Browse code

lavfi: add edgedetect filter.

Clément Bœsch authored on 2012/07/27 02:45:53
Showing 8 changed files
... ...
@@ -47,6 +47,7 @@ version next:
47 47
 - hue filter
48 48
 - ICO muxer
49 49
 - SubRip encoder and decoder without embedded timing
50
+- edge detection filter
50 51
 
51 52
 
52 53
 version 0.11:
... ...
@@ -1929,6 +1929,33 @@ For more information about libfreetype, check:
1929 1929
 For more information about fontconfig, check:
1930 1930
 @url{http://freedesktop.org/software/fontconfig/fontconfig-user.html}.
1931 1931
 
1932
+@section edgedetect
1933
+
1934
+Detect and draw edges. The filter uses the Canny Edge Detection algorithm.
1935
+
1936
+This filter accepts the following optional named parameters:
1937
+
1938
+@table @option
1939
+@item low, high
1940
+Set low and high threshold values used by the Canny thresholding
1941
+algorithm.
1942
+
1943
+The high threshold selects the "strong" edge pixels, which are then
1944
+connected through 8-connectivity with the "weak" edge pixels selected
1945
+by the low threshold.
1946
+
1947
+@var{low} and @var{high} threshold values must be choosen in the range
1948
+[0,1], and @var{low} should be lesser or equal to @var{high}.
1949
+
1950
+Default value for @var{low} is @code{20/255}, and default value for @var{high}
1951
+is @code{50/255}.
1952
+@end table
1953
+
1954
+Example:
1955
+@example
1956
+edgedetect=low=0.1:high=0.4
1957
+@end example
1958
+
1932 1959
 @section fade
1933 1960
 
1934 1961
 Apply fade-in/out effect to input video.
... ...
@@ -90,6 +90,7 @@ OBJS-$(CONFIG_DELOGO_FILTER)                 += vf_delogo.o
90 90
 OBJS-$(CONFIG_DESHAKE_FILTER)                += vf_deshake.o
91 91
 OBJS-$(CONFIG_DRAWBOX_FILTER)                += vf_drawbox.o
92 92
 OBJS-$(CONFIG_DRAWTEXT_FILTER)               += vf_drawtext.o
93
+OBJS-$(CONFIG_EDGEDETECT_FILTER)             += vf_edgedetect.o
93 94
 OBJS-$(CONFIG_FADE_FILTER)                   += vf_fade.o
94 95
 OBJS-$(CONFIG_FIELDORDER_FILTER)             += vf_fieldorder.o
95 96
 OBJS-$(CONFIG_FIFO_FILTER)                   += fifo.o
... ...
@@ -81,6 +81,7 @@ void avfilter_register_all(void)
81 81
     REGISTER_FILTER (DESHAKE,     deshake,     vf);
82 82
     REGISTER_FILTER (DRAWBOX,     drawbox,     vf);
83 83
     REGISTER_FILTER (DRAWTEXT,    drawtext,    vf);
84
+    REGISTER_FILTER (EDGEDETECT,  edgedetect,  vf);
84 85
     REGISTER_FILTER (FADE,        fade,        vf);
85 86
     REGISTER_FILTER (FIELDORDER,  fieldorder,  vf);
86 87
     REGISTER_FILTER (FIFO,        fifo,        vf);
... ...
@@ -29,8 +29,8 @@
29 29
 #include "libavutil/avutil.h"
30 30
 
31 31
 #define LIBAVFILTER_VERSION_MAJOR  3
32
-#define LIBAVFILTER_VERSION_MINOR  9
33
-#define LIBAVFILTER_VERSION_MICRO 102
32
+#define LIBAVFILTER_VERSION_MINOR  10
33
+#define LIBAVFILTER_VERSION_MICRO 100
34 34
 
35 35
 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \
36 36
                                                LIBAVFILTER_VERSION_MINOR, \
37 37
new file mode 100644
... ...
@@ -0,0 +1,323 @@
0
+/*
1
+ * Copyright (c) 2012 Clément Bœsch
2
+ *
3
+ * This file is part of FFmpeg.
4
+ *
5
+ * FFmpeg is free software; you can redistribute it and/or
6
+ * modify it under the terms of the GNU Lesser General Public
7
+ * License as published by the Free Software Foundation; either
8
+ * version 2.1 of the License, or (at your option) any later version.
9
+ *
10
+ * FFmpeg is distributed in the hope that it will be useful,
11
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
12
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13
+ * Lesser General Public License for more details.
14
+ *
15
+ * You should have received a copy of the GNU Lesser General Public
16
+ * License along with FFmpeg; if not, write to the Free Software
17
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18
+ */
19
+
20
+/**
21
+ * @file
22
+ * Edge detection filter
23
+ *
24
+ * @url https://en.wikipedia.org/wiki/Canny_edge_detector
25
+ */
26
+
27
+#include "libavutil/opt.h"
28
+#include "avfilter.h"
29
+#include "formats.h"
30
+#include "internal.h"
31
+#include "video.h"
32
+
33
+typedef struct {
34
+    const AVClass *class;
35
+    uint8_t  *tmpbuf;
36
+    uint16_t *gradients;
37
+    char     *directions;
38
+    double   low, high;
39
+    uint8_t  low_u8, high_u8;
40
+} EdgeDetectContext;
41
+
42
+#define OFFSET(x) offsetof(EdgeDetectContext, x)
43
+static const AVOption edgedetect_options[] = {
44
+    { "high", "set high threshold", OFFSET(high), AV_OPT_TYPE_DOUBLE, {.dbl=50/255.}, 0, 1 },
45
+    { "low",  "set low threshold",  OFFSET(low),  AV_OPT_TYPE_DOUBLE, {.dbl=20/255.}, 0, 1 },
46
+    { NULL },
47
+};
48
+
49
+AVFILTER_DEFINE_CLASS(edgedetect);
50
+
51
+static av_cold int init(AVFilterContext *ctx, const char *args)
52
+{
53
+    int ret;
54
+    EdgeDetectContext *edgedetect = ctx->priv;
55
+
56
+    edgedetect->class = &edgedetect_class;
57
+    av_opt_set_defaults(edgedetect);
58
+
59
+    if ((ret = av_set_options_string(edgedetect, args, "=", ":")) < 0)
60
+        return ret;
61
+
62
+    edgedetect->low_u8  = edgedetect->low  * 255.;
63
+    edgedetect->high_u8 = edgedetect->high * 255.;
64
+    return 0;
65
+}
66
+
67
+static int query_formats(AVFilterContext *ctx)
68
+{
69
+    static const enum PixelFormat pix_fmts[] = {PIX_FMT_GRAY8, PIX_FMT_NONE};
70
+    ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
71
+    return 0;
72
+}
73
+
74
+static int config_props(AVFilterLink *inlink)
75
+{
76
+    AVFilterContext *ctx = inlink->dst;
77
+    EdgeDetectContext *edgedetect = ctx->priv;
78
+
79
+    edgedetect->tmpbuf     = av_malloc(inlink->w * inlink->h);
80
+    edgedetect->gradients  = av_calloc(inlink->w * inlink->h, sizeof(*edgedetect->gradients));
81
+    edgedetect->directions = av_malloc(inlink->w * inlink->h);
82
+    if (!edgedetect->tmpbuf || !edgedetect->gradients || !edgedetect->directions)
83
+        return AVERROR(ENOMEM);
84
+    return 0;
85
+}
86
+
87
+static void gaussian_blur(AVFilterContext *ctx, int w, int h,
88
+                                uint8_t *dst, int dst_linesize,
89
+                          const uint8_t *src, int src_linesize)
90
+{
91
+    int i, j;
92
+
93
+    memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
94
+    memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
95
+    for (j = 2; j < h - 2; j++) {
96
+        dst[0] = src[0];
97
+        dst[1] = src[1];
98
+        for (i = 2; i < w - 2; i++) {
99
+            /* Gaussian mask of size 5x5 with sigma = 1.4 */
100
+            dst[i] = ((src[-2*src_linesize + i-2] + src[2*src_linesize + i-2]) * 2
101
+                    + (src[-2*src_linesize + i-1] + src[2*src_linesize + i-1]) * 4
102
+                    + (src[-2*src_linesize + i  ] + src[2*src_linesize + i  ]) * 5
103
+                    + (src[-2*src_linesize + i+1] + src[2*src_linesize + i+1]) * 4
104
+                    + (src[-2*src_linesize + i+2] + src[2*src_linesize + i+2]) * 2
105
+
106
+                    + (src[  -src_linesize + i-2] + src[  src_linesize + i-2]) *  4
107
+                    + (src[  -src_linesize + i-1] + src[  src_linesize + i-1]) *  9
108
+                    + (src[  -src_linesize + i  ] + src[  src_linesize + i  ]) * 12
109
+                    + (src[  -src_linesize + i+1] + src[  src_linesize + i+1]) *  9
110
+                    + (src[  -src_linesize + i+2] + src[  src_linesize + i+2]) *  4
111
+
112
+                    + src[i-2] *  5
113
+                    + src[i-1] * 12
114
+                    + src[i  ] * 15
115
+                    + src[i+1] * 12
116
+                    + src[i+2] *  5) / 159;
117
+        }
118
+        dst[i    ] = src[i    ];
119
+        dst[i + 1] = src[i + 1];
120
+
121
+        dst += dst_linesize;
122
+        src += src_linesize;
123
+    }
124
+    memcpy(dst, src, w); dst += dst_linesize; src += src_linesize;
125
+    memcpy(dst, src, w);
126
+}
127
+
128
+enum {
129
+    DIRECTION_45UP,
130
+    DIRECTION_45DOWN,
131
+    DIRECTION_HORIZONTAL,
132
+    DIRECTION_VERTICAL,
133
+};
134
+
135
+static int get_rounded_direction(int gx, int gy)
136
+{
137
+    /* reference angles:
138
+     *   tan( pi/8) = sqrt(2)-1
139
+     *   tan(3pi/8) = sqrt(2)+1
140
+     * Gy/Gx is the tangent of the angle (theta), so Gy/Gx is compared against
141
+     * <ref-angle>, or more simply Gy against <ref-angle>*Gx
142
+     *
143
+     * Gx and Gy bounds = [1020;1020], using 16-bit arithmetic:
144
+     *   round((sqrt(2)-1) * (1<<16)) =  27146
145
+     *   round((sqrt(2)+1) * (1<<16)) = 158218
146
+     */
147
+    if (gx) {
148
+        int tanpi8gx, tan3pi8gx;
149
+
150
+        if (gx < 0)
151
+            gx = -gx, gy = -gy;
152
+        gy <<= 16;
153
+        tanpi8gx  =  27146 * gx;
154
+        tan3pi8gx = 158218 * gx;
155
+        if (gy > -tan3pi8gx && gy < -tanpi8gx)  return DIRECTION_45UP;
156
+        if (gy > -tanpi8gx  && gy <  tanpi8gx)  return DIRECTION_HORIZONTAL;
157
+        if (gy >  tanpi8gx  && gy <  tan3pi8gx) return DIRECTION_45DOWN;
158
+    }
159
+    return DIRECTION_VERTICAL;
160
+}
161
+
162
+static void sobel(AVFilterContext *ctx, int w, int h,
163
+                        uint16_t *dst, int dst_linesize,
164
+                  const uint8_t  *src, int src_linesize)
165
+{
166
+    int i, j;
167
+    EdgeDetectContext *edgedetect = ctx->priv;
168
+
169
+    for (j = 1; j < h - 1; j++) {
170
+        dst += dst_linesize;
171
+        src += src_linesize;
172
+        for (i = 1; i < w - 1; i++) {
173
+            const int gx =
174
+                -1*src[-src_linesize + i-1] + 1*src[-src_linesize + i+1]
175
+                -2*src[                i-1] + 2*src[                i+1]
176
+                -1*src[ src_linesize + i-1] + 1*src[ src_linesize + i+1];
177
+            const int gy =
178
+                -1*src[-src_linesize + i-1] + 1*src[ src_linesize + i-1]
179
+                -2*src[-src_linesize + i  ] + 2*src[ src_linesize + i  ]
180
+                -1*src[-src_linesize + i+1] + 1*src[ src_linesize + i+1];
181
+
182
+            dst[i] = FFABS(gx) + FFABS(gy);
183
+            edgedetect->directions[j*w + i] = get_rounded_direction(gx, gy);
184
+        }
185
+    }
186
+}
187
+
188
+static void non_maximum_suppression(AVFilterContext *ctx, int w, int h,
189
+                                          uint8_t  *dst, int dst_linesize,
190
+                                    const uint16_t *src, int src_linesize)
191
+{
192
+    int i, j;
193
+    EdgeDetectContext *edgedetect = ctx->priv;
194
+
195
+#define COPY_MAXIMA(ay, ax, by, bx) do {                \
196
+    if (src[i] > src[(ay)*src_linesize + i+(ax)] &&     \
197
+        src[i] > src[(by)*src_linesize + i+(bx)])       \
198
+        dst[i] = av_clip_uint8(src[i]);                 \
199
+} while (0)
200
+
201
+    for (j = 1; j < h - 1; j++) {
202
+        dst += dst_linesize;
203
+        src += src_linesize;
204
+        for (i = 1; i < w - 1; i++) {
205
+            switch (edgedetect->directions[j*w + i]) {
206
+            case DIRECTION_45UP:        COPY_MAXIMA( 1, -1, -1,  1); break;
207
+            case DIRECTION_45DOWN:      COPY_MAXIMA(-1, -1,  1,  1); break;
208
+            case DIRECTION_HORIZONTAL:  COPY_MAXIMA( 0, -1,  0,  1); break;
209
+            case DIRECTION_VERTICAL:    COPY_MAXIMA(-1,  0,  1,  0); break;
210
+            }
211
+        }
212
+    }
213
+}
214
+
215
+static void double_threshold(AVFilterContext *ctx, int w, int h,
216
+                                   uint8_t *dst, int dst_linesize,
217
+                             const uint8_t *src, int src_linesize)
218
+{
219
+    int i, j;
220
+    EdgeDetectContext *edgedetect = ctx->priv;
221
+    const int low  = edgedetect->low_u8;
222
+    const int high = edgedetect->high_u8;
223
+
224
+    for (j = 0; j < h; j++) {
225
+        for (i = 0; i < w; i++) {
226
+            if (src[i] > high) {
227
+                dst[i] = src[i];
228
+                continue;
229
+            }
230
+
231
+            if ((!i || i == w - 1 || !j || j == h - 1) &&
232
+                src[i] > low &&
233
+                (src[-src_linesize + i-1] > high ||
234
+                 src[-src_linesize + i  ] > high ||
235
+                 src[-src_linesize + i+1] > high ||
236
+                 src[                i-1] > high ||
237
+                 src[                i+1] > high ||
238
+                 src[ src_linesize + i-1] > high ||
239
+                 src[ src_linesize + i  ] > high ||
240
+                 src[ src_linesize + i+1] > high))
241
+                dst[i] = src[i];
242
+            else
243
+                dst[i] = 0;
244
+        }
245
+        dst += dst_linesize;
246
+        src += src_linesize;
247
+    }
248
+}
249
+
250
+static int end_frame(AVFilterLink *inlink)
251
+{
252
+    AVFilterContext *ctx = inlink->dst;
253
+    EdgeDetectContext *edgedetect = ctx->priv;
254
+    AVFilterLink *outlink = inlink->dst->outputs[0];
255
+    AVFilterBufferRef  *inpicref = inlink->cur_buf;
256
+    AVFilterBufferRef *outpicref = outlink->out_buf;
257
+    uint8_t  *tmpbuf    = edgedetect->tmpbuf;
258
+    uint16_t *gradients = edgedetect->gradients;
259
+
260
+    /* gaussian filter to reduce noise  */
261
+    gaussian_blur(ctx, inlink->w, inlink->h,
262
+                  tmpbuf,            inlink->w,
263
+                  inpicref->data[0], inpicref->linesize[0]);
264
+
265
+    /* compute the 16-bits gradients and directions for the next step */
266
+    sobel(ctx, inlink->w, inlink->h,
267
+          gradients, inlink->w,
268
+          tmpbuf,    inlink->w);
269
+
270
+    /* non_maximum_suppression() will actually keep & clip what's necessary and
271
+     * ignore the rest, so we need a clean output buffer */
272
+    memset(tmpbuf, 0, inlink->w * inlink->h);
273
+    non_maximum_suppression(ctx, inlink->w, inlink->h,
274
+                            tmpbuf,    inlink->w,
275
+                            gradients, inlink->w);
276
+
277
+    /* keep high values, or low values surrounded by high values */
278
+    double_threshold(ctx, inlink->w, inlink->h,
279
+                     outpicref->data[0], outpicref->linesize[0],
280
+                     tmpbuf,             inlink->w);
281
+
282
+    ff_draw_slice(outlink, 0, outlink->h, 1);
283
+    return ff_end_frame(outlink);
284
+}
285
+
286
+static av_cold void uninit(AVFilterContext *ctx)
287
+{
288
+    EdgeDetectContext *edgedetect = ctx->priv;
289
+    av_freep(&edgedetect->tmpbuf);
290
+    av_freep(&edgedetect->gradients);
291
+    av_freep(&edgedetect->directions);
292
+}
293
+
294
+static int null_draw_slice(AVFilterLink *inlink, int y, int h, int slice_dir) { return 0; }
295
+
296
+AVFilter avfilter_vf_edgedetect = {
297
+    .name          = "edgedetect",
298
+    .description   = NULL_IF_CONFIG_SMALL("Detect and draw edge."),
299
+    .priv_size     = sizeof(EdgeDetectContext),
300
+    .init          = init,
301
+    .uninit        = uninit,
302
+    .query_formats = query_formats,
303
+
304
+    .inputs    = (const AVFilterPad[]) {
305
+       {
306
+           .name             = "default",
307
+           .type             = AVMEDIA_TYPE_VIDEO,
308
+           .draw_slice       = null_draw_slice,
309
+           .config_props     = config_props,
310
+           .end_frame        = end_frame,
311
+           .min_perms        = AV_PERM_READ
312
+        },
313
+        { .name = NULL }
314
+    },
315
+    .outputs   = (const AVFilterPad[]) {
316
+        {
317
+            .name            = "default",
318
+            .type            = AVMEDIA_TYPE_VIDEO,
319
+        },
320
+        { .name = NULL }
321
+    },
322
+};
... ...
@@ -42,6 +42,7 @@ do_lavfi "crop_scale"         "crop=iw-100:ih-100:100:100,scale=400:-1"
42 42
 do_lavfi "crop_scale_vflip"   "null,null,crop=iw-200:ih-200:200:200,crop=iw-20:ih-20:20:20,scale=200:200,scale=250:250,vflip,vflip,null,scale=200:200,crop=iw-100:ih-100:100:100,vflip,scale=200:200,null,vflip,crop=iw-100:ih-100:100:100,null"
43 43
 do_lavfi "crop_vflip"         "crop=iw-100:ih-100:100:100,vflip"
44 44
 do_lavfi "drawbox"            "drawbox=224:24:88:72:#FF8010@0.5"
45
+do_lavfi "edgedetect"         "edgedetect"
45 46
 do_lavfi "fade"               "fade=in:5:15,fade=out:30:15"
46 47
 do_lavfi "null"               "null"
47 48
 do_lavfi "overlay"            "split[m],scale=88:72,pad=96:80:4:4[o2];[m]fifo[o1],[o1][o2]overlay=240:16"
48 49
new file mode 100644
... ...
@@ -0,0 +1 @@
0
+edgedetect          16bce636eef1a82e18837d176f4187c1