1 /*
2 * Copyright (C) 2019 Leo Zhang <leozhang@qiyi.com>
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 * yaep(yet another edge preserving) blur filter
24 *
25 * This implementation is based on an algorithm described in
26 * "J. S. Lee, Digital image enhancement and noise filtering by use of local statistics, IEEE Trans. Pattern
27 * Anal. Mach. Intell. PAMI-2, 1980."
28 */
29
30 #include "libavutil/opt.h"
31 #include "libavutil/imgutils.h"
32 #include "avfilter.h"
33 #include "internal.h"
34
35 typedef struct YAEPContext {
36 const AVClass *class;
37
38 int planes;
39 int radius;
40 int sigma;
41
42 int nb_planes;
43 int planewidth[4];
44 int planeheight[4];
45 int depth;
46
47 uint64_t *sat; ///< summed area table
48 uint64_t *square_sat; ///< square summed area table
49 int sat_linesize;
50
51 int (*pre_calculate_row)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
52 int (*filter_slice )(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
53 } YAEPContext;
54
uninit(AVFilterContext * ctx)55 static av_cold void uninit(AVFilterContext *ctx)
56 {
57 YAEPContext *s = ctx->priv;
58 av_freep(&s->sat);
59 av_freep(&s->square_sat);
60 }
61
62 static const enum AVPixelFormat pix_fmts[] = {
63 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P,
64 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
65 AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUV420P,
66 AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
67 AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
68 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
69 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
70 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12,
71 AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
72 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
73 AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA444P9,
74 AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA444P10,
75 AV_PIX_FMT_YUVA422P12, AV_PIX_FMT_YUVA444P12,
76 AV_PIX_FMT_YUVA420P16, AV_PIX_FMT_YUVA422P16, AV_PIX_FMT_YUVA444P16,
77 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
78 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
79 AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
80 AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16,
81 AV_PIX_FMT_NONE
82 };
83
84 typedef struct ThreadData {
85 int width;
86 int height;
87 int src_linesize;
88 int dst_linesize;
89 uint8_t *src;
90 uint8_t *dst;
91 } ThreadData;
92
93 #define PRE_CALCULATE_ROW(type, name) \
94 static int pre_calculate_row_##name(AVFilterContext *ctx, void *arg, \
95 int jobnr, int nb_jobs) \
96 { \
97 ThreadData *td = arg; \
98 YAEPContext *s = ctx->priv; \
99 \
100 const int width = td->width; \
101 const int height = td->height; \
102 const int linesize = td->src_linesize / sizeof(type); \
103 const int sat_linesize = s->sat_linesize; \
104 \
105 const int starty = height * jobnr / nb_jobs; \
106 const int endy = height * (jobnr+1) / nb_jobs; \
107 \
108 uint64_t *sat = s->sat + (starty + 1) * sat_linesize; \
109 uint64_t *square_sat = s->square_sat + (starty + 1) * sat_linesize; \
110 const type *src = (const type *)td->src + starty * linesize; \
111 \
112 int x, y; \
113 \
114 for (y = starty; y < endy; y++) { \
115 for (x = 0; x < width; x++) { \
116 sat[x+1] = sat[x] + src[x]; \
117 square_sat[x+1] = square_sat[x] + (uint64_t)src[x] * src[x]; \
118 } \
119 sat += sat_linesize; \
120 square_sat += sat_linesize; \
121 src += linesize; \
122 } \
123 \
124 return 0; \
125 }
126
PRE_CALCULATE_ROW(uint8_t,byte)127 PRE_CALCULATE_ROW(uint8_t, byte)
128 PRE_CALCULATE_ROW(uint16_t, word)
129
130 static int pre_calculate_col(AVFilterContext *ctx, void *arg,
131 int jobnr, int nb_jobs)
132 {
133 ThreadData *td = arg;
134 YAEPContext *s = ctx->priv;
135
136 const int width = td->width;
137 const int height = td->height;
138 const int sat_linesize = s->sat_linesize;
139
140 const int startx = width * jobnr / nb_jobs;
141 const int endx = width * (jobnr + 1) / nb_jobs;
142
143 uint64_t *sat, *square_sat;
144 int x, y;
145
146 for (x = startx; x < endx; x++) {
147 sat = s->sat + x + 1;
148 square_sat = s->square_sat + x + 1;
149 for (y = 0; y < height; y++) {
150 *(sat+sat_linesize) += *sat;
151 *(square_sat+sat_linesize) += *square_sat;
152 sat += sat_linesize;
153 square_sat += sat_linesize;
154 }
155 }
156
157 return 0;
158 }
159
160 #define FILTER_SLICE(type, name) \
161 static int filter_slice_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
162 { \
163 ThreadData *td = arg; \
164 YAEPContext *s = ctx->priv; \
165 \
166 const int width = td->width; \
167 const int height = td->height; \
168 const int src_linesize = td->src_linesize / sizeof(type); \
169 const int dst_linesize = td->dst_linesize / sizeof(type); \
170 const int sat_linesize = s->sat_linesize; \
171 const int sigma = s->sigma; \
172 const int radius = s->radius; \
173 \
174 uint64_t *sat = s->sat; \
175 uint64_t *square_sat = s->square_sat; \
176 const type *src = (const type *)td->src; \
177 type *dst = (type *)td->dst; \
178 \
179 const int starty = height * jobnr / nb_jobs; \
180 const int endy = height * (jobnr + 1) / nb_jobs; \
181 \
182 int x, y; \
183 int lower_x, higher_x; \
184 int lower_y, higher_y; \
185 int dist_y, count; \
186 uint64_t sum, square_sum, mean, var; \
187 \
188 for (y = starty; y < endy; y++) { \
189 lower_y = y - radius < 0 ? 0 : y - radius; \
190 higher_y = y + radius + 1 > height ? height : y + radius + 1; \
191 dist_y = higher_y - lower_y; \
192 for (x = 0; x < width; x++) { \
193 lower_x = x - radius < 0 ? 0 : x - radius; \
194 higher_x = x + radius + 1 > width ? width : x + radius + 1; \
195 count = dist_y * (higher_x - lower_x); \
196 sum = sat[higher_y * sat_linesize + higher_x] \
197 - sat[higher_y * sat_linesize + lower_x] \
198 - sat[lower_y * sat_linesize + higher_x] \
199 + sat[lower_y * sat_linesize + lower_x]; \
200 square_sum = square_sat[higher_y * sat_linesize + higher_x] \
201 - square_sat[higher_y * sat_linesize + lower_x] \
202 - square_sat[lower_y * sat_linesize + higher_x] \
203 + square_sat[lower_y * sat_linesize + lower_x]; \
204 mean = sum / count; \
205 var = (square_sum - sum * sum / count) / count; \
206 dst[y * dst_linesize + x] = (sigma * mean + var * src[y * src_linesize + x]) / (sigma + var); \
207 } \
208 } \
209 return 0; \
210 }
211
FILTER_SLICE(uint8_t,byte)212 FILTER_SLICE(uint8_t, byte)
213 FILTER_SLICE(uint16_t, word)
214
215 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
216 {
217 AVFilterContext *ctx = inlink->dst;
218 YAEPContext *s = ctx->priv;
219 AVFilterLink *outlink = ctx->outputs[0];
220 AVFrame *out;
221 int plane;
222 const int nb_threads = ff_filter_get_nb_threads(ctx);
223 ThreadData td;
224
225 if (av_frame_is_writable(in)) {
226 out = in;
227 } else {
228 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
229 if (!out) {
230 av_frame_free(&in);
231 return AVERROR(ENOMEM);
232 }
233 av_frame_copy_props(out, in);
234 }
235
236 for (plane = 0; plane < s->nb_planes; plane++) {
237 if (!s->radius || !(s->planes & (1<<plane))) {
238 if (out != in) {
239 av_image_copy_plane(out->data[plane], out->linesize[plane],
240 in->data[plane], in->linesize[plane],
241 s->planewidth[plane] * ((s->depth + 7) / 8),
242 s->planeheight[plane]);
243 }
244 continue;
245 }
246
247 td.width = s->planewidth[plane];
248 td.height = s->planeheight[plane];
249 td.src = in->data[plane];
250 td.src_linesize = in->linesize[plane];
251 ff_filter_execute(ctx, s->pre_calculate_row, &td, NULL,
252 FFMIN(td.height, nb_threads));
253 ff_filter_execute(ctx, pre_calculate_col, &td, NULL,
254 FFMIN(td.width, nb_threads));
255
256 td.dst = out->data[plane];
257 td.dst_linesize = out->linesize[plane];
258 ff_filter_execute(ctx, s->filter_slice, &td, NULL,
259 FFMIN(td.height, nb_threads));
260 }
261
262 if (out != in)
263 av_frame_free(&in);
264
265 return ff_filter_frame(outlink, out);
266 }
267
config_input(AVFilterLink * inlink)268 static int config_input(AVFilterLink *inlink)
269 {
270 YAEPContext *s = inlink->dst->priv;
271 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
272
273 s->depth = desc->comp[0].depth;
274 s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
275 s->planewidth[0] = s->planewidth[3] = inlink->w;
276 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
277 s->planeheight[0] = s->planeheight[3] = inlink->h;
278 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
279
280 s->radius = FFMIN(s->radius, AV_CEIL_RSHIFT(FFMIN(inlink->w, inlink->h), 1));
281
282 if (s->depth <= 8) {
283 s->pre_calculate_row = pre_calculate_row_byte;
284 s->filter_slice = filter_slice_byte;
285 } else {
286 s->pre_calculate_row = pre_calculate_row_word;
287 s->filter_slice = filter_slice_word;
288 }
289
290 // padding one row on the top, and padding one col on the left, that is why + 1 below
291 s->sat_linesize = inlink->w + 1;
292 s->sat = av_calloc(inlink->h + 1, s->sat_linesize * sizeof(*s->sat));
293 if (!s->sat)
294 return AVERROR(ENOMEM);
295
296 s->square_sat = av_calloc(inlink->h + 1, s->sat_linesize * sizeof(*s->square_sat));
297 if (!s->square_sat)
298 return AVERROR(ENOMEM);
299
300 return 0;
301 }
302
303 static const AVFilterPad yaep_inputs[] = {
304 {
305 .name = "default",
306 .type = AVMEDIA_TYPE_VIDEO,
307 .config_props = config_input,
308 .filter_frame = filter_frame,
309 },
310 };
311
312 static const AVFilterPad yaep_outputs[] = {
313 {
314 .name = "default",
315 .type = AVMEDIA_TYPE_VIDEO,
316 },
317 };
318
319 #define OFFSET(x) offsetof(YAEPContext, x)
320 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
321
322 static const AVOption yaepblur_options[] = {
323 { "radius", "set window radius", OFFSET(radius), AV_OPT_TYPE_INT, {.i64=3}, 0, INT_MAX, .flags=FLAGS },
324 { "r" , "set window radius", OFFSET(radius), AV_OPT_TYPE_INT, {.i64=3}, 0, INT_MAX, .flags=FLAGS },
325 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=1}, 0, 0xF, .flags=FLAGS },
326 { "p", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=1}, 0, 0xF, .flags=FLAGS },
327 { "sigma", "set blur strength", OFFSET(sigma), AV_OPT_TYPE_INT, {.i64=128}, 1, INT_MAX, .flags=FLAGS },
328 { "s", "set blur strength", OFFSET(sigma), AV_OPT_TYPE_INT, {.i64=128}, 1, INT_MAX, .flags=FLAGS },
329 { NULL }
330 };
331
332 AVFILTER_DEFINE_CLASS(yaepblur);
333
334 const AVFilter ff_vf_yaepblur = {
335 .name = "yaepblur",
336 .description = NULL_IF_CONFIG_SMALL("Yet another edge preserving blur filter."),
337 .priv_size = sizeof(YAEPContext),
338 .priv_class = &yaepblur_class,
339 .uninit = uninit,
340 FILTER_INPUTS(yaep_inputs),
341 FILTER_OUTPUTS(yaep_outputs),
342 FILTER_PIXFMTS_ARRAY(pix_fmts),
343 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
344 .process_command = ff_filter_process_command,
345 };
346