History

avfilter: add corr video filter

Paul B Mahol committed 3 years ago

Commit f0f19f31 parent 9ca139b6 files · +366 −1Browse files

Unified · Split

6 files
  1. Changelog +1 −0
  2. doc/filters.texi +24 −0
  3. libavfilter/Makefile +1 −0
  4. libavfilter/allfilters.c +1 −0
  5. libavfilter/version.h +1 −1
  6. libavfilter/vf_corr.c +338 −0
Changelog+1−0
@@ -26,6 +26,7 @@ version <next>: · 25 unmodified lines
2626 - QSV AV1 encoder
2727 - QSV decoding and encoding for 10/12bit 422, 10/12bit 444 HEVC and VP9
2828 - showcwt multimedia filter
29+- corr video filter
2930
3031
3132 version 5.1:
doc/filters.texi+24−0
@@ -10110,6 +10110,30 @@ ffmpeg -f lavfi -i nullsrc=s=100x100,coreimage=filter=CIQRCodeGenerator@@inputMe · 10109 unmodified lines
1011010110 @end example
1011110111 @end itemize
1011210112
10113+@section corr
10114+
10115+Obtain the correlation between two input videos.
10116+
10117+This filter takes two input videos.
10118+
10119+Both input videos must have the same resolution and pixel format for
10120+this filter to work correctly. Also it assumes that both inputs
10121+have the same number of frames, which are compared one by one.
10122+
10123+The obtained per component, average, min and max correlation is printed through
10124+the logging system.
10125+
10126+The filter stores the calculated correlation of each frame in frame metadata.
10127+
10128+This filter also supports the @ref{framesync} options.
10129+
10130+In the below example the input file @file{main.mpg} being processed is compared
10131+with the reference file @file{ref.mpg}.
10132+
10133+@example
10134+ffmpeg -i main.mpg -i ref.mpg -lavfi corr -f null -
10135+@end example
10136+
1011310137 @section cover_rect
1011410138
1011510139 Cover a rectangular object
libavfilter/Makefile+1−0
@@ -241,6 +241,7 @@ OBJS-$(CONFIG_CONVOLUTION_OPENCL_FILTER) += vf_convolution_opencl.o opencl.o · 240 unmodified lines
241241 OBJS-$(CONFIG_CONVOLVE_FILTER) += vf_convolve.o framesync.o
242242 OBJS-$(CONFIG_COPY_FILTER) += vf_copy.o
243243 OBJS-$(CONFIG_COREIMAGE_FILTER) += vf_coreimage.o
244+OBJS-$(CONFIG_CORR_FILTER) += vf_corr.o framesync.o
244245 OBJS-$(CONFIG_COVER_RECT_FILTER) += vf_cover_rect.o lavfutils.o
245246 OBJS-$(CONFIG_CROP_FILTER) += vf_crop.o
246247 OBJS-$(CONFIG_CROPDETECT_FILTER) += vf_cropdetect.o
libavfilter/allfilters.c+1−0
@@ -221,6 +221,7 @@ extern const AVFilter ff_vf_convolution_opencl; · 220 unmodified lines
221221 extern const AVFilter ff_vf_convolve;
222222 extern const AVFilter ff_vf_copy;
223223 extern const AVFilter ff_vf_coreimage;
224+extern const AVFilter ff_vf_corr;
224225 extern const AVFilter ff_vf_cover_rect;
225226 extern const AVFilter ff_vf_crop;
226227 extern const AVFilter ff_vf_cropdetect;
libavfilter/version.h+1−1
@@ -31,7 +31,7 @@ · 30 unmodified lines
3131
3232 #include "version_major.h"
3333
34−#define LIBAVFILTER_VERSION_MINOR 51
34+#define LIBAVFILTER_VERSION_MINOR 52
3535 #define LIBAVFILTER_VERSION_MICRO 100
3636
3737
libavfilter/vf_corr.c+338−0
@@ -0,0 +1,338 @@
1+/*
2+ * This file is part of FFmpeg.
3+ *
4+ * FFmpeg is free software; you can redistribute it and/or
5+ * modify it under the terms of the GNU Lesser General Public
6+ * License as published by the Free Software Foundation; either
7+ * version 2.1 of the License, or (at your option) any later version.
8+ *
9+ * FFmpeg is distributed in the hope that it will be useful,
10+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
11+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12+ * Lesser General Public License for more details.
13+ *
14+ * You should have received a copy of the GNU Lesser General Public
15+ * License along with FFmpeg; if not, write to the Free Software
16+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17+ */
18+
19+/**
20+ * @file
21+ * Calculate the correlation between two input videos.
22+ */
23+
24+#include "libavutil/avstring.h"
25+#include "libavutil/opt.h"
26+#include "libavutil/pixdesc.h"
27+#include "avfilter.h"
28+#include "drawutils.h"
29+#include "formats.h"
30+#include "framesync.h"
31+#include "internal.h"
32+#include "video.h"
33+
34+typedef struct CorrContext {
35+ const AVClass *class;
36+ FFFrameSync fs;
37+ double score, min_score, max_score, score_comp[4];
38+ uint64_t nb_frames;
39+ int is_rgb;
40+ uint8_t rgba_map[4];
41+ int max[4];
42+ char comps[4];
43+ int nb_components;
44+ int planewidth[4];
45+ int planeheight[4];
46+ int (*filter_slice)(AVFilterContext *ctx, void *arg,
47+ int jobnr, int nb_jobs);
48+} CorrContext;
49+
50+#define OFFSET(x) offsetof(CorrContext, x)
51+#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
52+
53+static void set_meta(AVFilterContext *ctx,
54+ AVDictionary **metadata, const char *key, char comp, float d)
55+{
56+ char value[128];
57+ snprintf(value, sizeof(value), "%f", d);
58+ if (comp) {
59+ char key2[128];
60+ snprintf(key2, sizeof(key2), "lavfi.%s.%s%s%c",
61+ ctx->filter->name, ctx->filter->name, key, comp);
62+ av_dict_set(metadata, key2, value, 0);
63+ } else {
64+ char key2[128];
65+ snprintf(key2, sizeof(key2), "lavfi.%s.%s%s",
66+ ctx->filter->name, ctx->filter->name, key);
67+ av_dict_set(metadata, key2, value, 0);
68+ }
69+}
70+
71+#define CORR(type, name) \
72+static void f##name(AVFilterContext *ctx, AVFrame *master, \
73+ AVFrame *ref, double *comp_score) \
74+{ \
75+ CorrContext *s = ctx->priv; \
76+ \
77+ for (int c = 0; c < s->nb_components; c++) { \
78+ const ptrdiff_t linesize1 = master->linesize[c] / \
79+ sizeof(type); \
80+ const ptrdiff_t linesize2 = ref->linesize[c] / \
81+ sizeof(type); \
82+ const type *src1 = (const type *)master->data[c]; \
83+ const type *src2 = (const type *)ref->data[c]; \
84+ const int h = s->planeheight[c]; \
85+ const int w = s->planewidth[c]; \
86+ const float scale = 1.f / s->max[c]; \
87+ uint64_t sum1 = 0, sum2 = 0; \
88+ float sum12, sum1q, sum2q; \
89+ float sumq, mean1, mean2; \
90+ \
91+ for (int y = 0; y < h; y++) { \
92+ for (int x = 0; x < w; x++) { \
93+ sum1 += src1[x]; \
94+ sum2 += src2[x]; \
95+ } \
96+ \
97+ src1 += linesize1; \
98+ src2 += linesize2; \
99+ } \
100+ \
101+ mean1 = scale * (sum1 /(double)(w * h)); \
102+ mean2 = scale * (sum2 /(double)(w * h)); \
103+ \
104+ src1 = (const type *)master->data[c]; \
105+ src2 = (const type *)ref->data[c]; \
106+ \
107+ sum12 = 0.f; \
108+ sum1q = 0.f; \
109+ sum2q = 0.f; \
110+ \
111+ for (int y = 0; y < h; y++) { \
112+ for (int x = 0; x < w; x++) { \
113+ const float f1 = scale * src1[x] - mean1; \
114+ const float f2 = scale * src2[x] - mean2; \
115+ \
116+ sum12 += f1 * f2; \
117+ sum1q += f1 * f1; \
118+ sum2q += f2 * f2; \
119+ } \
120+ \
121+ src1 += linesize1; \
122+ src2 += linesize2; \
123+ } \
124+ \
125+ sumq = sqrtf(sum1q * sum2q); \
126+ if (sumq > 0.f) { \
127+ comp_score[c] = av_clipf(sum12 / sumq,-1.f,1.f); \
128+ } else { \
129+ comp_score[c] = sum1q == sum2q ? 1.f : 0.f; \
130+ } \
131+ } \
132+}
133+
134+CORR(uint8_t, corr8)
135+CORR(uint16_t, corr16)
136+
137+static int do_corr(FFFrameSync *fs)
138+{
139+ AVFilterContext *ctx = fs->parent;
140+ CorrContext *s = ctx->priv;
141+ AVFrame *master, *ref;
142+ double comp_score[4], score = 0.;
143+ AVDictionary **metadata;
144+ int ret;
145+
146+ ret = ff_framesync_dualinput_get(fs, &master, &ref);
147+ if (ret < 0)
148+ return ret;
149+ if (ctx->is_disabled || !ref)
150+ return ff_filter_frame(ctx->outputs[0], master);
151+ metadata = &master->metadata;
152+
153+ if (s->max[0] > 255) {
154+ fcorr16(ctx, master, ref, comp_score);
155+ } else {
156+ fcorr8(ctx, master, ref, comp_score);
157+ }
158+
159+ for (int c = 0; c < s->nb_components; c++)
160+ score += comp_score[c];
161+ score /= s->nb_components;
162+ s->score += score;
163+
164+ s->min_score = fmin(s->min_score, score);
165+ s->max_score = fmax(s->max_score, score);
166+
167+ for (int c = 0; c < s->nb_components; c++)
168+ s->score_comp[c] += comp_score[c];
169+ s->nb_frames++;
170+
171+ for (int j = 0; j < s->nb_components; j++) {
172+ int c = s->is_rgb ? s->rgba_map[j] : j;
173+ set_meta(ctx, metadata, ".", s->comps[j], comp_score[c]);
174+ }
175+ set_meta(ctx, metadata, "_avg", 0, score);
176+
177+ return ff_filter_frame(ctx->outputs[0], master);
178+}
179+
180+static av_cold int init(AVFilterContext *ctx)
181+{
182+ CorrContext *s = ctx->priv;
183+
184+ s->fs.on_event = do_corr;
185+
186+ return 0;
187+}
188+
189+static const enum AVPixelFormat pix_fmts[] = {
190+ AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16,
191+#define PF_NOALPHA(suf) AV_PIX_FMT_YUV420##suf, AV_PIX_FMT_YUV422##suf, AV_PIX_FMT_YUV444##suf
192+#define PF_ALPHA(suf) AV_PIX_FMT_YUVA420##suf, AV_PIX_FMT_YUVA422##suf, AV_PIX_FMT_YUVA444##suf
193+#define PF(suf) PF_NOALPHA(suf), PF_ALPHA(suf)
194+ PF(P), PF(P9), PF(P10), PF_NOALPHA(P12), PF_NOALPHA(P14), PF(P16),
195+ AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
196+ AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
197+ AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ444P,
198+ AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
199+ AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
200+ AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
201+ AV_PIX_FMT_NONE
202+};
203+
204+static int config_input_ref(AVFilterLink *inlink)
205+{
206+ const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
207+ AVFilterContext *ctx = inlink->dst;
208+ CorrContext *s = ctx->priv;
209+
210+ s->nb_components = desc->nb_components;
211+ if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
212+ ctx->inputs[0]->h != ctx->inputs[1]->h) {
213+ av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be same.\n");
214+ return AVERROR(EINVAL);
215+ }
216+
217+ s->is_rgb = ff_fill_rgba_map(s->rgba_map, inlink->format) >= 0;
218+ s->comps[0] = s->is_rgb ? 'R' : 'Y' ;
219+ s->comps[1] = s->is_rgb ? 'G' : 'U' ;
220+ s->comps[2] = s->is_rgb ? 'B' : 'V' ;
221+ s->comps[3] = 'A';
222+
223+ s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
224+ s->planeheight[0] = s->planeheight[3] = inlink->h;
225+ s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
226+ s->planewidth[0] = s->planewidth[3] = inlink->w;
227+
228+ s->min_score = +INFINITY;
229+ s->max_score = -INFINITY;
230+
231+ s->max[0] = (1 << desc->comp[0].depth) - 1;
232+ s->max[1] = (1 << desc->comp[1].depth) - 1;
233+ s->max[2] = (1 << desc->comp[2].depth) - 1;
234+ s->max[3] = (1 << desc->comp[3].depth) - 1;
235+
236+ return 0;
237+}
238+
239+static int config_output(AVFilterLink *outlink)
240+{
241+ AVFilterContext *ctx = outlink->src;
242+ CorrContext *s = ctx->priv;
243+ AVFilterLink *mainlink = ctx->inputs[0];
244+ int ret;
245+
246+ ret = ff_framesync_init_dualinput(&s->fs, ctx);
247+ if (ret < 0)
248+ return ret;
249+ outlink->w = mainlink->w;
250+ outlink->h = mainlink->h;
251+ outlink->time_base = mainlink->time_base;
252+ outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
253+ outlink->frame_rate = mainlink->frame_rate;
254+ if ((ret = ff_framesync_configure(&s->fs)) < 0)
255+ return ret;
256+
257+ outlink->time_base = s->fs.time_base;
258+
259+ if (av_cmp_q(mainlink->time_base, outlink->time_base) ||
260+ av_cmp_q(ctx->inputs[1]->time_base, outlink->time_base))
261+ av_log(ctx, AV_LOG_WARNING, "not matching timebases found between first input: %d/%d and second input %d/%d, results may be incorrect!\n",
262+ mainlink->time_base.num, mainlink->time_base.den,
263+ ctx->inputs[1]->time_base.num, ctx->inputs[1]->time_base.den);
264+
265+ return 0;
266+}
267+
268+static int activate(AVFilterContext *ctx)
269+{
270+ CorrContext *s = ctx->priv;
271+ return ff_framesync_activate(&s->fs);
272+}
273+
274+static av_cold void uninit(AVFilterContext *ctx)
275+{
276+ CorrContext *s = ctx->priv;
277+
278+ if (s->nb_frames > 0) {
279+ char buf[256];
280+
281+ buf[0] = 0;
282+ for (int j = 0; j < s->nb_components; j++) {
283+ int c = s->is_rgb ? s->rgba_map[j] : j;
284+ av_strlcatf(buf, sizeof(buf), " %c:%f", s->comps[j], s->score_comp[c] / s->nb_frames);
285+ }
286+
287+ av_log(ctx, AV_LOG_INFO, "%s%s average:%f min:%f max:%f\n",
288+ ctx->filter->name,
289+ buf,
290+ s->score / s->nb_frames,
291+ s->min_score,
292+ s->max_score);
293+ }
294+
295+ ff_framesync_uninit(&s->fs);
296+}
297+
298+static const AVFilterPad corr_inputs[] = {
299+ {
300+ .name = "main",
301+ .type = AVMEDIA_TYPE_VIDEO,
302+ },{
303+ .name = "reference",
304+ .type = AVMEDIA_TYPE_VIDEO,
305+ .config_props = config_input_ref,
306+ },
307+};
308+
309+static const AVFilterPad corr_outputs[] = {
310+ {
311+ .name = "default",
312+ .type = AVMEDIA_TYPE_VIDEO,
313+ .config_props = config_output,
314+ },
315+};
316+
317+static const AVOption options[] = {
318+ { NULL }
319+};
320+
321+#define corr_options options
322+FRAMESYNC_DEFINE_CLASS(corr, CorrContext, fs);
323+
324+const AVFilter ff_vf_corr = {
325+ .name = "corr",
326+ .description = NULL_IF_CONFIG_SMALL("Calculate the correlation between two video streams."),
327+ .preinit = corr_framesync_preinit,
328+ .init = init,
329+ .uninit = uninit,
330+ .activate = activate,
331+ .priv_size = sizeof(CorrContext),
332+ .priv_class = &corr_class,
333+ FILTER_INPUTS(corr_inputs),
334+ FILTER_OUTPUTS(corr_outputs),
335+ FILTER_PIXFMTS_ARRAY(pix_fmts),
336+ .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL |
337+ AVFILTER_FLAG_METADATA_ONLY,
338+};

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t