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