2 * Copyright (c) 2012 Clément Bœsch
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/opt.h"
40 uint8_t low_u8, high_u8;
43 #define OFFSET(x) offsetof(EdgeDetectContext, x)
44 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
45 static const AVOption edgedetect_options[] = {
46 { "high", "set high threshold", OFFSET(high), AV_OPT_TYPE_DOUBLE, {.dbl=50/255.}, 0, 1, FLAGS },
47 { "low", "set low threshold", OFFSET(low), AV_OPT_TYPE_DOUBLE, {.dbl=20/255.}, 0, 1, FLAGS },
51 AVFILTER_DEFINE_CLASS(edgedetect);
53 static av_cold int init(AVFilterContext *ctx, const char *args)
56 EdgeDetectContext *edgedetect = ctx->priv;
58 edgedetect->class = &edgedetect_class;
59 av_opt_set_defaults(edgedetect);
61 if ((ret = av_set_options_string(edgedetect, args, "=", ":")) < 0)
64 edgedetect->low_u8 = edgedetect->low * 255. + .5;
65 edgedetect->high_u8 = edgedetect->high * 255. + .5;
69 static int query_formats(AVFilterContext *ctx)
71 static const enum AVPixelFormat pix_fmts[] = {AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE};
72 ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
76 static int config_props(AVFilterLink *inlink)
78 AVFilterContext *ctx = inlink->dst;
79 EdgeDetectContext *edgedetect = ctx->priv;
81 edgedetect->tmpbuf = av_malloc(inlink->w * inlink->h);
82 edgedetect->gradients = av_calloc(inlink->w * inlink->h, sizeof(*edgedetect->gradients));
83 edgedetect->directions = av_malloc(inlink->w * inlink->h);
84 if (!edgedetect->tmpbuf || !edgedetect->gradients || !edgedetect->directions)
85 return AVERROR(ENOMEM);
89 static void gaussian_blur(AVFilterContext *ctx, int w, int h,
90 uint8_t *dst, int dst_linesize,
91 const uint8_t *src, int src_linesize)
95 memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
96 memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
97 for (j = 2; j < h - 2; j++) {
100 for (i = 2; i < w - 2; i++) {
101 /* Gaussian mask of size 5x5 with sigma = 1.4 */
102 dst[i] = ((src[-2*src_linesize + i-2] + src[2*src_linesize + i-2]) * 2
103 + (src[-2*src_linesize + i-1] + src[2*src_linesize + i-1]) * 4
104 + (src[-2*src_linesize + i ] + src[2*src_linesize + i ]) * 5
105 + (src[-2*src_linesize + i+1] + src[2*src_linesize + i+1]) * 4
106 + (src[-2*src_linesize + i+2] + src[2*src_linesize + i+2]) * 2
108 + (src[ -src_linesize + i-2] + src[ src_linesize + i-2]) * 4
109 + (src[ -src_linesize + i-1] + src[ src_linesize + i-1]) * 9
110 + (src[ -src_linesize + i ] + src[ src_linesize + i ]) * 12
111 + (src[ -src_linesize + i+1] + src[ src_linesize + i+1]) * 9
112 + (src[ -src_linesize + i+2] + src[ src_linesize + i+2]) * 4
118 + src[i+2] * 5) / 159;
121 dst[i + 1] = src[i + 1];
126 memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
133 DIRECTION_HORIZONTAL,
137 static int get_rounded_direction(int gx, int gy)
140 * tan( pi/8) = sqrt(2)-1
141 * tan(3pi/8) = sqrt(2)+1
142 * Gy/Gx is the tangent of the angle (theta), so Gy/Gx is compared against
143 * <ref-angle>, or more simply Gy against <ref-angle>*Gx
145 * Gx and Gy bounds = [-1020;1020], using 16-bit arithmetic:
146 * round((sqrt(2)-1) * (1<<16)) = 27146
147 * round((sqrt(2)+1) * (1<<16)) = 158218
150 int tanpi8gx, tan3pi8gx;
155 tanpi8gx = 27146 * gx;
156 tan3pi8gx = 158218 * gx;
157 if (gy > -tan3pi8gx && gy < -tanpi8gx) return DIRECTION_45UP;
158 if (gy > -tanpi8gx && gy < tanpi8gx) return DIRECTION_HORIZONTAL;
159 if (gy > tanpi8gx && gy < tan3pi8gx) return DIRECTION_45DOWN;
161 return DIRECTION_VERTICAL;
164 static void sobel(AVFilterContext *ctx, int w, int h,
165 uint16_t *dst, int dst_linesize,
166 const uint8_t *src, int src_linesize)
169 EdgeDetectContext *edgedetect = ctx->priv;
171 for (j = 1; j < h - 1; j++) {
174 for (i = 1; i < w - 1; i++) {
176 -1*src[-src_linesize + i-1] + 1*src[-src_linesize + i+1]
177 -2*src[ i-1] + 2*src[ i+1]
178 -1*src[ src_linesize + i-1] + 1*src[ src_linesize + i+1];
180 -1*src[-src_linesize + i-1] + 1*src[ src_linesize + i-1]
181 -2*src[-src_linesize + i ] + 2*src[ src_linesize + i ]
182 -1*src[-src_linesize + i+1] + 1*src[ src_linesize + i+1];
184 dst[i] = FFABS(gx) + FFABS(gy);
185 edgedetect->directions[j*w + i] = get_rounded_direction(gx, gy);
190 static void non_maximum_suppression(AVFilterContext *ctx, int w, int h,
191 uint8_t *dst, int dst_linesize,
192 const uint16_t *src, int src_linesize)
195 EdgeDetectContext *edgedetect = ctx->priv;
197 #define COPY_MAXIMA(ay, ax, by, bx) do { \
198 if (src[i] > src[(ay)*src_linesize + i+(ax)] && \
199 src[i] > src[(by)*src_linesize + i+(bx)]) \
200 dst[i] = av_clip_uint8(src[i]); \
203 for (j = 1; j < h - 1; j++) {
206 for (i = 1; i < w - 1; i++) {
207 switch (edgedetect->directions[j*w + i]) {
208 case DIRECTION_45UP: COPY_MAXIMA( 1, -1, -1, 1); break;
209 case DIRECTION_45DOWN: COPY_MAXIMA(-1, -1, 1, 1); break;
210 case DIRECTION_HORIZONTAL: COPY_MAXIMA( 0, -1, 0, 1); break;
211 case DIRECTION_VERTICAL: COPY_MAXIMA(-1, 0, 1, 0); break;
217 static void double_threshold(AVFilterContext *ctx, int w, int h,
218 uint8_t *dst, int dst_linesize,
219 const uint8_t *src, int src_linesize)
222 EdgeDetectContext *edgedetect = ctx->priv;
223 const int low = edgedetect->low_u8;
224 const int high = edgedetect->high_u8;
226 for (j = 0; j < h; j++) {
227 for (i = 0; i < w; i++) {
233 if ((!i || i == w - 1 || !j || j == h - 1) &&
235 (src[-src_linesize + i-1] > high ||
236 src[-src_linesize + i ] > high ||
237 src[-src_linesize + i+1] > high ||
240 src[ src_linesize + i-1] > high ||
241 src[ src_linesize + i ] > high ||
242 src[ src_linesize + i+1] > high))
252 static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *in)
254 AVFilterContext *ctx = inlink->dst;
255 EdgeDetectContext *edgedetect = ctx->priv;
256 AVFilterLink *outlink = inlink->dst->outputs[0];
257 uint8_t *tmpbuf = edgedetect->tmpbuf;
258 uint16_t *gradients = edgedetect->gradients;
259 AVFilterBufferRef *out;
261 out = ff_get_video_buffer(outlink, AV_PERM_WRITE, outlink->w, outlink->h);
263 avfilter_unref_bufferp(&in);
264 return AVERROR(ENOMEM);
266 avfilter_copy_buffer_ref_props(out, in);
268 /* gaussian filter to reduce noise */
269 gaussian_blur(ctx, inlink->w, inlink->h,
271 in->data[0], in->linesize[0]);
273 /* compute the 16-bits gradients and directions for the next step */
274 sobel(ctx, inlink->w, inlink->h,
275 gradients, inlink->w,
278 /* non_maximum_suppression() will actually keep & clip what's necessary and
279 * ignore the rest, so we need a clean output buffer */
280 memset(tmpbuf, 0, inlink->w * inlink->h);
281 non_maximum_suppression(ctx, inlink->w, inlink->h,
283 gradients, inlink->w);
285 /* keep high values, or low values surrounded by high values */
286 double_threshold(ctx, inlink->w, inlink->h,
287 out->data[0], out->linesize[0],
290 avfilter_unref_bufferp(&in);
291 return ff_filter_frame(outlink, out);
294 static av_cold void uninit(AVFilterContext *ctx)
296 EdgeDetectContext *edgedetect = ctx->priv;
297 av_freep(&edgedetect->tmpbuf);
298 av_freep(&edgedetect->gradients);
299 av_freep(&edgedetect->directions);
302 static const AVFilterPad edgedetect_inputs[] = {
305 .type = AVMEDIA_TYPE_VIDEO,
306 .config_props = config_props,
307 .filter_frame = filter_frame,
308 .min_perms = AV_PERM_READ,
313 static const AVFilterPad edgedetect_outputs[] = {
316 .type = AVMEDIA_TYPE_VIDEO,
321 AVFilter avfilter_vf_edgedetect = {
322 .name = "edgedetect",
323 .description = NULL_IF_CONFIG_SMALL("Detect and draw edge."),
324 .priv_size = sizeof(EdgeDetectContext),
327 .query_formats = query_formats,
328 .inputs = edgedetect_inputs,
329 .outputs = edgedetect_outputs,
330 .priv_class = &edgedetect_class,