FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
vf_edgedetect.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2012-2014 Clément Bœsch <u pkh me>
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 /**
22  * @file
23  * Edge detection filter
24  *
25  * @see https://en.wikipedia.org/wiki/Canny_edge_detector
26  */
27 
28 #include "libavutil/avassert.h"
29 #include "libavutil/opt.h"
30 #include "avfilter.h"
31 #include "formats.h"
32 #include "internal.h"
33 #include "video.h"
34 
35 enum FilterMode {
39 };
40 
41 struct plane_info {
43  uint16_t *gradients;
44  char *directions;
45 };
46 
47 typedef struct {
48  const AVClass *class;
49  struct plane_info planes[3];
50  int nb_planes;
51  double low, high;
52  uint8_t low_u8, high_u8;
55 
56 #define OFFSET(x) offsetof(EdgeDetectContext, x)
57 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
58 static const AVOption edgedetect_options[] = {
59  { "high", "set high threshold", OFFSET(high), AV_OPT_TYPE_DOUBLE, {.dbl=50/255.}, 0, 1, FLAGS },
60  { "low", "set low threshold", OFFSET(low), AV_OPT_TYPE_DOUBLE, {.dbl=20/255.}, 0, 1, FLAGS },
61  { "mode", "set mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64=MODE_WIRES}, 0, NB_MODE-1, FLAGS, "mode" },
62  { "wires", "white/gray wires on black", 0, AV_OPT_TYPE_CONST, {.i64=MODE_WIRES}, INT_MIN, INT_MAX, FLAGS, "mode" },
63  { "colormix", "mix colors", 0, AV_OPT_TYPE_CONST, {.i64=MODE_COLORMIX}, INT_MIN, INT_MAX, FLAGS, "mode" },
64  { NULL }
65 };
66 
67 AVFILTER_DEFINE_CLASS(edgedetect);
68 
69 static av_cold int init(AVFilterContext *ctx)
70 {
71  EdgeDetectContext *edgedetect = ctx->priv;
72 
73  edgedetect->low_u8 = edgedetect->low * 255. + .5;
74  edgedetect->high_u8 = edgedetect->high * 255. + .5;
75  return 0;
76 }
77 
79 {
80  const EdgeDetectContext *edgedetect = ctx->priv;
81 
82  if (edgedetect->mode == MODE_WIRES) {
83  static const enum AVPixelFormat pix_fmts[] = {AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE};
85  } else if (edgedetect->mode == MODE_COLORMIX) {
86  static const enum AVPixelFormat pix_fmts[] = {AV_PIX_FMT_GBRP, AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE};
88  } else {
89  av_assert0(0);
90  }
91  return 0;
92 }
93 
94 static int config_props(AVFilterLink *inlink)
95 {
96  int p;
97  AVFilterContext *ctx = inlink->dst;
98  EdgeDetectContext *edgedetect = ctx->priv;
99 
100  edgedetect->nb_planes = inlink->format == AV_PIX_FMT_GRAY8 ? 1 : 3;
101  for (p = 0; p < edgedetect->nb_planes; p++) {
102  struct plane_info *plane = &edgedetect->planes[p];
103 
104  plane->tmpbuf = av_malloc(inlink->w * inlink->h);
105  plane->gradients = av_calloc(inlink->w * inlink->h, sizeof(*plane->gradients));
106  plane->directions = av_malloc(inlink->w * inlink->h);
107  if (!plane->tmpbuf || !plane->gradients || !plane->directions)
108  return AVERROR(ENOMEM);
109  }
110  return 0;
111 }
112 
113 static void gaussian_blur(AVFilterContext *ctx, int w, int h,
114  uint8_t *dst, int dst_linesize,
115  const uint8_t *src, int src_linesize)
116 {
117  int i, j;
118 
119  memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
120  memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
121  for (j = 2; j < h - 2; j++) {
122  dst[0] = src[0];
123  dst[1] = src[1];
124  for (i = 2; i < w - 2; i++) {
125  /* Gaussian mask of size 5x5 with sigma = 1.4 */
126  dst[i] = ((src[-2*src_linesize + i-2] + src[2*src_linesize + i-2]) * 2
127  + (src[-2*src_linesize + i-1] + src[2*src_linesize + i-1]) * 4
128  + (src[-2*src_linesize + i ] + src[2*src_linesize + i ]) * 5
129  + (src[-2*src_linesize + i+1] + src[2*src_linesize + i+1]) * 4
130  + (src[-2*src_linesize + i+2] + src[2*src_linesize + i+2]) * 2
131 
132  + (src[ -src_linesize + i-2] + src[ src_linesize + i-2]) * 4
133  + (src[ -src_linesize + i-1] + src[ src_linesize + i-1]) * 9
134  + (src[ -src_linesize + i ] + src[ src_linesize + i ]) * 12
135  + (src[ -src_linesize + i+1] + src[ src_linesize + i+1]) * 9
136  + (src[ -src_linesize + i+2] + src[ src_linesize + i+2]) * 4
137 
138  + src[i-2] * 5
139  + src[i-1] * 12
140  + src[i ] * 15
141  + src[i+1] * 12
142  + src[i+2] * 5) / 159;
143  }
144  dst[i ] = src[i ];
145  dst[i + 1] = src[i + 1];
146 
147  dst += dst_linesize;
148  src += src_linesize;
149  }
150  memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
151  memcpy(dst, src, w);
152 }
153 
154 enum {
159 };
160 
161 static int get_rounded_direction(int gx, int gy)
162 {
163  /* reference angles:
164  * tan( pi/8) = sqrt(2)-1
165  * tan(3pi/8) = sqrt(2)+1
166  * Gy/Gx is the tangent of the angle (theta), so Gy/Gx is compared against
167  * <ref-angle>, or more simply Gy against <ref-angle>*Gx
168  *
169  * Gx and Gy bounds = [-1020;1020], using 16-bit arithmetic:
170  * round((sqrt(2)-1) * (1<<16)) = 27146
171  * round((sqrt(2)+1) * (1<<16)) = 158218
172  */
173  if (gx) {
174  int tanpi8gx, tan3pi8gx;
175 
176  if (gx < 0)
177  gx = -gx, gy = -gy;
178  gy <<= 16;
179  tanpi8gx = 27146 * gx;
180  tan3pi8gx = 158218 * gx;
181  if (gy > -tan3pi8gx && gy < -tanpi8gx) return DIRECTION_45UP;
182  if (gy > -tanpi8gx && gy < tanpi8gx) return DIRECTION_HORIZONTAL;
183  if (gy > tanpi8gx && gy < tan3pi8gx) return DIRECTION_45DOWN;
184  }
185  return DIRECTION_VERTICAL;
186 }
187 
188 static void sobel(int w, int h,
189  uint16_t *dst, int dst_linesize,
190  int8_t *dir, int dir_linesize,
191  const uint8_t *src, int src_linesize)
192 {
193  int i, j;
194 
195  for (j = 1; j < h - 1; j++) {
196  dst += dst_linesize;
197  dir += dir_linesize;
198  src += src_linesize;
199  for (i = 1; i < w - 1; i++) {
200  const int gx =
201  -1*src[-src_linesize + i-1] + 1*src[-src_linesize + i+1]
202  -2*src[ i-1] + 2*src[ i+1]
203  -1*src[ src_linesize + i-1] + 1*src[ src_linesize + i+1];
204  const int gy =
205  -1*src[-src_linesize + i-1] + 1*src[ src_linesize + i-1]
206  -2*src[-src_linesize + i ] + 2*src[ src_linesize + i ]
207  -1*src[-src_linesize + i+1] + 1*src[ src_linesize + i+1];
208 
209  dst[i] = FFABS(gx) + FFABS(gy);
210  dir[i] = get_rounded_direction(gx, gy);
211  }
212  }
213 }
214 
215 static void non_maximum_suppression(int w, int h,
216  uint8_t *dst, int dst_linesize,
217  const int8_t *dir, int dir_linesize,
218  const uint16_t *src, int src_linesize)
219 {
220  int i, j;
221 
222 #define COPY_MAXIMA(ay, ax, by, bx) do { \
223  if (src[i] > src[(ay)*src_linesize + i+(ax)] && \
224  src[i] > src[(by)*src_linesize + i+(bx)]) \
225  dst[i] = av_clip_uint8(src[i]); \
226 } while (0)
227 
228  for (j = 1; j < h - 1; j++) {
229  dst += dst_linesize;
230  dir += dir_linesize;
231  src += src_linesize;
232  for (i = 1; i < w - 1; i++) {
233  switch (dir[i]) {
234  case DIRECTION_45UP: COPY_MAXIMA( 1, -1, -1, 1); break;
235  case DIRECTION_45DOWN: COPY_MAXIMA(-1, -1, 1, 1); break;
236  case DIRECTION_HORIZONTAL: COPY_MAXIMA( 0, -1, 0, 1); break;
237  case DIRECTION_VERTICAL: COPY_MAXIMA(-1, 0, 1, 0); break;
238  }
239  }
240  }
241 }
242 
243 static void double_threshold(int low, int high, int w, int h,
244  uint8_t *dst, int dst_linesize,
245  const uint8_t *src, int src_linesize)
246 {
247  int i, j;
248 
249  for (j = 0; j < h; j++) {
250  for (i = 0; i < w; i++) {
251  if (src[i] > high) {
252  dst[i] = src[i];
253  continue;
254  }
255 
256  if ((!i || i == w - 1 || !j || j == h - 1) &&
257  src[i] > low &&
258  (src[-src_linesize + i-1] > high ||
259  src[-src_linesize + i ] > high ||
260  src[-src_linesize + i+1] > high ||
261  src[ i-1] > high ||
262  src[ i+1] > high ||
263  src[ src_linesize + i-1] > high ||
264  src[ src_linesize + i ] > high ||
265  src[ src_linesize + i+1] > high))
266  dst[i] = src[i];
267  else
268  dst[i] = 0;
269  }
270  dst += dst_linesize;
271  src += src_linesize;
272  }
273 }
274 
275 static void color_mix(int w, int h,
276  uint8_t *dst, int dst_linesize,
277  const uint8_t *src, int src_linesize)
278 {
279  int i, j;
280 
281  for (j = 0; j < h; j++) {
282  for (i = 0; i < w; i++)
283  dst[i] = (dst[i] + src[i]) >> 1;
284  dst += dst_linesize;
285  src += src_linesize;
286  }
287 }
288 
289 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
290 {
291  AVFilterContext *ctx = inlink->dst;
292  EdgeDetectContext *edgedetect = ctx->priv;
293  AVFilterLink *outlink = ctx->outputs[0];
294  int p, direct = 0;
295  AVFrame *out;
296 
297  if (edgedetect->mode != MODE_COLORMIX && av_frame_is_writable(in)) {
298  direct = 1;
299  out = in;
300  } else {
301  out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
302  if (!out) {
303  av_frame_free(&in);
304  return AVERROR(ENOMEM);
305  }
306  av_frame_copy_props(out, in);
307  }
308 
309  for (p = 0; p < edgedetect->nb_planes; p++) {
310  struct plane_info *plane = &edgedetect->planes[p];
311  uint8_t *tmpbuf = plane->tmpbuf;
312  uint16_t *gradients = plane->gradients;
313  int8_t *directions = plane->directions;
314 
315  /* gaussian filter to reduce noise */
316  gaussian_blur(ctx, inlink->w, inlink->h,
317  tmpbuf, inlink->w,
318  in->data[p], in->linesize[p]);
319 
320  /* compute the 16-bits gradients and directions for the next step */
321  sobel(inlink->w, inlink->h,
322  gradients, inlink->w,
323  directions,inlink->w,
324  tmpbuf, inlink->w);
325 
326  /* non_maximum_suppression() will actually keep & clip what's necessary and
327  * ignore the rest, so we need a clean output buffer */
328  memset(tmpbuf, 0, inlink->w * inlink->h);
329  non_maximum_suppression(inlink->w, inlink->h,
330  tmpbuf, inlink->w,
331  directions,inlink->w,
332  gradients, inlink->w);
333 
334  /* keep high values, or low values surrounded by high values */
335  double_threshold(edgedetect->low_u8, edgedetect->high_u8,
336  inlink->w, inlink->h,
337  out->data[p], out->linesize[p],
338  tmpbuf, inlink->w);
339 
340  if (edgedetect->mode == MODE_COLORMIX) {
341  color_mix(inlink->w, inlink->h,
342  out->data[p], out->linesize[p],
343  in->data[p], in->linesize[p]);
344  }
345  }
346 
347  if (!direct)
348  av_frame_free(&in);
349  return ff_filter_frame(outlink, out);
350 }
351 
352 static av_cold void uninit(AVFilterContext *ctx)
353 {
354  int p;
355  EdgeDetectContext *edgedetect = ctx->priv;
356 
357  for (p = 0; p < edgedetect->nb_planes; p++) {
358  struct plane_info *plane = &edgedetect->planes[p];
359  av_freep(&plane->tmpbuf);
360  av_freep(&plane->gradients);
361  av_freep(&plane->directions);
362  }
363 }
364 
365 static const AVFilterPad edgedetect_inputs[] = {
366  {
367  .name = "default",
368  .type = AVMEDIA_TYPE_VIDEO,
369  .config_props = config_props,
370  .filter_frame = filter_frame,
371  },
372  { NULL }
373 };
374 
375 static const AVFilterPad edgedetect_outputs[] = {
376  {
377  .name = "default",
378  .type = AVMEDIA_TYPE_VIDEO,
379  },
380  { NULL }
381 };
382 
384  .name = "edgedetect",
385  .description = NULL_IF_CONFIG_SMALL("Detect and draw edge."),
386  .priv_size = sizeof(EdgeDetectContext),
387  .init = init,
388  .uninit = uninit,
390  .inputs = edgedetect_inputs,
391  .outputs = edgedetect_outputs,
392  .priv_class = &edgedetect_class,
394 };