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/imgutils.h"
30 #include "libavutil/opt.h"
31 #include "avfilter.h"
32 #include "formats.h"
33 #include "internal.h"
34 #include "video.h"
35
36 #define PLANE_R 0x4
37 #define PLANE_G 0x1
38 #define PLANE_B 0x2
39 #define PLANE_Y 0x1
40 #define PLANE_U 0x2
41 #define PLANE_V 0x4
42 #define PLANE_A 0x8
43
44 enum FilterMode {
45 MODE_WIRES,
46 MODE_COLORMIX,
47 MODE_CANNY,
48 NB_MODE
49 };
50
51 struct plane_info {
52 uint8_t *tmpbuf;
53 uint16_t *gradients;
54 char *directions;
55 int width, height;
56 };
57
58 typedef struct EdgeDetectContext {
59 const AVClass *class;
60 struct plane_info planes[3];
61 int filter_planes;
62 int nb_planes;
63 double low, high;
64 uint8_t low_u8, high_u8;
65 int mode;
66 } EdgeDetectContext;
67
68 #define OFFSET(x) offsetof(EdgeDetectContext, x)
69 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
70 static const AVOption edgedetect_options[] = {
71 { "high", "set high threshold", OFFSET(high), AV_OPT_TYPE_DOUBLE, {.dbl=50/255.}, 0, 1, FLAGS },
72 { "low", "set low threshold", OFFSET(low), AV_OPT_TYPE_DOUBLE, {.dbl=20/255.}, 0, 1, FLAGS },
73 { "mode", "set mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64=MODE_WIRES}, 0, NB_MODE-1, FLAGS, "mode" },
74 { "wires", "white/gray wires on black", 0, AV_OPT_TYPE_CONST, {.i64=MODE_WIRES}, INT_MIN, INT_MAX, FLAGS, "mode" },
75 { "colormix", "mix colors", 0, AV_OPT_TYPE_CONST, {.i64=MODE_COLORMIX}, INT_MIN, INT_MAX, FLAGS, "mode" },
76 { "canny", "detect edges on planes", 0, AV_OPT_TYPE_CONST, {.i64=MODE_CANNY}, INT_MIN, INT_MAX, FLAGS, "mode" },
77 { "planes", "set planes to filter", OFFSET(filter_planes), AV_OPT_TYPE_FLAGS, {.i64=7}, 1, 0x7, FLAGS, "flags" },
78 { "y", "filter luma plane", 0, AV_OPT_TYPE_CONST, {.i64=PLANE_Y}, 0, 0, FLAGS, "flags" },
79 { "u", "filter u plane", 0, AV_OPT_TYPE_CONST, {.i64=PLANE_U}, 0, 0, FLAGS, "flags" },
80 { "v", "filter v plane", 0, AV_OPT_TYPE_CONST, {.i64=PLANE_V}, 0, 0, FLAGS, "flags" },
81 { "r", "filter red plane", 0, AV_OPT_TYPE_CONST, {.i64=PLANE_R}, 0, 0, FLAGS, "flags" },
82 { "g", "filter green plane", 0, AV_OPT_TYPE_CONST, {.i64=PLANE_G}, 0, 0, FLAGS, "flags" },
83 { "b", "filter blue plane", 0, AV_OPT_TYPE_CONST, {.i64=PLANE_B}, 0, 0, FLAGS, "flags" },
84 { NULL }
85 };
86
87 AVFILTER_DEFINE_CLASS(edgedetect);
88
init(AVFilterContext * ctx)89 static av_cold int init(AVFilterContext *ctx)
90 {
91 EdgeDetectContext *edgedetect = ctx->priv;
92
93 edgedetect->low_u8 = edgedetect->low * 255. + .5;
94 edgedetect->high_u8 = edgedetect->high * 255. + .5;
95 return 0;
96 }
97
query_formats(AVFilterContext * ctx)98 static int query_formats(AVFilterContext *ctx)
99 {
100 const EdgeDetectContext *edgedetect = ctx->priv;
101 static const enum AVPixelFormat wires_pix_fmts[] = {AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE};
102 static const enum AVPixelFormat canny_pix_fmts[] = {AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_GBRP, AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE};
103 static const enum AVPixelFormat colormix_pix_fmts[] = {AV_PIX_FMT_GBRP, AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE};
104 AVFilterFormats *fmts_list;
105 const enum AVPixelFormat *pix_fmts = NULL;
106
107 if (edgedetect->mode == MODE_WIRES) {
108 pix_fmts = wires_pix_fmts;
109 } else if (edgedetect->mode == MODE_COLORMIX) {
110 pix_fmts = colormix_pix_fmts;
111 } else if (edgedetect->mode == MODE_CANNY) {
112 pix_fmts = canny_pix_fmts;
113 } else {
114 av_assert0(0);
115 }
116 fmts_list = ff_make_format_list(pix_fmts);
117 if (!fmts_list)
118 return AVERROR(ENOMEM);
119 return ff_set_common_formats(ctx, fmts_list);
120 }
121
config_props(AVFilterLink * inlink)122 static int config_props(AVFilterLink *inlink)
123 {
124 int p;
125 AVFilterContext *ctx = inlink->dst;
126 EdgeDetectContext *edgedetect = ctx->priv;
127 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
128
129 edgedetect->nb_planes = inlink->format == AV_PIX_FMT_GRAY8 ? 1 : 3;
130 for (p = 0; p < edgedetect->nb_planes; p++) {
131 struct plane_info *plane = &edgedetect->planes[p];
132 int vsub = p ? desc->log2_chroma_h : 0;
133 int hsub = p ? desc->log2_chroma_w : 0;
134
135 plane->width = AV_CEIL_RSHIFT(inlink->w, hsub);
136 plane->height = AV_CEIL_RSHIFT(inlink->h, vsub);
137 plane->tmpbuf = av_malloc(plane->width * plane->height);
138 plane->gradients = av_calloc(plane->width * plane->height, sizeof(*plane->gradients));
139 plane->directions = av_malloc(plane->width * plane->height);
140 if (!plane->tmpbuf || !plane->gradients || !plane->directions)
141 return AVERROR(ENOMEM);
142 }
143 return 0;
144 }
145
gaussian_blur(AVFilterContext * ctx,int w,int h,uint8_t * dst,int dst_linesize,const uint8_t * src,int src_linesize)146 static void gaussian_blur(AVFilterContext *ctx, int w, int h,
147 uint8_t *dst, int dst_linesize,
148 const uint8_t *src, int src_linesize)
149 {
150 int i, j;
151
152 memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
153 if (h > 1) {
154 memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
155 }
156 for (j = 2; j < h - 2; j++) {
157 dst[0] = src[0];
158 if (w > 1)
159 dst[1] = src[1];
160 for (i = 2; i < w - 2; i++) {
161 /* Gaussian mask of size 5x5 with sigma = 1.4 */
162 dst[i] = ((src[-2*src_linesize + i-2] + src[2*src_linesize + i-2]) * 2
163 + (src[-2*src_linesize + i-1] + src[2*src_linesize + i-1]) * 4
164 + (src[-2*src_linesize + i ] + src[2*src_linesize + i ]) * 5
165 + (src[-2*src_linesize + i+1] + src[2*src_linesize + i+1]) * 4
166 + (src[-2*src_linesize + i+2] + src[2*src_linesize + i+2]) * 2
167
168 + (src[ -src_linesize + i-2] + src[ src_linesize + i-2]) * 4
169 + (src[ -src_linesize + i-1] + src[ src_linesize + i-1]) * 9
170 + (src[ -src_linesize + i ] + src[ src_linesize + i ]) * 12
171 + (src[ -src_linesize + i+1] + src[ src_linesize + i+1]) * 9
172 + (src[ -src_linesize + i+2] + src[ src_linesize + i+2]) * 4
173
174 + src[i-2] * 5
175 + src[i-1] * 12
176 + src[i ] * 15
177 + src[i+1] * 12
178 + src[i+2] * 5) / 159;
179 }
180 if (w > 2)
181 dst[i ] = src[i ];
182 if (w > 3)
183 dst[i + 1] = src[i + 1];
184
185 dst += dst_linesize;
186 src += src_linesize;
187 }
188 if (h > 2) {
189 memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
190 }
191 if (h > 3)
192 memcpy(dst, src, w);
193 }
194
195 enum {
196 DIRECTION_45UP,
197 DIRECTION_45DOWN,
198 DIRECTION_HORIZONTAL,
199 DIRECTION_VERTICAL,
200 };
201
get_rounded_direction(int gx,int gy)202 static int get_rounded_direction(int gx, int gy)
203 {
204 /* reference angles:
205 * tan( pi/8) = sqrt(2)-1
206 * tan(3pi/8) = sqrt(2)+1
207 * Gy/Gx is the tangent of the angle (theta), so Gy/Gx is compared against
208 * <ref-angle>, or more simply Gy against <ref-angle>*Gx
209 *
210 * Gx and Gy bounds = [-1020;1020], using 16-bit arithmetic:
211 * round((sqrt(2)-1) * (1<<16)) = 27146
212 * round((sqrt(2)+1) * (1<<16)) = 158218
213 */
214 if (gx) {
215 int tanpi8gx, tan3pi8gx;
216
217 if (gx < 0)
218 gx = -gx, gy = -gy;
219 gy *= (1 << 16);
220 tanpi8gx = 27146 * gx;
221 tan3pi8gx = 158218 * gx;
222 if (gy > -tan3pi8gx && gy < -tanpi8gx) return DIRECTION_45UP;
223 if (gy > -tanpi8gx && gy < tanpi8gx) return DIRECTION_HORIZONTAL;
224 if (gy > tanpi8gx && gy < tan3pi8gx) return DIRECTION_45DOWN;
225 }
226 return DIRECTION_VERTICAL;
227 }
228
sobel(int w,int h,uint16_t * dst,int dst_linesize,int8_t * dir,int dir_linesize,const uint8_t * src,int src_linesize)229 static void sobel(int w, int h,
230 uint16_t *dst, int dst_linesize,
231 int8_t *dir, int dir_linesize,
232 const uint8_t *src, int src_linesize)
233 {
234 int i, j;
235
236 for (j = 1; j < h - 1; j++) {
237 dst += dst_linesize;
238 dir += dir_linesize;
239 src += src_linesize;
240 for (i = 1; i < w - 1; i++) {
241 const int gx =
242 -1*src[-src_linesize + i-1] + 1*src[-src_linesize + i+1]
243 -2*src[ i-1] + 2*src[ i+1]
244 -1*src[ src_linesize + i-1] + 1*src[ src_linesize + i+1];
245 const int gy =
246 -1*src[-src_linesize + i-1] + 1*src[ src_linesize + i-1]
247 -2*src[-src_linesize + i ] + 2*src[ src_linesize + i ]
248 -1*src[-src_linesize + i+1] + 1*src[ src_linesize + i+1];
249
250 dst[i] = FFABS(gx) + FFABS(gy);
251 dir[i] = get_rounded_direction(gx, gy);
252 }
253 }
254 }
255
non_maximum_suppression(int w,int h,uint8_t * dst,int dst_linesize,const int8_t * dir,int dir_linesize,const uint16_t * src,int src_linesize)256 static void non_maximum_suppression(int w, int h,
257 uint8_t *dst, int dst_linesize,
258 const int8_t *dir, int dir_linesize,
259 const uint16_t *src, int src_linesize)
260 {
261 int i, j;
262
263 #define COPY_MAXIMA(ay, ax, by, bx) do { \
264 if (src[i] > src[(ay)*src_linesize + i+(ax)] && \
265 src[i] > src[(by)*src_linesize + i+(bx)]) \
266 dst[i] = av_clip_uint8(src[i]); \
267 } while (0)
268
269 for (j = 1; j < h - 1; j++) {
270 dst += dst_linesize;
271 dir += dir_linesize;
272 src += src_linesize;
273 for (i = 1; i < w - 1; i++) {
274 switch (dir[i]) {
275 case DIRECTION_45UP: COPY_MAXIMA( 1, -1, -1, 1); break;
276 case DIRECTION_45DOWN: COPY_MAXIMA(-1, -1, 1, 1); break;
277 case DIRECTION_HORIZONTAL: COPY_MAXIMA( 0, -1, 0, 1); break;
278 case DIRECTION_VERTICAL: COPY_MAXIMA(-1, 0, 1, 0); break;
279 }
280 }
281 }
282 }
283
double_threshold(int low,int high,int w,int h,uint8_t * dst,int dst_linesize,const uint8_t * src,int src_linesize)284 static void double_threshold(int low, int high, int w, int h,
285 uint8_t *dst, int dst_linesize,
286 const uint8_t *src, int src_linesize)
287 {
288 int i, j;
289
290 for (j = 0; j < h; j++) {
291 for (i = 0; i < w; i++) {
292 if (src[i] > high) {
293 dst[i] = src[i];
294 continue;
295 }
296
297 if (!(!i || i == w - 1 || !j || j == h - 1) &&
298 src[i] > low &&
299 (src[-src_linesize + i-1] > high ||
300 src[-src_linesize + i ] > high ||
301 src[-src_linesize + i+1] > high ||
302 src[ i-1] > high ||
303 src[ i+1] > high ||
304 src[ src_linesize + i-1] > high ||
305 src[ src_linesize + i ] > high ||
306 src[ src_linesize + i+1] > high))
307 dst[i] = src[i];
308 else
309 dst[i] = 0;
310 }
311 dst += dst_linesize;
312 src += src_linesize;
313 }
314 }
315
color_mix(int w,int h,uint8_t * dst,int dst_linesize,const uint8_t * src,int src_linesize)316 static void color_mix(int w, int h,
317 uint8_t *dst, int dst_linesize,
318 const uint8_t *src, int src_linesize)
319 {
320 int i, j;
321
322 for (j = 0; j < h; j++) {
323 for (i = 0; i < w; i++)
324 dst[i] = (dst[i] + src[i]) >> 1;
325 dst += dst_linesize;
326 src += src_linesize;
327 }
328 }
329
filter_frame(AVFilterLink * inlink,AVFrame * in)330 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
331 {
332 AVFilterContext *ctx = inlink->dst;
333 EdgeDetectContext *edgedetect = ctx->priv;
334 AVFilterLink *outlink = ctx->outputs[0];
335 int p, direct = 0;
336 AVFrame *out;
337
338 if (edgedetect->mode != MODE_COLORMIX && av_frame_is_writable(in)) {
339 direct = 1;
340 out = in;
341 } else {
342 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
343 if (!out) {
344 av_frame_free(&in);
345 return AVERROR(ENOMEM);
346 }
347 av_frame_copy_props(out, in);
348 }
349
350 for (p = 0; p < edgedetect->nb_planes; p++) {
351 struct plane_info *plane = &edgedetect->planes[p];
352 uint8_t *tmpbuf = plane->tmpbuf;
353 uint16_t *gradients = plane->gradients;
354 int8_t *directions = plane->directions;
355 const int width = plane->width;
356 const int height = plane->height;
357
358 if (!((1 << p) & edgedetect->filter_planes)) {
359 if (!direct)
360 av_image_copy_plane(out->data[p], out->linesize[p],
361 in->data[p], in->linesize[p],
362 width, height);
363 continue;
364 }
365
366 /* gaussian filter to reduce noise */
367 gaussian_blur(ctx, width, height,
368 tmpbuf, width,
369 in->data[p], in->linesize[p]);
370
371 /* compute the 16-bits gradients and directions for the next step */
372 sobel(width, height,
373 gradients, width,
374 directions,width,
375 tmpbuf, width);
376
377 /* non_maximum_suppression() will actually keep & clip what's necessary and
378 * ignore the rest, so we need a clean output buffer */
379 memset(tmpbuf, 0, width * height);
380 non_maximum_suppression(width, height,
381 tmpbuf, width,
382 directions,width,
383 gradients, width);
384
385 /* keep high values, or low values surrounded by high values */
386 double_threshold(edgedetect->low_u8, edgedetect->high_u8,
387 width, height,
388 out->data[p], out->linesize[p],
389 tmpbuf, width);
390
391 if (edgedetect->mode == MODE_COLORMIX) {
392 color_mix(width, height,
393 out->data[p], out->linesize[p],
394 in->data[p], in->linesize[p]);
395 }
396 }
397
398 if (!direct)
399 av_frame_free(&in);
400 return ff_filter_frame(outlink, out);
401 }
402
uninit(AVFilterContext * ctx)403 static av_cold void uninit(AVFilterContext *ctx)
404 {
405 int p;
406 EdgeDetectContext *edgedetect = ctx->priv;
407
408 for (p = 0; p < edgedetect->nb_planes; p++) {
409 struct plane_info *plane = &edgedetect->planes[p];
410 av_freep(&plane->tmpbuf);
411 av_freep(&plane->gradients);
412 av_freep(&plane->directions);
413 }
414 }
415
416 static const AVFilterPad edgedetect_inputs[] = {
417 {
418 .name = "default",
419 .type = AVMEDIA_TYPE_VIDEO,
420 .config_props = config_props,
421 .filter_frame = filter_frame,
422 },
423 { NULL }
424 };
425
426 static const AVFilterPad edgedetect_outputs[] = {
427 {
428 .name = "default",
429 .type = AVMEDIA_TYPE_VIDEO,
430 },
431 { NULL }
432 };
433
434 AVFilter ff_vf_edgedetect = {
435 .name = "edgedetect",
436 .description = NULL_IF_CONFIG_SMALL("Detect and draw edge."),
437 .priv_size = sizeof(EdgeDetectContext),
438 .init = init,
439 .uninit = uninit,
440 .query_formats = query_formats,
441 .inputs = edgedetect_inputs,
442 .outputs = edgedetect_outputs,
443 .priv_class = &edgedetect_class,
444 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
445 };
446