blob: 8128030b8c9b7827af3ffabcf2f07d6f17eead70
1 | /* |
2 | * Copyright (c) 2010 Stefano Sabatini |
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 | * libopencv wrapper functions |
24 | */ |
25 | |
26 | #include "config.h" |
27 | #if HAVE_OPENCV2_CORE_CORE_C_H |
28 | #include <opencv2/core/core_c.h> |
29 | #include <opencv2/imgproc/imgproc_c.h> |
30 | #else |
31 | #include <opencv/cv.h> |
32 | #include <opencv/cxcore.h> |
33 | #endif |
34 | #include "libavutil/avstring.h" |
35 | #include "libavutil/common.h" |
36 | #include "libavutil/file.h" |
37 | #include "libavutil/opt.h" |
38 | #include "avfilter.h" |
39 | #include "formats.h" |
40 | #include "internal.h" |
41 | #include "video.h" |
42 | |
43 | static void fill_iplimage_from_frame(IplImage *img, const AVFrame *frame, enum AVPixelFormat pixfmt) |
44 | { |
45 | IplImage *tmpimg; |
46 | int depth, channels_nb; |
47 | |
48 | if (pixfmt == AV_PIX_FMT_GRAY8) { depth = IPL_DEPTH_8U; channels_nb = 1; } |
49 | else if (pixfmt == AV_PIX_FMT_BGRA) { depth = IPL_DEPTH_8U; channels_nb = 4; } |
50 | else if (pixfmt == AV_PIX_FMT_BGR24) { depth = IPL_DEPTH_8U; channels_nb = 3; } |
51 | else return; |
52 | |
53 | tmpimg = cvCreateImageHeader((CvSize){frame->width, frame->height}, depth, channels_nb); |
54 | *img = *tmpimg; |
55 | img->imageData = img->imageDataOrigin = frame->data[0]; |
56 | img->dataOrder = IPL_DATA_ORDER_PIXEL; |
57 | img->origin = IPL_ORIGIN_TL; |
58 | img->widthStep = frame->linesize[0]; |
59 | } |
60 | |
61 | static void fill_frame_from_iplimage(AVFrame *frame, const IplImage *img, enum AVPixelFormat pixfmt) |
62 | { |
63 | frame->linesize[0] = img->widthStep; |
64 | frame->data[0] = img->imageData; |
65 | } |
66 | |
67 | static int query_formats(AVFilterContext *ctx) |
68 | { |
69 | static const enum AVPixelFormat pix_fmts[] = { |
70 | AV_PIX_FMT_BGR24, AV_PIX_FMT_BGRA, AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE |
71 | }; |
72 | AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts); |
73 | if (!fmts_list) |
74 | return AVERROR(ENOMEM); |
75 | return ff_set_common_formats(ctx, fmts_list); |
76 | } |
77 | |
78 | typedef struct OCVContext { |
79 | const AVClass *class; |
80 | char *name; |
81 | char *params; |
82 | int (*init)(AVFilterContext *ctx, const char *args); |
83 | void (*uninit)(AVFilterContext *ctx); |
84 | void (*end_frame_filter)(AVFilterContext *ctx, IplImage *inimg, IplImage *outimg); |
85 | void *priv; |
86 | } OCVContext; |
87 | |
88 | typedef struct SmoothContext { |
89 | int type; |
90 | int param1, param2; |
91 | double param3, param4; |
92 | } SmoothContext; |
93 | |
94 | static av_cold int smooth_init(AVFilterContext *ctx, const char *args) |
95 | { |
96 | OCVContext *s = ctx->priv; |
97 | SmoothContext *smooth = s->priv; |
98 | char type_str[128] = "gaussian"; |
99 | |
100 | smooth->param1 = 3; |
101 | smooth->param2 = 0; |
102 | smooth->param3 = 0.0; |
103 | smooth->param4 = 0.0; |
104 | |
105 | if (args) |
106 | sscanf(args, "%127[^|]|%d|%d|%lf|%lf", type_str, &smooth->param1, &smooth->param2, &smooth->param3, &smooth->param4); |
107 | |
108 | if (!strcmp(type_str, "blur" )) smooth->type = CV_BLUR; |
109 | else if (!strcmp(type_str, "blur_no_scale")) smooth->type = CV_BLUR_NO_SCALE; |
110 | else if (!strcmp(type_str, "median" )) smooth->type = CV_MEDIAN; |
111 | else if (!strcmp(type_str, "gaussian" )) smooth->type = CV_GAUSSIAN; |
112 | else if (!strcmp(type_str, "bilateral" )) smooth->type = CV_BILATERAL; |
113 | else { |
114 | av_log(ctx, AV_LOG_ERROR, "Smoothing type '%s' unknown.\n", type_str); |
115 | return AVERROR(EINVAL); |
116 | } |
117 | |
118 | if (smooth->param1 < 0 || !(smooth->param1%2)) { |
119 | av_log(ctx, AV_LOG_ERROR, |
120 | "Invalid value '%d' for param1, it has to be a positive odd number\n", |
121 | smooth->param1); |
122 | return AVERROR(EINVAL); |
123 | } |
124 | if ((smooth->type == CV_BLUR || smooth->type == CV_BLUR_NO_SCALE || smooth->type == CV_GAUSSIAN) && |
125 | (smooth->param2 < 0 || (smooth->param2 && !(smooth->param2%2)))) { |
126 | av_log(ctx, AV_LOG_ERROR, |
127 | "Invalid value '%d' for param2, it has to be zero or a positive odd number\n", |
128 | smooth->param2); |
129 | return AVERROR(EINVAL); |
130 | } |
131 | |
132 | av_log(ctx, AV_LOG_VERBOSE, "type:%s param1:%d param2:%d param3:%f param4:%f\n", |
133 | type_str, smooth->param1, smooth->param2, smooth->param3, smooth->param4); |
134 | return 0; |
135 | } |
136 | |
137 | static void smooth_end_frame_filter(AVFilterContext *ctx, IplImage *inimg, IplImage *outimg) |
138 | { |
139 | OCVContext *s = ctx->priv; |
140 | SmoothContext *smooth = s->priv; |
141 | cvSmooth(inimg, outimg, smooth->type, smooth->param1, smooth->param2, smooth->param3, smooth->param4); |
142 | } |
143 | |
144 | static int read_shape_from_file(int *cols, int *rows, int **values, const char *filename, |
145 | void *log_ctx) |
146 | { |
147 | uint8_t *buf, *p, *pend; |
148 | size_t size; |
149 | int ret, i, j, w; |
150 | |
151 | if ((ret = av_file_map(filename, &buf, &size, 0, log_ctx)) < 0) |
152 | return ret; |
153 | |
154 | /* prescan file to get the number of lines and the maximum width */ |
155 | w = 0; |
156 | for (i = 0; i < size; i++) { |
157 | if (buf[i] == '\n') { |
158 | if (*rows == INT_MAX) { |
159 | av_log(log_ctx, AV_LOG_ERROR, "Overflow on the number of rows in the file\n"); |
160 | ret = AVERROR_INVALIDDATA; |
161 | goto end; |
162 | } |
163 | ++(*rows); |
164 | *cols = FFMAX(*cols, w); |
165 | w = 0; |
166 | } else if (w == INT_MAX) { |
167 | av_log(log_ctx, AV_LOG_ERROR, "Overflow on the number of columns in the file\n"); |
168 | return AVERROR_INVALIDDATA; |
169 | } |
170 | w++; |
171 | } |
172 | if (*rows > (SIZE_MAX / sizeof(int) / *cols)) { |
173 | av_log(log_ctx, AV_LOG_ERROR, "File with size %dx%d is too big\n", |
174 | *rows, *cols); |
175 | ret = AVERROR_INVALIDDATA; |
176 | goto end; |
177 | } |
178 | if (!(*values = av_mallocz_array(sizeof(int) * *rows, *cols))) { |
179 | ret = AVERROR(ENOMEM); |
180 | goto end; |
181 | } |
182 | |
183 | /* fill *values */ |
184 | p = buf; |
185 | pend = buf + size-1; |
186 | for (i = 0; i < *rows; i++) { |
187 | for (j = 0;; j++) { |
188 | if (p > pend || *p == '\n') { |
189 | p++; |
190 | break; |
191 | } else |
192 | (*values)[*cols*i + j] = !!av_isgraph(*(p++)); |
193 | } |
194 | } |
195 | |
196 | end: |
197 | av_file_unmap(buf, size); |
198 | |
199 | #ifdef DEBUG |
200 | { |
201 | char *line; |
202 | if (!(line = av_malloc(*cols + 1))) |
203 | return AVERROR(ENOMEM); |
204 | for (i = 0; i < *rows; i++) { |
205 | for (j = 0; j < *cols; j++) |
206 | line[j] = (*values)[i * *cols + j] ? '@' : ' '; |
207 | line[j] = 0; |
208 | av_log(log_ctx, AV_LOG_DEBUG, "%3d: %s\n", i, line); |
209 | } |
210 | av_free(line); |
211 | } |
212 | #endif |
213 | |
214 | return 0; |
215 | } |
216 | |
217 | static int parse_iplconvkernel(IplConvKernel **kernel, char *buf, void *log_ctx) |
218 | { |
219 | char shape_filename[128] = "", shape_str[32] = "rect"; |
220 | int cols = 0, rows = 0, anchor_x = 0, anchor_y = 0, shape = CV_SHAPE_RECT; |
221 | int *values = NULL, ret = 0; |
222 | |
223 | sscanf(buf, "%dx%d+%dx%d/%32[^=]=%127s", &cols, &rows, &anchor_x, &anchor_y, shape_str, shape_filename); |
224 | |
225 | if (!strcmp(shape_str, "rect" )) shape = CV_SHAPE_RECT; |
226 | else if (!strcmp(shape_str, "cross" )) shape = CV_SHAPE_CROSS; |
227 | else if (!strcmp(shape_str, "ellipse")) shape = CV_SHAPE_ELLIPSE; |
228 | else if (!strcmp(shape_str, "custom" )) { |
229 | shape = CV_SHAPE_CUSTOM; |
230 | if ((ret = read_shape_from_file(&cols, &rows, &values, shape_filename, log_ctx)) < 0) |
231 | return ret; |
232 | } else { |
233 | av_log(log_ctx, AV_LOG_ERROR, |
234 | "Shape unspecified or type '%s' unknown.\n", shape_str); |
235 | ret = AVERROR(EINVAL); |
236 | goto out; |
237 | } |
238 | |
239 | if (rows <= 0 || cols <= 0) { |
240 | av_log(log_ctx, AV_LOG_ERROR, |
241 | "Invalid non-positive values for shape size %dx%d\n", cols, rows); |
242 | ret = AVERROR(EINVAL); |
243 | goto out; |
244 | } |
245 | |
246 | if (anchor_x < 0 || anchor_y < 0 || anchor_x >= cols || anchor_y >= rows) { |
247 | av_log(log_ctx, AV_LOG_ERROR, |
248 | "Shape anchor %dx%d is not inside the rectangle with size %dx%d.\n", |
249 | anchor_x, anchor_y, cols, rows); |
250 | ret = AVERROR(EINVAL); |
251 | goto out; |
252 | } |
253 | |
254 | *kernel = cvCreateStructuringElementEx(cols, rows, anchor_x, anchor_y, shape, values); |
255 | if (!*kernel) { |
256 | ret = AVERROR(ENOMEM); |
257 | goto out; |
258 | } |
259 | |
260 | av_log(log_ctx, AV_LOG_VERBOSE, "Structuring element: w:%d h:%d x:%d y:%d shape:%s\n", |
261 | rows, cols, anchor_x, anchor_y, shape_str); |
262 | out: |
263 | av_freep(&values); |
264 | return ret; |
265 | } |
266 | |
267 | typedef struct DilateContext { |
268 | int nb_iterations; |
269 | IplConvKernel *kernel; |
270 | } DilateContext; |
271 | |
272 | static av_cold int dilate_init(AVFilterContext *ctx, const char *args) |
273 | { |
274 | OCVContext *s = ctx->priv; |
275 | DilateContext *dilate = s->priv; |
276 | char default_kernel_str[] = "3x3+0x0/rect"; |
277 | char *kernel_str = NULL; |
278 | const char *buf = args; |
279 | int ret; |
280 | |
281 | if (args) { |
282 | kernel_str = av_get_token(&buf, "|"); |
283 | |
284 | if (!kernel_str) |
285 | return AVERROR(ENOMEM); |
286 | } |
287 | |
288 | ret = parse_iplconvkernel(&dilate->kernel, |
289 | (!kernel_str || !*kernel_str) ? default_kernel_str |
290 | : kernel_str, |
291 | ctx); |
292 | av_free(kernel_str); |
293 | if (ret < 0) |
294 | return ret; |
295 | |
296 | if (!buf || sscanf(buf, "|%d", &dilate->nb_iterations) != 1) |
297 | dilate->nb_iterations = 1; |
298 | av_log(ctx, AV_LOG_VERBOSE, "iterations_nb:%d\n", dilate->nb_iterations); |
299 | if (dilate->nb_iterations <= 0) { |
300 | av_log(ctx, AV_LOG_ERROR, "Invalid non-positive value '%d' for nb_iterations\n", |
301 | dilate->nb_iterations); |
302 | return AVERROR(EINVAL); |
303 | } |
304 | return 0; |
305 | } |
306 | |
307 | static av_cold void dilate_uninit(AVFilterContext *ctx) |
308 | { |
309 | OCVContext *s = ctx->priv; |
310 | DilateContext *dilate = s->priv; |
311 | |
312 | cvReleaseStructuringElement(&dilate->kernel); |
313 | } |
314 | |
315 | static void dilate_end_frame_filter(AVFilterContext *ctx, IplImage *inimg, IplImage *outimg) |
316 | { |
317 | OCVContext *s = ctx->priv; |
318 | DilateContext *dilate = s->priv; |
319 | cvDilate(inimg, outimg, dilate->kernel, dilate->nb_iterations); |
320 | } |
321 | |
322 | static void erode_end_frame_filter(AVFilterContext *ctx, IplImage *inimg, IplImage *outimg) |
323 | { |
324 | OCVContext *s = ctx->priv; |
325 | DilateContext *dilate = s->priv; |
326 | cvErode(inimg, outimg, dilate->kernel, dilate->nb_iterations); |
327 | } |
328 | |
329 | typedef struct OCVFilterEntry { |
330 | const char *name; |
331 | size_t priv_size; |
332 | int (*init)(AVFilterContext *ctx, const char *args); |
333 | void (*uninit)(AVFilterContext *ctx); |
334 | void (*end_frame_filter)(AVFilterContext *ctx, IplImage *inimg, IplImage *outimg); |
335 | } OCVFilterEntry; |
336 | |
337 | static const OCVFilterEntry ocv_filter_entries[] = { |
338 | { "dilate", sizeof(DilateContext), dilate_init, dilate_uninit, dilate_end_frame_filter }, |
339 | { "erode", sizeof(DilateContext), dilate_init, dilate_uninit, erode_end_frame_filter }, |
340 | { "smooth", sizeof(SmoothContext), smooth_init, NULL, smooth_end_frame_filter }, |
341 | }; |
342 | |
343 | static av_cold int init(AVFilterContext *ctx) |
344 | { |
345 | OCVContext *s = ctx->priv; |
346 | int i; |
347 | |
348 | if (!s->name) { |
349 | av_log(ctx, AV_LOG_ERROR, "No libopencv filter name specified\n"); |
350 | return AVERROR(EINVAL); |
351 | } |
352 | for (i = 0; i < FF_ARRAY_ELEMS(ocv_filter_entries); i++) { |
353 | const OCVFilterEntry *entry = &ocv_filter_entries[i]; |
354 | if (!strcmp(s->name, entry->name)) { |
355 | s->init = entry->init; |
356 | s->uninit = entry->uninit; |
357 | s->end_frame_filter = entry->end_frame_filter; |
358 | |
359 | if (!(s->priv = av_mallocz(entry->priv_size))) |
360 | return AVERROR(ENOMEM); |
361 | return s->init(ctx, s->params); |
362 | } |
363 | } |
364 | |
365 | av_log(ctx, AV_LOG_ERROR, "No libopencv filter named '%s'\n", s->name); |
366 | return AVERROR(EINVAL); |
367 | } |
368 | |
369 | static av_cold void uninit(AVFilterContext *ctx) |
370 | { |
371 | OCVContext *s = ctx->priv; |
372 | |
373 | if (s->uninit) |
374 | s->uninit(ctx); |
375 | av_freep(&s->priv); |
376 | } |
377 | |
378 | static int filter_frame(AVFilterLink *inlink, AVFrame *in) |
379 | { |
380 | AVFilterContext *ctx = inlink->dst; |
381 | OCVContext *s = ctx->priv; |
382 | AVFilterLink *outlink= inlink->dst->outputs[0]; |
383 | AVFrame *out; |
384 | IplImage inimg, outimg; |
385 | |
386 | out = ff_get_video_buffer(outlink, outlink->w, outlink->h); |
387 | if (!out) { |
388 | av_frame_free(&in); |
389 | return AVERROR(ENOMEM); |
390 | } |
391 | av_frame_copy_props(out, in); |
392 | |
393 | fill_iplimage_from_frame(&inimg , in , inlink->format); |
394 | fill_iplimage_from_frame(&outimg, out, inlink->format); |
395 | s->end_frame_filter(ctx, &inimg, &outimg); |
396 | fill_frame_from_iplimage(out, &outimg, inlink->format); |
397 | |
398 | av_frame_free(&in); |
399 | |
400 | return ff_filter_frame(outlink, out); |
401 | } |
402 | |
403 | #define OFFSET(x) offsetof(OCVContext, x) |
404 | #define FLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_FILTERING_PARAM |
405 | static const AVOption ocv_options[] = { |
406 | { "filter_name", NULL, OFFSET(name), AV_OPT_TYPE_STRING, .flags = FLAGS }, |
407 | { "filter_params", NULL, OFFSET(params), AV_OPT_TYPE_STRING, .flags = FLAGS }, |
408 | { NULL } |
409 | }; |
410 | |
411 | AVFILTER_DEFINE_CLASS(ocv); |
412 | |
413 | static const AVFilterPad avfilter_vf_ocv_inputs[] = { |
414 | { |
415 | .name = "default", |
416 | .type = AVMEDIA_TYPE_VIDEO, |
417 | .filter_frame = filter_frame, |
418 | }, |
419 | { NULL } |
420 | }; |
421 | |
422 | static const AVFilterPad avfilter_vf_ocv_outputs[] = { |
423 | { |
424 | .name = "default", |
425 | .type = AVMEDIA_TYPE_VIDEO, |
426 | }, |
427 | { NULL } |
428 | }; |
429 | |
430 | AVFilter ff_vf_ocv = { |
431 | .name = "ocv", |
432 | .description = NULL_IF_CONFIG_SMALL("Apply transform using libopencv."), |
433 | .priv_size = sizeof(OCVContext), |
434 | .priv_class = &ocv_class, |
435 | .query_formats = query_formats, |
436 | .init = init, |
437 | .uninit = uninit, |
438 | .inputs = avfilter_vf_ocv_inputs, |
439 | .outputs = avfilter_vf_ocv_outputs, |
440 | }; |
441 |