FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
vf_xbr.c
Go to the documentation of this file.
1 /*
2  * This file is part of FFmpeg.
3  *
4  * Copyright (c) 2011, 2012 Hyllian/Jararaca <sergiogdb@gmail.com>
5  * Copyright (c) 2014 Arwa Arif <arwaarif1994@gmail.com>
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 /**
23  * @file
24  * XBR Filter is used for depixelization of image.
25  * This is based on Hyllian's xBR shader.
26  *
27  * @see http://www.libretro.com/forums/viewtopic.php?f=6&t=134
28  * @see https://github.com/yoyofr/iFBA/blob/master/fba_src/src/intf/video/scalers/xbr.cpp
29  */
30 
31 #include "libavutil/opt.h"
32 #include "libavutil/avassert.h"
33 #include "libavutil/pixdesc.h"
34 #include "internal.h"
35 
36 #define LB_MASK 0x00FEFEFE
37 #define RED_BLUE_MASK 0x00FF00FF
38 #define GREEN_MASK 0x0000FF00
39 
40 typedef int (*xbrfunc_t)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
41 
42 typedef struct {
43  const AVClass *class;
44  int n;
45  xbrfunc_t func;
46  uint32_t rgbtoyuv[1<<24];
47 } XBRContext;
48 
49 typedef struct ThreadData {
50  AVFrame *in, *out;
51  const uint32_t *rgbtoyuv;
52 } ThreadData;
53 
54 #define OFFSET(x) offsetof(XBRContext, x)
55 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
56 static const AVOption xbr_options[] = {
57  { "n", "set scale factor", OFFSET(n), AV_OPT_TYPE_INT, {.i64 = 3}, 2, 4, .flags = FLAGS },
58  { NULL }
59 };
60 
61 AVFILTER_DEFINE_CLASS(xbr);
62 
63 static uint32_t pixel_diff(uint32_t x, uint32_t y, const uint32_t *r2y)
64 {
65 #define YMASK 0xff0000
66 #define UMASK 0x00ff00
67 #define VMASK 0x0000ff
68 
69  uint32_t yuv1 = r2y[x & 0xffffff];
70  uint32_t yuv2 = r2y[y & 0xffffff];
71 
72  return (abs((yuv1 & YMASK) - (yuv2 & YMASK)) >> 16) +
73  (abs((yuv1 & UMASK) - (yuv2 & UMASK)) >> 8) +
74  abs((yuv1 & VMASK) - (yuv2 & VMASK));
75 }
76 
77 #define ALPHA_BLEND_128_W(a, b) ((((a) & LB_MASK) >> 1) + (((b) & LB_MASK) >> 1))
78 #define ALPHA_BLEND_BASE(a, b, m, s) ( (RED_BLUE_MASK & (((a) & RED_BLUE_MASK) + (((((b) & RED_BLUE_MASK) - ((a) & RED_BLUE_MASK)) * (m)) >> (s)))) \
79  | (GREEN_MASK & (((a) & GREEN_MASK) + (((((b) & GREEN_MASK) - ((a) & GREEN_MASK)) * (m)) >> (s)))))
80 #define ALPHA_BLEND_32_W(a, b) ALPHA_BLEND_BASE(a, b, 1, 3)
81 #define ALPHA_BLEND_64_W(a, b) ALPHA_BLEND_BASE(a, b, 1, 2)
82 #define ALPHA_BLEND_192_W(a, b) ALPHA_BLEND_BASE(a, b, 3, 2)
83 #define ALPHA_BLEND_224_W(a, b) ALPHA_BLEND_BASE(a, b, 7, 3)
84 
85 #define df(A, B) pixel_diff(A, B, r2y)
86 #define eq(A, B) (df(A, B) < 155)
87 
88 #define FILT2(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, \
89  N0, N1, N2, N3) do { \
90  if (PE != PH && PE != PF) { \
91  const unsigned e = df(PE,PC) + df(PE,PG) + df(PI,H5) + df(PI,F4) + (df(PH,PF)<<2); \
92  const unsigned i = df(PH,PD) + df(PH,I5) + df(PF,I4) + df(PF,PB) + (df(PE,PI)<<2); \
93  if (e <= i) { \
94  const unsigned px = df(PE,PF) <= df(PE,PH) ? PF : PH; \
95  if (e < i && (!eq(PF,PB) && !eq(PH,PD) || eq(PE,PI) \
96  && (!eq(PF,I4) && !eq(PH,I5)) \
97  || eq(PE,PG) || eq(PE,PC))) { \
98  const unsigned ke = df(PF,PG); \
99  const unsigned ki = df(PH,PC); \
100  const int left = ke<<1 <= ki && PE != PG && PD != PG; \
101  const int up = ke >= ki<<1 && PE != PC && PB != PC; \
102  if (left && up) { \
103  E[N3] = ALPHA_BLEND_224_W(E[N3], px); \
104  E[N2] = ALPHA_BLEND_64_W( E[N2], px); \
105  E[N1] = E[N2]; \
106  } else if (left) { \
107  E[N3] = ALPHA_BLEND_192_W(E[N3], px); \
108  E[N2] = ALPHA_BLEND_64_W( E[N2], px); \
109  } else if (up) { \
110  E[N3] = ALPHA_BLEND_192_W(E[N3], px); \
111  E[N1] = ALPHA_BLEND_64_W( E[N1], px); \
112  } else { /* diagonal */ \
113  E[N3] = ALPHA_BLEND_128_W(E[N3], px); \
114  } \
115  } else { \
116  E[N3] = ALPHA_BLEND_128_W(E[N3], px); \
117  } \
118  } \
119  } \
120 } while (0)
121 
122 #define FILT3(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, \
123  N0, N1, N2, N3, N4, N5, N6, N7, N8) do { \
124  if (PE != PH && PE != PF) { \
125  const unsigned e = df(PE,PC) + df(PE,PG) + df(PI,H5) + df(PI,F4) + (df(PH,PF)<<2); \
126  const unsigned i = df(PH,PD) + df(PH,I5) + df(PF,I4) + df(PF,PB) + (df(PE,PI)<<2); \
127  if (e <= i) { \
128  const unsigned px = df(PE,PF) <= df(PE,PH) ? PF : PH; \
129  if (e < i && (!eq(PF,PB) && !eq(PF,PC) || !eq(PH,PD) && !eq(PH,PG) || eq(PE,PI) \
130  && (!eq(PF,F4) && !eq(PF,I4) || !eq(PH,H5) && !eq(PH,I5)) \
131  || eq(PE,PG) || eq(PE,PC))) { \
132  const unsigned ke = df(PF,PG); \
133  const unsigned ki = df(PH,PC); \
134  const int left = ke<<1 <= ki && PE != PG && PD != PG; \
135  const int up = ke >= ki<<1 && PE != PC && PB != PC; \
136  if (left && up) { \
137  E[N7] = ALPHA_BLEND_192_W(E[N7], px); \
138  E[N6] = ALPHA_BLEND_64_W( E[N6], px); \
139  E[N5] = E[N7]; \
140  E[N2] = E[N6]; \
141  E[N8] = px; \
142  } else if (left) { \
143  E[N7] = ALPHA_BLEND_192_W(E[N7], px); \
144  E[N5] = ALPHA_BLEND_64_W( E[N5], px); \
145  E[N6] = ALPHA_BLEND_64_W( E[N6], px); \
146  E[N8] = px; \
147  } else if (up) { \
148  E[N5] = ALPHA_BLEND_192_W(E[N5], px); \
149  E[N7] = ALPHA_BLEND_64_W( E[N7], px); \
150  E[N2] = ALPHA_BLEND_64_W( E[N2], px); \
151  E[N8] = px; \
152  } else { /* diagonal */ \
153  E[N8] = ALPHA_BLEND_224_W(E[N8], px); \
154  E[N5] = ALPHA_BLEND_32_W( E[N5], px); \
155  E[N7] = ALPHA_BLEND_32_W( E[N7], px); \
156  } \
157  } else { \
158  E[N8] = ALPHA_BLEND_128_W(E[N8], px); \
159  } \
160  } \
161  } \
162 } while (0)
163 
164 #define FILT4(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, \
165  N15, N14, N11, N3, N7, N10, N13, N12, N9, N6, N2, N1, N5, N8, N4, N0) do { \
166  if (PE != PH && PE != PF) { \
167  const unsigned e = df(PE,PC) + df(PE,PG) + df(PI,H5) + df(PI,F4) + (df(PH,PF)<<2); \
168  const unsigned i = df(PH,PD) + df(PH,I5) + df(PF,I4) + df(PF,PB) + (df(PE,PI)<<2); \
169  if (e <= i) { \
170  const unsigned px = df(PE,PF) <= df(PE,PH) ? PF : PH; \
171  if (e < i && (!eq(PF,PB) && !eq(PH,PD) || eq(PE,PI) \
172  && (!eq(PF,I4) && !eq(PH,I5)) \
173  || eq(PE,PG) || eq(PE,PC))) { \
174  const unsigned ke = df(PF,PG); \
175  const unsigned ki = df(PH,PC); \
176  const int left = ke<<1 <= ki && PE != PG && PD != PG; \
177  const int up = ke >= ki<<1 && PE != PC && PB != PC; \
178  if (left && up) { \
179  E[N13] = ALPHA_BLEND_192_W(E[N13], px); \
180  E[N12] = ALPHA_BLEND_64_W( E[N12], px); \
181  E[N15] = E[N14] = E[N11] = px; \
182  E[N10] = E[N3] = E[N12]; \
183  E[N7] = E[N13]; \
184  } else if (left) { \
185  E[N11] = ALPHA_BLEND_192_W(E[N11], px); \
186  E[N13] = ALPHA_BLEND_192_W(E[N13], px); \
187  E[N10] = ALPHA_BLEND_64_W( E[N10], px); \
188  E[N12] = ALPHA_BLEND_64_W( E[N12], px); \
189  E[N14] = px; \
190  E[N15] = px; \
191  } else if (up) { \
192  E[N14] = ALPHA_BLEND_192_W(E[N14], px); \
193  E[N7 ] = ALPHA_BLEND_192_W(E[N7 ], px); \
194  E[N10] = ALPHA_BLEND_64_W( E[N10], px); \
195  E[N3 ] = ALPHA_BLEND_64_W( E[N3 ], px); \
196  E[N11] = px; \
197  E[N15] = px; \
198  } else { /* diagonal */ \
199  E[N11] = ALPHA_BLEND_128_W(E[N11], px); \
200  E[N14] = ALPHA_BLEND_128_W(E[N14], px); \
201  E[N15] = px; \
202  } \
203  } else { \
204  E[N15] = ALPHA_BLEND_128_W(E[N15], px); \
205  } \
206  } \
207  } \
208 } while (0)
209 
210 static av_always_inline void xbr_filter(const ThreadData *td, int jobnr, int nb_jobs, int n)
211 {
212  int x, y;
213  const AVFrame *input = td->in;
214  AVFrame *output = td->out;
215  const uint32_t *r2y = td->rgbtoyuv;
216  const int slice_start = (input->height * jobnr ) / nb_jobs;
217  const int slice_end = (input->height * (jobnr+1)) / nb_jobs;
218  const int nl = output->linesize[0] >> 2;
219  const int nl1 = nl + nl;
220  const int nl2 = nl1 + nl;
221 
222  for (y = slice_start; y < slice_end; y++) {
223 
224  uint32_t *E = (uint32_t *)(output->data[0] + y * output->linesize[0] * n);
225  const uint32_t *sa2 = (uint32_t *)(input->data[0] + y * input->linesize[0] - 8); /* center */
226  const uint32_t *sa1 = sa2 - (input->linesize[0]>>2); /* up x1 */
227  const uint32_t *sa0 = sa1 - (input->linesize[0]>>2); /* up x2 */
228  const uint32_t *sa3 = sa2 + (input->linesize[0]>>2); /* down x1 */
229  const uint32_t *sa4 = sa3 + (input->linesize[0]>>2); /* down x2 */
230 
231  if (y <= 1) {
232  sa0 = sa1;
233  if (y == 0) {
234  sa0 = sa1 = sa2;
235  }
236  }
237 
238  if (y >= input->height - 2) {
239  sa4 = sa3;
240  if (y == input->height - 1) {
241  sa4 = sa3 = sa2;
242  }
243  }
244 
245  for (x = 0; x < input->width; x++) {
246  const uint32_t B1 = sa0[2];
247  const uint32_t PB = sa1[2];
248  const uint32_t PE = sa2[2];
249  const uint32_t PH = sa3[2];
250  const uint32_t H5 = sa4[2];
251 
252  const int pprev = 2 - (x > 0);
253  const uint32_t A1 = sa0[pprev];
254  const uint32_t PA = sa1[pprev];
255  const uint32_t PD = sa2[pprev];
256  const uint32_t PG = sa3[pprev];
257  const uint32_t G5 = sa4[pprev];
258 
259  const int pprev2 = pprev - (x > 1);
260  const uint32_t A0 = sa1[pprev2];
261  const uint32_t D0 = sa2[pprev2];
262  const uint32_t G0 = sa3[pprev2];
263 
264  const int pnext = 3 - (x == input->width - 1);
265  const uint32_t C1 = sa0[pnext];
266  const uint32_t PC = sa1[pnext];
267  const uint32_t PF = sa2[pnext];
268  const uint32_t PI = sa3[pnext];
269  const uint32_t I5 = sa4[pnext];
270 
271  const int pnext2 = pnext + 1 - (x >= input->width - 2);
272  const uint32_t C4 = sa1[pnext2];
273  const uint32_t F4 = sa2[pnext2];
274  const uint32_t I4 = sa3[pnext2];
275 
276  if (n == 2) {
277  E[0] = E[1] = // 0, 1
278  E[nl] = E[nl + 1] = PE; // 2, 3
279 
280  FILT2(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, 0, 1, nl, nl+1);
281  FILT2(PE, PC, PF, PB, PI, PA, PH, PD, PG, I4, A1, I5, H5, A0, D0, B1, C1, F4, C4, G5, G0, nl, 0, nl+1, 1);
282  FILT2(PE, PA, PB, PD, PC, PG, PF, PH, PI, C1, G0, C4, F4, G5, H5, D0, A0, B1, A1, I4, I5, nl+1, nl, 1, 0);
283  FILT2(PE, PG, PD, PH, PA, PI, PB, PF, PC, A0, I5, A1, B1, I4, F4, H5, G5, D0, G0, C1, C4, 1, nl+1, 0, nl);
284  } else if (n == 3) {
285  E[0] = E[1] = E[2] = // 0, 1, 2
286  E[nl] = E[nl+1] = E[nl+2] = // 3, 4, 5
287  E[nl1] = E[nl1+1] = E[nl1+2] = PE; // 6, 7, 8
288 
289  FILT3(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, 0, 1, 2, nl, nl+1, nl+2, nl1, nl1+1, nl1+2);
290  FILT3(PE, PC, PF, PB, PI, PA, PH, PD, PG, I4, A1, I5, H5, A0, D0, B1, C1, F4, C4, G5, G0, nl1, nl, 0, nl1+1, nl+1, 1, nl1+2, nl+2, 2);
291  FILT3(PE, PA, PB, PD, PC, PG, PF, PH, PI, C1, G0, C4, F4, G5, H5, D0, A0, B1, A1, I4, I5, nl1+2, nl1+1, nl1, nl+2, nl+1, nl, 2, 1, 0);
292  FILT3(PE, PG, PD, PH, PA, PI, PB, PF, PC, A0, I5, A1, B1, I4, F4, H5, G5, D0, G0, C1, C4, 2, nl+2, nl1+2, 1, nl+1, nl1+1, 0, nl, nl1);
293  } else if (n == 4) {
294  E[0] = E[1] = E[2] = E[3] = // 0, 1, 2, 3
295  E[nl] = E[nl+1] = E[nl+2] = E[nl+3] = // 4, 5, 6, 7
296  E[nl1] = E[nl1+1] = E[nl1+2] = E[nl1+3] = // 8, 9, 10, 11
297  E[nl2] = E[nl2+1] = E[nl2+2] = E[nl2+3] = PE; // 12, 13, 14, 15
298 
299  FILT4(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, nl2+3, nl2+2, nl1+3, 3, nl+3, nl1+2, nl2+1, nl2, nl1+1, nl+2, 2, 1, nl+1, nl1, nl, 0);
300  FILT4(PE, PC, PF, PB, PI, PA, PH, PD, PG, I4, A1, I5, H5, A0, D0, B1, C1, F4, C4, G5, G0, 3, nl+3, 2, 0, 1, nl+2, nl1+3, nl2+3, nl1+2, nl+1, nl, nl1, nl1+1, nl2+2, nl2+1, nl2);
301  FILT4(PE, PA, PB, PD, PC, PG, PF, PH, PI, C1, G0, C4, F4, G5, H5, D0, A0, B1, A1, I4, I5, 0, 1, nl, nl2, nl1, nl+1, 2, 3, nl+2, nl1+1, nl2+1, nl2+2, nl1+2, nl+3, nl1+3, nl2+3);
302  FILT4(PE, PG, PD, PH, PA, PI, PB, PF, PC, A0, I5, A1, B1, I4, F4, H5, G5, D0, G0, C1, C4, nl2, nl1, nl2+1, nl2+3, nl2+2, nl1+1, nl, 0, nl+1, nl1+2, nl1+3, nl+3, nl+2, 1, 2, 3);
303  }
304 
305  sa0 += 1;
306  sa1 += 1;
307  sa2 += 1;
308  sa3 += 1;
309  sa4 += 1;
310 
311  E += n;
312  }
313  }
314 }
315 
316 #define XBR_FUNC(size) \
317 static int xbr##size##x(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
318 { \
319  xbr_filter(arg, jobnr, nb_jobs, size); \
320  return 0; \
321 }
322 
323 XBR_FUNC(2)
324 XBR_FUNC(3)
325 XBR_FUNC(4)
326 
327 
328 static int config_output(AVFilterLink *outlink)
329 {
330  AVFilterContext *ctx = outlink->src;
331  XBRContext *xbr = ctx->priv;
332  AVFilterLink *inlink = ctx->inputs[0];
333 
334  outlink->w = inlink->w * xbr->n;
335  outlink->h = inlink->h * xbr->n;
336  return 0;
337 }
338 
339 static int query_formats(AVFilterContext *ctx)
340 {
341  static const enum AVPixelFormat pix_fmts[] = {
342  AV_PIX_FMT_0RGB32, AV_PIX_FMT_NONE,
343  };
344 
345  ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
346  return 0;
347 }
348 
349 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
350 {
351  AVFilterContext *ctx = inlink->dst;
352  AVFilterLink *outlink = ctx->outputs[0];
353  XBRContext *xbr = ctx->priv;
354  ThreadData td;
355 
356  AVFrame *out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
357  if (!out) {
358  av_frame_free(&in);
359  return AVERROR(ENOMEM);
360  }
361 
362  av_frame_copy_props(out, in);
363 
364  td.in = in;
365  td.out = out;
366  td.rgbtoyuv = xbr->rgbtoyuv;
367  ctx->internal->execute(ctx, xbr->func, &td, NULL, FFMIN(inlink->h, ctx->graph->nb_threads));
368 
369  out->width = outlink->w;
370  out->height = outlink->h;
371 
372  av_frame_free(&in);
373  return ff_filter_frame(outlink, out);
374 }
375 
376 static int init(AVFilterContext *ctx)
377 {
378  XBRContext *xbr = ctx->priv;
379  static const xbrfunc_t xbrfuncs[] = {xbr2x, xbr3x, xbr4x};
380 
381  uint32_t c;
382  int bg, rg, g;
383 
384  for (bg = -255; bg < 256; bg++) {
385  for (rg = -255; rg < 256; rg++) {
386  const uint32_t u = (uint32_t)((-169*rg + 500*bg)/1000) + 128;
387  const uint32_t v = (uint32_t)(( 500*rg - 81*bg)/1000) + 128;
388  int startg = FFMAX3(-bg, -rg, 0);
389  int endg = FFMIN3(255-bg, 255-rg, 255);
390  uint32_t y = (uint32_t)(( 299*rg + 1000*startg + 114*bg)/1000);
391  c = bg + (rg<<16) + 0x010101 * startg;
392  for (g = startg; g <= endg; g++) {
393  xbr->rgbtoyuv[c] = ((y++) << 16) + (u << 8) + v;
394  c+= 0x010101;
395  }
396  }
397  }
398 
399  xbr->func = xbrfuncs[xbr->n - 2];
400  return 0;
401 }
402 
403 static const AVFilterPad xbr_inputs[] = {
404  {
405  .name = "default",
406  .type = AVMEDIA_TYPE_VIDEO,
407  .filter_frame = filter_frame,
408  },
409  { NULL }
410 };
411 
412 static const AVFilterPad xbr_outputs[] = {
413  {
414  .name = "default",
415  .type = AVMEDIA_TYPE_VIDEO,
416  .config_props = config_output,
417  },
418  { NULL }
419 };
420 
421 AVFilter ff_vf_xbr = {
422  .name = "xbr",
423  .description = NULL_IF_CONFIG_SMALL("Scale the input using xBR algorithm."),
424  .inputs = xbr_inputs,
425  .outputs = xbr_outputs,
426  .query_formats = query_formats,
427  .priv_size = sizeof(XBRContext),
428  .priv_class = &xbr_class,
429  .init = init,
430  .flags = AVFILTER_FLAG_SLICE_THREADS,
431 };
Generated on Sun Mar 8 2015 02:35:08 for FFmpeg by   doxygen 1.8.2