Statistics
| Branch: | Revision:

ffmpeg / libavfilter / vf_libopencv.c @ 2912e87a

History | View | Annotate | Download (13 KB)

1
/*
2
 * Copyright (c) 2010 Stefano Sabatini
3
 *
4
 * This file is part of Libav.
5
 *
6
 * Libav 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
 * Libav 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 Libav; 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
/* #define DEBUG */
27

    
28
#include <opencv/cv.h>
29
#include <opencv/cxtypes.h>
30
#include "libavutil/avstring.h"
31
#include "libavutil/file.h"
32
#include "avfilter.h"
33

    
34
static void fill_iplimage_from_picref(IplImage *img, const AVFilterBufferRef *picref, enum PixelFormat pixfmt)
35
{
36
    IplImage *tmpimg;
37
    int depth, channels_nb;
38

    
39
    if      (pixfmt == PIX_FMT_GRAY8) { depth = IPL_DEPTH_8U;  channels_nb = 1; }
40
    else if (pixfmt == PIX_FMT_BGRA)  { depth = IPL_DEPTH_8U;  channels_nb = 4; }
41
    else if (pixfmt == PIX_FMT_BGR24) { depth = IPL_DEPTH_8U;  channels_nb = 3; }
42
    else return;
43

    
44
    tmpimg = cvCreateImageHeader((CvSize){picref->video->w, picref->video->h}, depth, channels_nb);
45
    *img = *tmpimg;
46
    img->imageData = img->imageDataOrigin = picref->data[0];
47
    img->dataOrder = IPL_DATA_ORDER_PIXEL;
48
    img->origin    = IPL_ORIGIN_TL;
49
    img->widthStep = picref->linesize[0];
50
}
51

    
52
static void fill_picref_from_iplimage(AVFilterBufferRef *picref, const IplImage *img, enum PixelFormat pixfmt)
53
{
54
    picref->linesize[0] = img->widthStep;
55
    picref->data[0]     = img->imageData;
56
}
57

    
58
static int query_formats(AVFilterContext *ctx)
59
{
60
    static const enum PixelFormat pix_fmts[] = {
61
        PIX_FMT_BGR24, PIX_FMT_BGRA, PIX_FMT_GRAY8, PIX_FMT_NONE
62
    };
63

    
64
    avfilter_set_common_formats(ctx, avfilter_make_format_list(pix_fmts));
65
    return 0;
66
}
67

    
68
static void null_draw_slice(AVFilterLink *link, int y, int h, int slice_dir) { }
69

    
70
typedef struct {
71
    const char *name;
72
    int (*init)(AVFilterContext *ctx, const char *args, void *opaque);
73
    void (*uninit)(AVFilterContext *ctx);
74
    void (*end_frame_filter)(AVFilterContext *ctx, IplImage *inimg, IplImage *outimg);
75
    void *priv;
76
} OCVContext;
77

    
78
typedef struct {
79
    int type;
80
    int    param1, param2;
81
    double param3, param4;
82
} SmoothContext;
83

    
84
static av_cold int smooth_init(AVFilterContext *ctx, const char *args, void *opaque)
85
{
86
    OCVContext *ocv = ctx->priv;
87
    SmoothContext *smooth = ocv->priv;
88
    char type_str[128] = "gaussian";
89

    
90
    smooth->param1 = 3;
91
    smooth->param2 = 0;
92
    smooth->param3 = 0.0;
93
    smooth->param4 = 0.0;
94

    
95
    if (args)
96
        sscanf(args, "%127[^:]:%d:%d:%lf:%lf", type_str, &smooth->param1, &smooth->param2, &smooth->param3, &smooth->param4);
97

    
98
    if      (!strcmp(type_str, "blur"         )) smooth->type = CV_BLUR;
99
    else if (!strcmp(type_str, "blur_no_scale")) smooth->type = CV_BLUR_NO_SCALE;
100
    else if (!strcmp(type_str, "median"       )) smooth->type = CV_MEDIAN;
101
    else if (!strcmp(type_str, "gaussian"     )) smooth->type = CV_GAUSSIAN;
102
    else if (!strcmp(type_str, "bilateral"    )) smooth->type = CV_BILATERAL;
103
    else {
104
        av_log(ctx, AV_LOG_ERROR, "Smoothing type '%s' unknown\n.", type_str);
105
        return AVERROR(EINVAL);
106
    }
107

    
108
    if (smooth->param1 < 0 || !(smooth->param1%2)) {
109
        av_log(ctx, AV_LOG_ERROR,
110
               "Invalid value '%d' for param1, it has to be a positive odd number\n",
111
               smooth->param1);
112
        return AVERROR(EINVAL);
113
    }
114
    if ((smooth->type == CV_BLUR || smooth->type == CV_BLUR_NO_SCALE || smooth->type == CV_GAUSSIAN) &&
115
        (smooth->param2 < 0 || (smooth->param2 && !(smooth->param2%2)))) {
116
        av_log(ctx, AV_LOG_ERROR,
117
               "Invalid value '%d' for param2, it has to be zero or a positive odd number\n",
118
               smooth->param2);
119
        return AVERROR(EINVAL);
120
    }
121

    
122
    av_log(ctx, AV_LOG_INFO, "type:%s param1:%d param2:%d param3:%f param4:%f\n",
123
           type_str, smooth->param1, smooth->param2, smooth->param3, smooth->param4);
124
    return 0;
125
}
126

    
127
static void smooth_end_frame_filter(AVFilterContext *ctx, IplImage *inimg, IplImage *outimg)
128
{
129
    OCVContext *ocv = ctx->priv;
130
    SmoothContext *smooth = ocv->priv;
131
    cvSmooth(inimg, outimg, smooth->type, smooth->param1, smooth->param2, smooth->param3, smooth->param4);
132
}
133

    
134
static int read_shape_from_file(int *cols, int *rows, int **values, const char *filename,
135
                                void *log_ctx)
136
{
137
    uint8_t *buf, *p, *pend;
138
    size_t size;
139
    int ret, i, j, w;
140

    
141
    if ((ret = av_file_map(filename, &buf, &size, 0, log_ctx)) < 0)
142
        return ret;
143

    
144
    /* prescan file to get the number of lines and the maximum width */
145
    w = 0;
146
    for (i = 0; i < size; i++) {
147
        if (buf[i] == '\n') {
148
            if (*rows == INT_MAX) {
149
                av_log(log_ctx, AV_LOG_ERROR, "Overflow on the number of rows in the file\n");
150
                return AVERROR_INVALIDDATA;
151
            }
152
            ++(*rows);
153
            *cols = FFMAX(*cols, w);
154
            w = 0;
155
        } else if (w == INT_MAX) {
156
            av_log(log_ctx, AV_LOG_ERROR, "Overflow on the number of columns in the file\n");
157
            return AVERROR_INVALIDDATA;
158
        }
159
        w++;
160
    }
161
    if (*rows > (FF_INTERNAL_MEM_TYPE_MAX_VALUE / (sizeof(int)) / *cols)) {
162
        av_log(log_ctx, AV_LOG_ERROR, "File with size %dx%d is too big\n",
163
               *rows, *cols);
164
        return AVERROR_INVALIDDATA;
165
    }
166
    if (!(*values = av_mallocz(sizeof(int) * *rows * *cols)))
167
        return AVERROR(ENOMEM);
168

    
169
    /* fill *values */
170
    p    = buf;
171
    pend = buf + size-1;
172
    for (i = 0; i < *rows; i++) {
173
        for (j = 0;; j++) {
174
            if (p > pend || *p == '\n') {
175
                p++;
176
                break;
177
            } else
178
                (*values)[*cols*i + j] = !!isgraph(*(p++));
179
        }
180
    }
181
    av_file_unmap(buf, size);
182

    
183
#ifdef DEBUG
184
    {
185
        char *line;
186
        if (!(line = av_malloc(*cols + 1)))
187
            return AVERROR(ENOMEM);
188
        for (i = 0; i < *rows; i++) {
189
            for (j = 0; j < *cols; j++)
190
                line[j] = (*values)[i * *cols + j] ? '@' : ' ';
191
            line[j] = 0;
192
            av_log(log_ctx, AV_LOG_DEBUG, "%3d: %s\n", i, line);
193
        }
194
        av_free(line);
195
    }
196
#endif
197

    
198
    return 0;
199
}
200

    
201
static int parse_iplconvkernel(IplConvKernel **kernel, char *buf, void *log_ctx)
202
{
203
    char shape_filename[128] = "", shape_str[32] = "rect";
204
    int cols = 0, rows = 0, anchor_x = 0, anchor_y = 0, shape = CV_SHAPE_RECT;
205
    int *values = NULL, ret;
206

    
207
    sscanf(buf, "%dx%d+%dx%d/%32[^=]=%127s", &cols, &rows, &anchor_x, &anchor_y, shape_str, shape_filename);
208

    
209
    if      (!strcmp(shape_str, "rect"   )) shape = CV_SHAPE_RECT;
210
    else if (!strcmp(shape_str, "cross"  )) shape = CV_SHAPE_CROSS;
211
    else if (!strcmp(shape_str, "ellipse")) shape = CV_SHAPE_ELLIPSE;
212
    else if (!strcmp(shape_str, "custom" )) {
213
        shape = CV_SHAPE_CUSTOM;
214
        if ((ret = read_shape_from_file(&cols, &rows, &values, shape_filename, log_ctx)) < 0)
215
            return ret;
216
    } else {
217
        av_log(log_ctx, AV_LOG_ERROR,
218
               "Shape unspecified or type '%s' unknown\n.", shape_str);
219
        return AVERROR(EINVAL);
220
    }
221

    
222
    if (rows <= 0 || cols <= 0) {
223
        av_log(log_ctx, AV_LOG_ERROR,
224
               "Invalid non-positive values for shape size %dx%d\n", cols, rows);
225
        return AVERROR(EINVAL);
226
    }
227

    
228
    if (anchor_x < 0 || anchor_y < 0 || anchor_x >= cols || anchor_y >= rows) {
229
        av_log(log_ctx, AV_LOG_ERROR,
230
               "Shape anchor %dx%d is not inside the rectangle with size %dx%d.\n",
231
               anchor_x, anchor_y, cols, rows);
232
        return AVERROR(EINVAL);
233
    }
234

    
235
    *kernel = cvCreateStructuringElementEx(cols, rows, anchor_x, anchor_y, shape, values);
236
    av_freep(&values);
237
    if (!*kernel)
238
        return AVERROR(ENOMEM);
239

    
240
    av_log(log_ctx, AV_LOG_INFO, "Structuring element: w:%d h:%d x:%d y:%d shape:%s\n",
241
           rows, cols, anchor_x, anchor_y, shape_str);
242
    return 0;
243
}
244

    
245
typedef struct {
246
    int nb_iterations;
247
    IplConvKernel *kernel;
248
} DilateContext;
249

    
250
static av_cold int dilate_init(AVFilterContext *ctx, const char *args, void *opaque)
251
{
252
    OCVContext *ocv = ctx->priv;
253
    DilateContext *dilate = ocv->priv;
254
    char default_kernel_str[] = "3x3+0x0/rect";
255
    char *kernel_str;
256
    const char *buf = args;
257
    int ret;
258

    
259
    dilate->nb_iterations = 1;
260

    
261
    if (args)
262
        kernel_str = av_get_token(&buf, ":");
263
    if ((ret = parse_iplconvkernel(&dilate->kernel,
264
                                   *kernel_str ? kernel_str : default_kernel_str,
265
                                   ctx)) < 0)
266
        return ret;
267
    av_free(kernel_str);
268

    
269
    sscanf(buf, ":%d", &dilate->nb_iterations);
270
    av_log(ctx, AV_LOG_INFO, "iterations_nb:%d\n", dilate->nb_iterations);
271
    if (dilate->nb_iterations <= 0) {
272
        av_log(ctx, AV_LOG_ERROR, "Invalid non-positive value '%d' for nb_iterations\n",
273
               dilate->nb_iterations);
274
        return AVERROR(EINVAL);
275
    }
276
    return 0;
277
}
278

    
279
static av_cold void dilate_uninit(AVFilterContext *ctx)
280
{
281
    OCVContext *ocv = ctx->priv;
282
    DilateContext *dilate = ocv->priv;
283

    
284
    cvReleaseStructuringElement(&dilate->kernel);
285
}
286

    
287
static void dilate_end_frame_filter(AVFilterContext *ctx, IplImage *inimg, IplImage *outimg)
288
{
289
    OCVContext *ocv = ctx->priv;
290
    DilateContext *dilate = ocv->priv;
291
    cvDilate(inimg, outimg, dilate->kernel, dilate->nb_iterations);
292
}
293

    
294
static void erode_end_frame_filter(AVFilterContext *ctx, IplImage *inimg, IplImage *outimg)
295
{
296
    OCVContext *ocv = ctx->priv;
297
    DilateContext *dilate = ocv->priv;
298
    cvErode(inimg, outimg, dilate->kernel, dilate->nb_iterations);
299
}
300

    
301
typedef struct {
302
    const char *name;
303
    size_t priv_size;
304
    int  (*init)(AVFilterContext *ctx, const char *args, void *opaque);
305
    void (*uninit)(AVFilterContext *ctx);
306
    void (*end_frame_filter)(AVFilterContext *ctx, IplImage *inimg, IplImage *outimg);
307
} OCVFilterEntry;
308

    
309
static OCVFilterEntry ocv_filter_entries[] = {
310
    { "dilate", sizeof(DilateContext), dilate_init, dilate_uninit, dilate_end_frame_filter },
311
    { "erode",  sizeof(DilateContext), dilate_init, dilate_uninit, erode_end_frame_filter  },
312
    { "smooth", sizeof(SmoothContext), smooth_init, NULL, smooth_end_frame_filter },
313
};
314

    
315
static av_cold int init(AVFilterContext *ctx, const char *args, void *opaque)
316
{
317
    OCVContext *ocv = ctx->priv;
318
    char name[128], priv_args[1024];
319
    int i;
320
    char c;
321

    
322
    sscanf(args, "%127[^=:]%c%1023s", name, &c, priv_args);
323

    
324
    for (i = 0; i < FF_ARRAY_ELEMS(ocv_filter_entries); i++) {
325
        OCVFilterEntry *entry = &ocv_filter_entries[i];
326
        if (!strcmp(name, entry->name)) {
327
            ocv->name             = entry->name;
328
            ocv->init             = entry->init;
329
            ocv->uninit           = entry->uninit;
330
            ocv->end_frame_filter = entry->end_frame_filter;
331

    
332
            if (!(ocv->priv = av_mallocz(entry->priv_size)))
333
                return AVERROR(ENOMEM);
334
            return ocv->init(ctx, priv_args, opaque);
335
        }
336
    }
337

    
338
    av_log(ctx, AV_LOG_ERROR, "No libopencv filter named '%s'\n", name);
339
    return AVERROR(EINVAL);
340
}
341

    
342
static av_cold void uninit(AVFilterContext *ctx)
343
{
344
    OCVContext *ocv = ctx->priv;
345

    
346
    if (ocv->uninit)
347
        ocv->uninit(ctx);
348
    av_free(ocv->priv);
349
    memset(ocv, 0, sizeof(*ocv));
350
}
351

    
352
static void end_frame(AVFilterLink *inlink)
353
{
354
    AVFilterContext *ctx = inlink->dst;
355
    OCVContext *ocv = ctx->priv;
356
    AVFilterLink *outlink= inlink->dst->outputs[0];
357
    AVFilterBufferRef *inpicref  = inlink ->cur_buf;
358
    AVFilterBufferRef *outpicref = outlink->out_buf;
359
    IplImage inimg, outimg;
360

    
361
    fill_iplimage_from_picref(&inimg , inpicref , inlink->format);
362
    fill_iplimage_from_picref(&outimg, outpicref, inlink->format);
363
    ocv->end_frame_filter(ctx, &inimg, &outimg);
364
    fill_picref_from_iplimage(outpicref, &outimg, inlink->format);
365

    
366
    avfilter_unref_buffer(inpicref);
367
    avfilter_draw_slice(outlink, 0, outlink->h, 1);
368
    avfilter_end_frame(outlink);
369
    avfilter_unref_buffer(outpicref);
370
}
371

    
372
AVFilter avfilter_vf_ocv = {
373
    .name        = "ocv",
374
    .description = NULL_IF_CONFIG_SMALL("Apply transform using libopencv."),
375

    
376
    .priv_size = sizeof(OCVContext),
377

    
378
    .query_formats = query_formats,
379
    .init = init,
380
    .uninit = uninit,
381

    
382
    .inputs    = (AVFilterPad[]) {{ .name             = "default",
383
                                    .type             = AVMEDIA_TYPE_VIDEO,
384
                                    .draw_slice       = null_draw_slice,
385
                                    .end_frame        = end_frame,
386
                                    .min_perms        = AV_PERM_READ },
387
                                  { .name = NULL}},
388

    
389
    .outputs   = (AVFilterPad[]) {{ .name             = "default",
390
                                    .type             = AVMEDIA_TYPE_VIDEO, },
391
                                  { .name = NULL}},
392
};