2 * Copyright (c) 2012-2014 Clément Bœsch <u pkh me>
4 * This file is part of FFmpeg.
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.
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.
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
23 * Edge detection filter
25 * @see https://en.wikipedia.org/wiki/Canny_edge_detector
28 #include "libavutil/avassert.h"
29 #include "libavutil/opt.h"
49 struct plane_info planes[3];
52 uint8_t low_u8, high_u8;
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" },
67 AVFILTER_DEFINE_CLASS(edgedetect);
69 static av_cold int init(AVFilterContext *ctx)
71 EdgeDetectContext *edgedetect = ctx->priv;
73 edgedetect->low_u8 = edgedetect->low * 255. + .5;
74 edgedetect->high_u8 = edgedetect->high * 255. + .5;
78 static int query_formats(AVFilterContext *ctx)
80 const EdgeDetectContext *edgedetect = ctx->priv;
82 if (edgedetect->mode == MODE_WIRES) {
83 static const enum AVPixelFormat pix_fmts[] = {AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE};
84 ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
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};
87 ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
94 static int config_props(AVFilterLink *inlink)
97 AVFilterContext *ctx = inlink->dst;
98 EdgeDetectContext *edgedetect = ctx->priv;
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];
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);
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)
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++) {
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
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
142 + src[i+2] * 5) / 159;
145 dst[i + 1] = src[i + 1];
150 memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
157 DIRECTION_HORIZONTAL,
161 static int get_rounded_direction(int gx, int gy)
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
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
174 int tanpi8gx, tan3pi8gx;
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;
185 return DIRECTION_VERTICAL;
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)
195 for (j = 1; j < h - 1; j++) {
199 for (i = 1; i < w - 1; i++) {
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];
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];
209 dst[i] = FFABS(gx) + FFABS(gy);
210 dir[i] = get_rounded_direction(gx, gy);
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)
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]); \
228 for (j = 1; j < h - 1; j++) {
232 for (i = 1; i < w - 1; 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;
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)
249 for (j = 0; j < h; j++) {
250 for (i = 0; i < w; i++) {
256 if ((!i || i == w - 1 || !j || j == h - 1) &&
258 (src[-src_linesize + i-1] > high ||
259 src[-src_linesize + i ] > high ||
260 src[-src_linesize + i+1] > high ||
263 src[ src_linesize + i-1] > high ||
264 src[ src_linesize + i ] > high ||
265 src[ src_linesize + i+1] > high))
275 static void color_mix(int w, int h,
276 uint8_t *dst, int dst_linesize,
277 const uint8_t *src, int src_linesize)
281 for (j = 0; j < h; j++) {
282 for (i = 0; i < w; i++)
283 dst[i] = (dst[i] + src[i]) >> 1;
289 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
291 AVFilterContext *ctx = inlink->dst;
292 EdgeDetectContext *edgedetect = ctx->priv;
293 AVFilterLink *outlink = ctx->outputs[0];
297 if (edgedetect->mode != MODE_COLORMIX && av_frame_is_writable(in)) {
301 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
304 return AVERROR(ENOMEM);
306 av_frame_copy_props(out, in);
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;
315 /* gaussian filter to reduce noise */
316 gaussian_blur(ctx, inlink->w, inlink->h,
318 in->data[p], in->linesize[p]);
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,
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,
331 directions,inlink->w,
332 gradients, inlink->w);
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],
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]);
349 return ff_filter_frame(outlink, out);
352 static av_cold void uninit(AVFilterContext *ctx)
355 EdgeDetectContext *edgedetect = ctx->priv;
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);
365 static const AVFilterPad edgedetect_inputs[] = {
368 .type = AVMEDIA_TYPE_VIDEO,
369 .config_props = config_props,
370 .filter_frame = filter_frame,
375 static const AVFilterPad edgedetect_outputs[] = {
378 .type = AVMEDIA_TYPE_VIDEO,
383 AVFilter ff_vf_edgedetect = {
384 .name = "edgedetect",
385 .description = NULL_IF_CONFIG_SMALL("Detect and draw edge."),
386 .priv_size = sizeof(EdgeDetectContext),
389 .query_formats = query_formats,
390 .inputs = edgedetect_inputs,
391 .outputs = edgedetect_outputs,
392 .priv_class = &edgedetect_class,
393 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,