blob: ac88e02a11b5f441f2c3c399b47e6533a58e2cac
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 { |
36 | MODE_WIRES, |
37 | MODE_COLORMIX, |
38 | NB_MODE |
39 | }; |
40 | |
41 | struct plane_info { |
42 | uint8_t *tmpbuf; |
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; |
53 | int mode; |
54 | } EdgeDetectContext; |
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 | |
78 | static int query_formats(AVFilterContext *ctx) |
79 | { |
80 | const EdgeDetectContext *edgedetect = ctx->priv; |
81 | static const enum AVPixelFormat wires_pix_fmts[] = {AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE}; |
82 | static const enum AVPixelFormat colormix_pix_fmts[] = {AV_PIX_FMT_GBRP, AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE}; |
83 | AVFilterFormats *fmts_list; |
84 | const enum AVPixelFormat *pix_fmts = NULL; |
85 | |
86 | if (edgedetect->mode == MODE_WIRES) { |
87 | pix_fmts = wires_pix_fmts; |
88 | } else if (edgedetect->mode == MODE_COLORMIX) { |
89 | pix_fmts = colormix_pix_fmts; |
90 | } else { |
91 | av_assert0(0); |
92 | } |
93 | fmts_list = ff_make_format_list(pix_fmts); |
94 | if (!fmts_list) |
95 | return AVERROR(ENOMEM); |
96 | return ff_set_common_formats(ctx, fmts_list); |
97 | } |
98 | |
99 | static int config_props(AVFilterLink *inlink) |
100 | { |
101 | int p; |
102 | AVFilterContext *ctx = inlink->dst; |
103 | EdgeDetectContext *edgedetect = ctx->priv; |
104 | |
105 | edgedetect->nb_planes = inlink->format == AV_PIX_FMT_GRAY8 ? 1 : 3; |
106 | for (p = 0; p < edgedetect->nb_planes; p++) { |
107 | struct plane_info *plane = &edgedetect->planes[p]; |
108 | |
109 | plane->tmpbuf = av_malloc(inlink->w * inlink->h); |
110 | plane->gradients = av_calloc(inlink->w * inlink->h, sizeof(*plane->gradients)); |
111 | plane->directions = av_malloc(inlink->w * inlink->h); |
112 | if (!plane->tmpbuf || !plane->gradients || !plane->directions) |
113 | return AVERROR(ENOMEM); |
114 | } |
115 | return 0; |
116 | } |
117 | |
118 | static void gaussian_blur(AVFilterContext *ctx, int w, int h, |
119 | uint8_t *dst, int dst_linesize, |
120 | const uint8_t *src, int src_linesize) |
121 | { |
122 | int i, j; |
123 | |
124 | memcpy(dst, src, w); dst += dst_linesize; src += src_linesize; |
125 | memcpy(dst, src, w); dst += dst_linesize; src += src_linesize; |
126 | for (j = 2; j < h - 2; j++) { |
127 | dst[0] = src[0]; |
128 | dst[1] = src[1]; |
129 | for (i = 2; i < w - 2; i++) { |
130 | /* Gaussian mask of size 5x5 with sigma = 1.4 */ |
131 | dst[i] = ((src[-2*src_linesize + i-2] + src[2*src_linesize + i-2]) * 2 |
132 | + (src[-2*src_linesize + i-1] + src[2*src_linesize + i-1]) * 4 |
133 | + (src[-2*src_linesize + i ] + src[2*src_linesize + i ]) * 5 |
134 | + (src[-2*src_linesize + i+1] + src[2*src_linesize + i+1]) * 4 |
135 | + (src[-2*src_linesize + i+2] + src[2*src_linesize + i+2]) * 2 |
136 | |
137 | + (src[ -src_linesize + i-2] + src[ src_linesize + i-2]) * 4 |
138 | + (src[ -src_linesize + i-1] + src[ src_linesize + i-1]) * 9 |
139 | + (src[ -src_linesize + i ] + src[ src_linesize + i ]) * 12 |
140 | + (src[ -src_linesize + i+1] + src[ src_linesize + i+1]) * 9 |
141 | + (src[ -src_linesize + i+2] + src[ src_linesize + i+2]) * 4 |
142 | |
143 | + src[i-2] * 5 |
144 | + src[i-1] * 12 |
145 | + src[i ] * 15 |
146 | + src[i+1] * 12 |
147 | + src[i+2] * 5) / 159; |
148 | } |
149 | dst[i ] = src[i ]; |
150 | dst[i + 1] = src[i + 1]; |
151 | |
152 | dst += dst_linesize; |
153 | src += src_linesize; |
154 | } |
155 | memcpy(dst, src, w); dst += dst_linesize; src += src_linesize; |
156 | memcpy(dst, src, w); |
157 | } |
158 | |
159 | enum { |
160 | DIRECTION_45UP, |
161 | DIRECTION_45DOWN, |
162 | DIRECTION_HORIZONTAL, |
163 | DIRECTION_VERTICAL, |
164 | }; |
165 | |
166 | static int get_rounded_direction(int gx, int gy) |
167 | { |
168 | /* reference angles: |
169 | * tan( pi/8) = sqrt(2)-1 |
170 | * tan(3pi/8) = sqrt(2)+1 |
171 | * Gy/Gx is the tangent of the angle (theta), so Gy/Gx is compared against |
172 | * <ref-angle>, or more simply Gy against <ref-angle>*Gx |
173 | * |
174 | * Gx and Gy bounds = [-1020;1020], using 16-bit arithmetic: |
175 | * round((sqrt(2)-1) * (1<<16)) = 27146 |
176 | * round((sqrt(2)+1) * (1<<16)) = 158218 |
177 | */ |
178 | if (gx) { |
179 | int tanpi8gx, tan3pi8gx; |
180 | |
181 | if (gx < 0) |
182 | gx = -gx, gy = -gy; |
183 | gy <<= 16; |
184 | tanpi8gx = 27146 * gx; |
185 | tan3pi8gx = 158218 * gx; |
186 | if (gy > -tan3pi8gx && gy < -tanpi8gx) return DIRECTION_45UP; |
187 | if (gy > -tanpi8gx && gy < tanpi8gx) return DIRECTION_HORIZONTAL; |
188 | if (gy > tanpi8gx && gy < tan3pi8gx) return DIRECTION_45DOWN; |
189 | } |
190 | return DIRECTION_VERTICAL; |
191 | } |
192 | |
193 | static void sobel(int w, int h, |
194 | uint16_t *dst, int dst_linesize, |
195 | int8_t *dir, int dir_linesize, |
196 | const uint8_t *src, int src_linesize) |
197 | { |
198 | int i, j; |
199 | |
200 | for (j = 1; j < h - 1; j++) { |
201 | dst += dst_linesize; |
202 | dir += dir_linesize; |
203 | src += src_linesize; |
204 | for (i = 1; i < w - 1; i++) { |
205 | const int gx = |
206 | -1*src[-src_linesize + i-1] + 1*src[-src_linesize + i+1] |
207 | -2*src[ i-1] + 2*src[ i+1] |
208 | -1*src[ src_linesize + i-1] + 1*src[ src_linesize + i+1]; |
209 | const int gy = |
210 | -1*src[-src_linesize + i-1] + 1*src[ src_linesize + i-1] |
211 | -2*src[-src_linesize + i ] + 2*src[ src_linesize + i ] |
212 | -1*src[-src_linesize + i+1] + 1*src[ src_linesize + i+1]; |
213 | |
214 | dst[i] = FFABS(gx) + FFABS(gy); |
215 | dir[i] = get_rounded_direction(gx, gy); |
216 | } |
217 | } |
218 | } |
219 | |
220 | static void non_maximum_suppression(int w, int h, |
221 | uint8_t *dst, int dst_linesize, |
222 | const int8_t *dir, int dir_linesize, |
223 | const uint16_t *src, int src_linesize) |
224 | { |
225 | int i, j; |
226 | |
227 | #define COPY_MAXIMA(ay, ax, by, bx) do { \ |
228 | if (src[i] > src[(ay)*src_linesize + i+(ax)] && \ |
229 | src[i] > src[(by)*src_linesize + i+(bx)]) \ |
230 | dst[i] = av_clip_uint8(src[i]); \ |
231 | } while (0) |
232 | |
233 | for (j = 1; j < h - 1; j++) { |
234 | dst += dst_linesize; |
235 | dir += dir_linesize; |
236 | src += src_linesize; |
237 | for (i = 1; i < w - 1; i++) { |
238 | switch (dir[i]) { |
239 | case DIRECTION_45UP: COPY_MAXIMA( 1, -1, -1, 1); break; |
240 | case DIRECTION_45DOWN: COPY_MAXIMA(-1, -1, 1, 1); break; |
241 | case DIRECTION_HORIZONTAL: COPY_MAXIMA( 0, -1, 0, 1); break; |
242 | case DIRECTION_VERTICAL: COPY_MAXIMA(-1, 0, 1, 0); break; |
243 | } |
244 | } |
245 | } |
246 | } |
247 | |
248 | static void double_threshold(int low, int high, int w, int h, |
249 | uint8_t *dst, int dst_linesize, |
250 | const uint8_t *src, int src_linesize) |
251 | { |
252 | int i, j; |
253 | |
254 | for (j = 0; j < h; j++) { |
255 | for (i = 0; i < w; i++) { |
256 | if (src[i] > high) { |
257 | dst[i] = src[i]; |
258 | continue; |
259 | } |
260 | |
261 | if ((!i || i == w - 1 || !j || j == h - 1) && |
262 | src[i] > low && |
263 | (src[-src_linesize + i-1] > high || |
264 | src[-src_linesize + i ] > high || |
265 | src[-src_linesize + i+1] > high || |
266 | src[ i-1] > high || |
267 | src[ i+1] > high || |
268 | src[ src_linesize + i-1] > high || |
269 | src[ src_linesize + i ] > high || |
270 | src[ src_linesize + i+1] > high)) |
271 | dst[i] = src[i]; |
272 | else |
273 | dst[i] = 0; |
274 | } |
275 | dst += dst_linesize; |
276 | src += src_linesize; |
277 | } |
278 | } |
279 | |
280 | static void color_mix(int w, int h, |
281 | uint8_t *dst, int dst_linesize, |
282 | const uint8_t *src, int src_linesize) |
283 | { |
284 | int i, j; |
285 | |
286 | for (j = 0; j < h; j++) { |
287 | for (i = 0; i < w; i++) |
288 | dst[i] = (dst[i] + src[i]) >> 1; |
289 | dst += dst_linesize; |
290 | src += src_linesize; |
291 | } |
292 | } |
293 | |
294 | static int filter_frame(AVFilterLink *inlink, AVFrame *in) |
295 | { |
296 | AVFilterContext *ctx = inlink->dst; |
297 | EdgeDetectContext *edgedetect = ctx->priv; |
298 | AVFilterLink *outlink = ctx->outputs[0]; |
299 | int p, direct = 0; |
300 | AVFrame *out; |
301 | |
302 | if (edgedetect->mode != MODE_COLORMIX && av_frame_is_writable(in)) { |
303 | direct = 1; |
304 | out = in; |
305 | } else { |
306 | out = ff_get_video_buffer(outlink, outlink->w, outlink->h); |
307 | if (!out) { |
308 | av_frame_free(&in); |
309 | return AVERROR(ENOMEM); |
310 | } |
311 | av_frame_copy_props(out, in); |
312 | } |
313 | |
314 | for (p = 0; p < edgedetect->nb_planes; p++) { |
315 | struct plane_info *plane = &edgedetect->planes[p]; |
316 | uint8_t *tmpbuf = plane->tmpbuf; |
317 | uint16_t *gradients = plane->gradients; |
318 | int8_t *directions = plane->directions; |
319 | |
320 | /* gaussian filter to reduce noise */ |
321 | gaussian_blur(ctx, inlink->w, inlink->h, |
322 | tmpbuf, inlink->w, |
323 | in->data[p], in->linesize[p]); |
324 | |
325 | /* compute the 16-bits gradients and directions for the next step */ |
326 | sobel(inlink->w, inlink->h, |
327 | gradients, inlink->w, |
328 | directions,inlink->w, |
329 | tmpbuf, inlink->w); |
330 | |
331 | /* non_maximum_suppression() will actually keep & clip what's necessary and |
332 | * ignore the rest, so we need a clean output buffer */ |
333 | memset(tmpbuf, 0, inlink->w * inlink->h); |
334 | non_maximum_suppression(inlink->w, inlink->h, |
335 | tmpbuf, inlink->w, |
336 | directions,inlink->w, |
337 | gradients, inlink->w); |
338 | |
339 | /* keep high values, or low values surrounded by high values */ |
340 | double_threshold(edgedetect->low_u8, edgedetect->high_u8, |
341 | inlink->w, inlink->h, |
342 | out->data[p], out->linesize[p], |
343 | tmpbuf, inlink->w); |
344 | |
345 | if (edgedetect->mode == MODE_COLORMIX) { |
346 | color_mix(inlink->w, inlink->h, |
347 | out->data[p], out->linesize[p], |
348 | in->data[p], in->linesize[p]); |
349 | } |
350 | } |
351 | |
352 | if (!direct) |
353 | av_frame_free(&in); |
354 | return ff_filter_frame(outlink, out); |
355 | } |
356 | |
357 | static av_cold void uninit(AVFilterContext *ctx) |
358 | { |
359 | int p; |
360 | EdgeDetectContext *edgedetect = ctx->priv; |
361 | |
362 | for (p = 0; p < edgedetect->nb_planes; p++) { |
363 | struct plane_info *plane = &edgedetect->planes[p]; |
364 | av_freep(&plane->tmpbuf); |
365 | av_freep(&plane->gradients); |
366 | av_freep(&plane->directions); |
367 | } |
368 | } |
369 | |
370 | static const AVFilterPad edgedetect_inputs[] = { |
371 | { |
372 | .name = "default", |
373 | .type = AVMEDIA_TYPE_VIDEO, |
374 | .config_props = config_props, |
375 | .filter_frame = filter_frame, |
376 | }, |
377 | { NULL } |
378 | }; |
379 | |
380 | static const AVFilterPad edgedetect_outputs[] = { |
381 | { |
382 | .name = "default", |
383 | .type = AVMEDIA_TYPE_VIDEO, |
384 | }, |
385 | { NULL } |
386 | }; |
387 | |
388 | AVFilter ff_vf_edgedetect = { |
389 | .name = "edgedetect", |
390 | .description = NULL_IF_CONFIG_SMALL("Detect and draw edge."), |
391 | .priv_size = sizeof(EdgeDetectContext), |
392 | .init = init, |
393 | .uninit = uninit, |
394 | .query_formats = query_formats, |
395 | .inputs = edgedetect_inputs, |
396 | .outputs = edgedetect_outputs, |
397 | .priv_class = &edgedetect_class, |
398 | .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC, |
399 | }; |
400 |