FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
vf_perspective.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2002 Michael Niedermayer <michaelni@gmx.at>
3  * Copyright (c) 2013 Paul B Mahol
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public
9  * License as published by the Free Software Foundation; either
10  * version 2 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
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License along
18  * 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 #include "libavutil/eval.h"
23 #include "libavutil/imgutils.h"
24 #include "libavutil/pixdesc.h"
25 #include "libavutil/opt.h"
26 #include "avfilter.h"
27 #include "formats.h"
28 #include "internal.h"
29 #include "video.h"
30 
31 #define SUB_PIXEL_BITS 8
32 #define SUB_PIXELS (1 << SUB_PIXEL_BITS)
33 #define COEFF_BITS 11
34 
35 #define LINEAR 0
36 #define CUBIC 1
37 
38 typedef struct PerspectiveContext {
39  const AVClass *class;
40  char *expr_str[4][2];
41  double ref[4][2];
42  int32_t (*pv)[2];
45  int linesize[4];
46  int height[4];
47  int hsub, vsub;
48  int nb_planes;
49 
51  void *arg, int job, int nb_jobs);
53 
54 #define OFFSET(x) offsetof(PerspectiveContext, x)
55 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
56 
57 static const AVOption perspective_options[] = {
58  { "x0", "set top left x coordinate", OFFSET(expr_str[0][0]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
59  { "y0", "set top left y coordinate", OFFSET(expr_str[0][1]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
60  { "x1", "set top right x coordinate", OFFSET(expr_str[1][0]), AV_OPT_TYPE_STRING, {.str="W"}, 0, 0, FLAGS },
61  { "y1", "set top right y coordinate", OFFSET(expr_str[1][1]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
62  { "x2", "set bottom left x coordinate", OFFSET(expr_str[2][0]), AV_OPT_TYPE_STRING, {.str="0"}, 0, 0, FLAGS },
63  { "y2", "set bottom left y coordinate", OFFSET(expr_str[2][1]), AV_OPT_TYPE_STRING, {.str="H"}, 0, 0, FLAGS },
64  { "x3", "set bottom right x coordinate", OFFSET(expr_str[3][0]), AV_OPT_TYPE_STRING, {.str="W"}, 0, 0, FLAGS },
65  { "y3", "set bottom right y coordinate", OFFSET(expr_str[3][1]), AV_OPT_TYPE_STRING, {.str="H"}, 0, 0, FLAGS },
66  { "interpolation", "set interpolation", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=LINEAR}, 0, 1, FLAGS, "interpolation" },
67  { "linear", "", 0, AV_OPT_TYPE_CONST, {.i64=LINEAR}, 0, 0, FLAGS, "interpolation" },
68  { "cubic", "", 0, AV_OPT_TYPE_CONST, {.i64=CUBIC}, 0, 0, FLAGS, "interpolation" },
69  { NULL }
70 };
71 
72 AVFILTER_DEFINE_CLASS(perspective);
73 
75 {
76  static const enum AVPixelFormat pix_fmts[] = {
81  };
82 
84  return 0;
85 }
86 
87 static inline double get_coeff(double d)
88 {
89  double coeff, A = -0.60;
90 
91  d = fabs(d);
92 
93  if (d < 1.0)
94  coeff = (1.0 - (A + 3.0) * d * d + (A + 2.0) * d * d * d);
95  else if (d < 2.0)
96  coeff = (-4.0 * A + 8.0 * A * d - 5.0 * A * d * d + A * d * d * d);
97  else
98  coeff = 0.0;
99 
100  return coeff;
101 }
102 
103 static const char *const var_names[] = { "W", "H", NULL };
105 
106 static int config_input(AVFilterLink *inlink)
107 {
108  double x0, x1, x2, x3, x4, x5, x6, x7, q;
109  AVFilterContext *ctx = inlink->dst;
110  PerspectiveContext *s = ctx->priv;
111  double (*ref)[2] = s->ref;
112  const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
113  double values[VAR_VARS_NB] = { [VAR_W] = inlink->w, [VAR_H] = inlink->h };
114  int h = inlink->h;
115  int w = inlink->w;
116  int x, y, i, j, ret;
117 
118  for (i = 0; i < 4; i++) {
119  for (j = 0; j < 2; j++) {
120  if (!s->expr_str[i][j])
121  return AVERROR(EINVAL);
122  ret = av_expr_parse_and_eval(&s->ref[i][j], s->expr_str[i][j],
123  var_names, &values[0],
124  NULL, NULL, NULL, NULL,
125  0, 0, ctx);
126  if (ret < 0)
127  return ret;
128  }
129  }
130 
131  s->hsub = desc->log2_chroma_w;
132  s->vsub = desc->log2_chroma_h;
134  if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
135  return ret;
136 
137  s->height[1] = s->height[2] = FF_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
138  s->height[0] = s->height[3] = inlink->h;
139 
140  s->pv = av_realloc_f(s->pv, w * h, 2 * sizeof(*s->pv));
141  if (!s->pv)
142  return AVERROR(ENOMEM);
143 
144  x6 = ((ref[0][0] - ref[1][0] - ref[2][0] + ref[3][0]) *
145  (ref[2][1] - ref[3][1]) -
146  ( ref[0][1] - ref[1][1] - ref[2][1] + ref[3][1]) *
147  (ref[2][0] - ref[3][0])) * h;
148  x7 = ((ref[0][1] - ref[1][1] - ref[2][1] + ref[3][1]) *
149  (ref[1][0] - ref[3][0]) -
150  ( ref[0][0] - ref[1][0] - ref[2][0] + ref[3][0]) *
151  (ref[1][1] - ref[3][1])) * w;
152  q = ( ref[1][0] - ref[3][0]) * (ref[2][1] - ref[3][1]) -
153  ( ref[2][0] - ref[3][0]) * (ref[1][1] - ref[3][1]);
154 
155  x0 = q * (ref[1][0] - ref[0][0]) * h + x6 * ref[1][0];
156  x1 = q * (ref[2][0] - ref[0][0]) * w + x7 * ref[2][0];
157  x2 = q * ref[0][0] * w * h;
158  x3 = q * (ref[1][1] - ref[0][1]) * h + x6 * ref[1][1];
159  x4 = q * (ref[2][1] - ref[0][1]) * w + x7 * ref[2][1];
160  x5 = q * ref[0][1] * w * h;
161 
162  for (y = 0; y < h; y++){
163  for (x = 0; x < w; x++){
164  int u, v;
165 
166  u = (int)floor(SUB_PIXELS * (x0 * x + x1 * y + x2) /
167  (x6 * x + x7 * y + q * w * h) + 0.5);
168  v = (int)floor(SUB_PIXELS * (x3 * x + x4 * y + x5) /
169  (x6 * x + x7 * y + q * w * h) + 0.5);
170 
171  s->pv[x + y * w][0] = u;
172  s->pv[x + y * w][1] = v;
173  }
174  }
175 
176  for (i = 0; i < SUB_PIXELS; i++){
177  double d = i / (double)SUB_PIXELS;
178  double temp[4];
179  double sum = 0;
180 
181  for (j = 0; j < 4; j++)
182  temp[j] = get_coeff(j - d - 1);
183 
184  for (j = 0; j < 4; j++)
185  sum += temp[j];
186 
187  for (j = 0; j < 4; j++)
188  s->coeff[i][j] = (int)floor((1 << COEFF_BITS) * temp[j] / sum + 0.5);
189  }
190 
191  return 0;
192 }
193 
194 typedef struct ThreadData {
199  int w, h;
200  int hsub, vsub;
201 } ThreadData;
202 
203 static int resample_cubic(AVFilterContext *ctx, void *arg,
204  int job, int nb_jobs)
205 {
206  PerspectiveContext *s = ctx->priv;
207  ThreadData *td = arg;
208  uint8_t *dst = td->dst;
209  int dst_linesize = td->dst_linesize;
210  uint8_t *src = td->src;
211  int src_linesize = td->src_linesize;
212  int w = td->w;
213  int h = td->h;
214  int hsub = td->hsub;
215  int vsub = td->vsub;
216  int start = (h * job) / nb_jobs;
217  int end = (h * (job+1)) / nb_jobs;
218  const int linesize = s->linesize[0];
219  int x, y;
220 
221  for (y = start; y < end; y++) {
222  int sy = y << vsub;
223  for (x = 0; x < w; x++) {
224  int u, v, subU, subV, sum, sx;
225 
226  sx = x << hsub;
227  u = s->pv[sx + sy * linesize][0] >> hsub;
228  v = s->pv[sx + sy * linesize][1] >> vsub;
229  subU = u & (SUB_PIXELS - 1);
230  subV = v & (SUB_PIXELS - 1);
231  u >>= SUB_PIXEL_BITS;
232  v >>= SUB_PIXEL_BITS;
233 
234  if (u > 0 && v > 0 && u < w - 2 && v < h - 2){
235  const int index = u + v*src_linesize;
236  const int a = s->coeff[subU][0];
237  const int b = s->coeff[subU][1];
238  const int c = s->coeff[subU][2];
239  const int d = s->coeff[subU][3];
240 
241  sum = s->coeff[subV][0] * (a * src[index - 1 - src_linesize] + b * src[index - 0 - src_linesize] +
242  c * src[index + 1 - src_linesize] + d * src[index + 2 - src_linesize]) +
243  s->coeff[subV][1] * (a * src[index - 1 ] + b * src[index - 0 ] +
244  c * src[index + 1 ] + d * src[index + 2 ]) +
245  s->coeff[subV][2] * (a * src[index - 1 + src_linesize] + b * src[index - 0 + src_linesize] +
246  c * src[index + 1 + src_linesize] + d * src[index + 2 + src_linesize]) +
247  s->coeff[subV][3] * (a * src[index - 1 + 2 * src_linesize] + b * src[index - 0 + 2 * src_linesize] +
248  c * src[index + 1 + 2 * src_linesize] + d * src[index + 2 + 2 * src_linesize]);
249  } else {
250  int dx, dy;
251 
252  sum = 0;
253 
254  for (dy = 0; dy < 4; dy++) {
255  int iy = v + dy - 1;
256 
257  if (iy < 0)
258  iy = 0;
259  else if (iy >= h)
260  iy = h-1;
261  for (dx = 0; dx < 4; dx++) {
262  int ix = u + dx - 1;
263 
264  if (ix < 0)
265  ix = 0;
266  else if (ix >= w)
267  ix = w - 1;
268 
269  sum += s->coeff[subU][dx] * s->coeff[subV][dy] * src[ ix + iy * src_linesize];
270  }
271  }
272  }
273 
274  sum = (sum + (1<<(COEFF_BITS * 2 - 1))) >> (COEFF_BITS * 2);
275  sum = av_clip(sum, 0, 255);
276  dst[x + y * dst_linesize] = sum;
277  }
278  }
279  return 0;
280 }
281 
282 static int resample_linear(AVFilterContext *ctx, void *arg,
283  int job, int nb_jobs)
284 {
285  PerspectiveContext *s = ctx->priv;
286  ThreadData *td = arg;
287  uint8_t *dst = td->dst;
288  int dst_linesize = td->dst_linesize;
289  uint8_t *src = td->src;
290  int src_linesize = td->src_linesize;
291  int w = td->w;
292  int h = td->h;
293  int hsub = td->hsub;
294  int vsub = td->vsub;
295  int start = (h * job) / nb_jobs;
296  int end = (h * (job+1)) / nb_jobs;
297  const int linesize = s->linesize[0];
298  int x, y;
299 
300  for (y = start; y < end; y++){
301  int sy = y << vsub;
302  for (x = 0; x < w; x++){
303  int u, v, subU, subV, sum, sx, index, subUI, subVI;
304 
305  sx = x << hsub;
306  u = s->pv[sx + sy * linesize][0] >> hsub;
307  v = s->pv[sx + sy * linesize][1] >> vsub;
308  subU = u & (SUB_PIXELS - 1);
309  subV = v & (SUB_PIXELS - 1);
310  u >>= SUB_PIXEL_BITS;
311  v >>= SUB_PIXEL_BITS;
312 
313  index = u + v * src_linesize;
314  subUI = SUB_PIXELS - subU;
315  subVI = SUB_PIXELS - subV;
316 
317  if ((unsigned)u < (unsigned)(w - 1)){
318  if((unsigned)v < (unsigned)(h - 1)){
319  sum = subVI * (subUI * src[index] + subU * src[index + 1]) +
320  subV * (subUI * src[index + src_linesize] + subU * src[index + src_linesize + 1]);
321  sum = (sum + (1 << (SUB_PIXEL_BITS * 2 - 1)))>> (SUB_PIXEL_BITS * 2);
322  } else {
323  if (v < 0)
324  v = 0;
325  else
326  v = h - 1;
327  index = u + v * src_linesize;
328  sum = subUI * src[index] + subU * src[index + 1];
329  sum = (sum + (1 << (SUB_PIXEL_BITS - 1))) >> SUB_PIXEL_BITS;
330  }
331  } else {
332  if (u < 0)
333  u = 0;
334  else
335  u = w - 1;
336  if ((unsigned)v < (unsigned)(h - 1)){
337  index = u + v * src_linesize;
338  sum = subVI * src[index] + subV * src[index + src_linesize];
339  sum = (sum + (1 << (SUB_PIXEL_BITS - 1))) >> SUB_PIXEL_BITS;
340  } else {
341  if (v < 0)
342  v = 0;
343  else
344  v = h - 1;
345  index = u + v * src_linesize;
346  sum = src[index];
347  }
348  }
349 
350  sum = av_clip(sum, 0, 255);
351  dst[x + y * dst_linesize] = sum;
352  }
353  }
354  return 0;
355 }
356 
357 static av_cold int init(AVFilterContext *ctx)
358 {
359  PerspectiveContext *s = ctx->priv;
360 
361  switch (s->interpolation) {
362  case LINEAR: s->perspective = resample_linear; break;
363  case CUBIC: s->perspective = resample_cubic; break;
364  }
365 
366  return 0;
367 }
368 
369 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
370 {
371  AVFilterContext *ctx = inlink->dst;
372  AVFilterLink *outlink = ctx->outputs[0];
373  PerspectiveContext *s = ctx->priv;
374  AVFrame *out;
375  int plane;
376 
377  out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
378  if (!out) {
379  av_frame_free(&frame);
380  return AVERROR(ENOMEM);
381  }
382  av_frame_copy_props(out, frame);
383 
384  for (plane = 0; plane < s->nb_planes; plane++) {
385  int hsub = plane == 1 || plane == 2 ? s->hsub : 0;
386  int vsub = plane == 1 || plane == 2 ? s->vsub : 0;
387  ThreadData td = {.dst = out->data[plane],
388  .dst_linesize = out->linesize[plane],
389  .src = frame->data[plane],
390  .src_linesize = frame->linesize[plane],
391  .w = s->linesize[plane],
392  .h = s->height[plane],
393  .hsub = hsub,
394  .vsub = vsub };
395  ctx->internal->execute(ctx, s->perspective, &td, NULL, FFMIN(td.h, ctx->graph->nb_threads));
396  }
397 
398  av_frame_free(&frame);
399  return ff_filter_frame(outlink, out);
400 }
401 
402 static av_cold void uninit(AVFilterContext *ctx)
403 {
404  PerspectiveContext *s = ctx->priv;
405 
406  av_freep(&s->pv);
407 }
408 
409 static const AVFilterPad perspective_inputs[] = {
410  {
411  .name = "default",
412  .type = AVMEDIA_TYPE_VIDEO,
413  .filter_frame = filter_frame,
414  .config_props = config_input,
415  },
416  { NULL }
417 };
418 
420  {
421  .name = "default",
422  .type = AVMEDIA_TYPE_VIDEO,
423  },
424  { NULL }
425 };
426 
428  .name = "perspective",
429  .description = NULL_IF_CONFIG_SMALL("Correct the perspective of video."),
430  .priv_size = sizeof(PerspectiveContext),
431  .init = init,
432  .uninit = uninit,
434  .inputs = perspective_inputs,
435  .outputs = perspective_outputs,
436  .priv_class = &perspective_class,
438 };