idet->cur; |
266 | idet->cur = idet->next; |
267 | idet->next = picref; |
268 | |
269 | if (!idet->cur && |
270 | !(idet->cur = av_frame_clone(idet->next))) |
271 | return AVERROR(ENOMEM); |
272 | |
273 | if (!idet->prev) |
274 | return 0; |
275 | |
276 | if (!idet->csp) |
277 | idet->csp = av_pix_fmt_desc_get(link->format); |
278 | if (idet->csp->comp[0].depth > 8){ |
279 | idet->filter_line = (ff_idet_filter_func)ff_idet_filter_line_c_16bit; |
280 | if (ARCH_X86) |
281 | ff_idet_init_x86(idet, 1); |
282 | } |
283 | |
284 | if (idet->analyze_interlaced_flag) { |
285 | if (idet->cur->interlaced_frame) { |
286 | idet->cur->interlaced_frame = 0; |
287 | filter(ctx); |
288 | if (idet->last_type == PROGRESSIVE) { |
289 | idet->interlaced_flag_accuracy --; |
290 | idet->analyze_interlaced_flag --; |
291 | } else if (idet->last_type != UNDETERMINED) { |
292 | idet->interlaced_flag_accuracy ++; |
293 | idet->analyze_interlaced_flag --; |
294 | } |
295 | if (idet->analyze_interlaced_flag == 1) { |
296 | ff_filter_frame(ctx->outputs[0], av_frame_clone(idet->cur)); |
297 | |
298 | if (idet->next->interlaced_frame && idet->interlaced_flag_accuracy < 0) |
299 | idet->next->interlaced_frame = 0; |
300 | idet->analyze_interlaced_flag_done = 1; |
301 | av_log(ctx, AV_LOG_INFO, "Final flag accuracy %d\n", idet->interlaced_flag_accuracy); |
302 | return ff_filter_frame(ctx->outputs[0], av_frame_clone(idet->next)); |
303 | } |
304 | } |
305 | } else { |
306 | filter(ctx); |
307 | } |
308 | |
309 | return ff_filter_frame(ctx->outputs[0], av_frame_clone(idet->cur)); |
310 | } |
311 | |
312 | static int request_frame(AVFilterLink *link) |
313 | { |
314 | AVFilterContext *ctx = link->src; |
315 | IDETContext *idet = ctx->priv; |
316 | int ret; |
317 | |
318 | if (idet->eof) |
319 | return AVERROR_EOF; |
320 | |
321 | ret = ff_request_frame(link->src->inputs[0]); |
322 | |
323 | if (ret == AVERROR_EOF && idet->cur && !idet->analyze_interlaced_flag_done) { |
324 | AVFrame *next = av_frame_clone(idet->next); |
325 | |
326 | if (!next) |
327 | return AVERROR(ENOMEM); |
328 | |
329 | ret = filter_frame(link->src->inputs[0], next); |
330 | idet->eof = 1; |
331 | } |
332 | |
333 | return ret; |
334 | } |
335 | |
336 | static av_cold void uninit(AVFilterContext *ctx) |
337 | { |
338 | IDETContext *idet = ctx->priv; |
339 | int level = strncmp(ctx->name, "auto-inserted", 13) ? AV_LOG_INFO : AV_LOG_DEBUG; |
340 | |
341 | av_log(ctx, level, "Repeated Fields: Neither:%6"PRId64" Top:%6"PRId64" Bottom:%6"PRId64"\n", |
342 | idet->total_repeats[REPEAT_NONE], |
343 | idet->total_repeats[REPEAT_TOP], |
344 | idet->total_repeats[REPEAT_BOTTOM] |
345 | ); |
346 | av_log(ctx, level, "Single frame detection: TFF:%6"PRId64" BFF:%6"PRId64" Progressive:%6"PRId64" Undetermined:%6"PRId64"\n", |
347 | idet->total_prestat[TFF], |
348 | idet->total_prestat[BFF], |
349 | idet->total_prestat[PROGRESSIVE], |
350 | idet->total_prestat[UNDETERMINED] |
351 | ); |
352 | av_log(ctx, level, "Multi frame detection: TFF:%6"PRId64" BFF:%6"PRId64" Progressive:%6"PRId64" Undetermined:%6"PRId64"\n", |
353 | idet->total_poststat[TFF], |
354 | idet->total_poststat[BFF], |
355 | idet->total_poststat[PROGRESSIVE], |
356 | idet->total_poststat[UNDETERMINED] |
357 | ); |
358 | |
359 | av_frame_free(&idet->prev); |
360 | av_frame_free(&idet->cur ); |
361 | av_frame_free(&idet->next); |
362 | } |
363 | |
364 | static int query_formats(AVFilterContext *ctx) |
365 | { |
366 | static const enum AVPixelFormat pix_fmts[] = { |
367 | AV_PIX_FMT_YUV420P, |
368 | AV_PIX_FMT_YUV422P, |
369 | AV_PIX_FMT_YUV444P, |
370 | AV_PIX_FMT_YUV410P, |
371 | AV_PIX_FMT_YUV411P, |
372 | AV_PIX_FMT_GRAY8, |
373 | AV_PIX_FMT_YUVJ420P, |
374 | AV_PIX_FMT_YUVJ422P, |
375 | AV_PIX_FMT_YUVJ444P, |
376 | AV_PIX_FMT_GRAY16, |
377 | AV_PIX_FMT_YUV440P, |
378 | AV_PIX_FMT_YUVJ440P, |
379 | AV_PIX_FMT_YUV420P9, |
380 | AV_PIX_FMT_YUV422P9, |
381 | AV_PIX_FMT_YUV444P9, |
382 | AV_PIX_FMT_YUV420P10, |
383 | AV_PIX_FMT_YUV422P10, |
384 | AV_PIX_FMT_YUV444P10, |
385 | AV_PIX_FMT_YUV420P12, |
386 | AV_PIX_FMT_YUV422P12, |
387 | AV_PIX_FMT_YUV444P12, |
388 | AV_PIX_FMT_YUV420P14, |
389 | AV_PIX_FMT_YUV422P14, |
390 | AV_PIX_FMT_YUV444P14, |
391 | AV_PIX_FMT_YUV420P16, |
392 | AV_PIX_FMT_YUV422P16, |
393 | AV_PIX_FMT_YUV444P16, |
394 | AV_PIX_FMT_YUVA420P, |
395 | AV_PIX_FMT_NONE |
396 | }; |
397 | AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts); |
398 | if (!fmts_list) |
399 | return AVERROR(ENOMEM); |
400 | return ff_set_common_formats(ctx, fmts_list); |
401 | } |
402 | |
403 | static av_cold int init(AVFilterContext *ctx) |
404 | { |
405 | IDETContext *idet = ctx->priv; |
406 | |
407 | idet->eof = 0; |
408 | idet->last_type = UNDETERMINED; |
409 | memset(idet->history, UNDETERMINED, HIST_SIZE); |
410 | |
411 | if( idet->half_life > 0 ) |
412 | idet->decay_coefficient = lrint( PRECISION * exp2(-1.0 / idet->half_life) ); |
413 | else |
414 | idet->decay_coefficient = PRECISION; |
415 | |
416 | idet->filter_line = ff_idet_filter_line_c; |
417 | |
418 | if (ARCH_X86) |
419 | ff_idet_init_x86(idet, 0); |
420 | |
421 | return 0; |
422 | } |
423 | |
424 | static const AVFilterPad idet_inputs[] = { |
425 | { |
426 | .name = "default", |
427 | .type = AVMEDIA_TYPE_VIDEO, |
428 | .filter_frame = filter_frame, |
429 | }, |
430 | { NULL } |
431 | }; |
432 | |
433 | static const AVFilterPad idet_outputs[] = { |
434 | { |
435 | .name = "default", |
436 | .type = AVMEDIA_TYPE_VIDEO, |
437 | .request_frame = request_frame |
438 | }, |
439 | { NULL } |
440 | }; |
441 | |
442 | AVFilter ff_vf_idet = { |
443 | .name = "idet", |
444 | .description = NULL_IF_CONFIG_SMALL("Interlace detect Filter."), |
445 | .priv_size = sizeof(IDETContext), |
446 | .init = init, |
447 | .uninit = uninit, |
448 | .query_formats = query_formats, |
449 | .inputs = idet_inputs, |
450 | .outputs = idet_outputs, |
451 | .priv_class = &idet_class, |
452 | }; |
453 |