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
query_formats(AVFilterContext * ctx)62 static int query_formats(AVFilterContext *ctx)
63 {
64 static const enum AVPixelFormat pix_fmts[] = {
65 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P,
66 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
67 AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUV420P,
68 AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
69 AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
70 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
71 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
72 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12,
73 AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
74 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
75 AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA444P9,
76 AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA444P10,
77 AV_PIX_FMT_YUVA422P12, AV_PIX_FMT_YUVA444P12,
78 AV_PIX_FMT_YUVA420P16, AV_PIX_FMT_YUVA422P16, AV_PIX_FMT_YUVA444P16,
79 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
80 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
81 AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
82 AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16,
83 AV_PIX_FMT_NONE
84 };
85
86 return ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
87 }
88
89 typedef struct ThreadData {
90 int width;
91 int height;
92 int src_linesize;
93 int dst_linesize;
94 uint8_t *src;
95 uint8_t *dst;
96 } ThreadData;
97
98 #define PRE_CALCULATE_ROW(type, name) \
99 static int pre_calculate_row_##name(AVFilterContext *ctx, void *arg, \
100 int jobnr, int nb_jobs) \
101 { \
102 ThreadData *td = arg; \
103 YAEPContext *s = ctx->priv; \
104 \
105 const int width = td->width; \
106 const int height = td->height; \
107 const int linesize = td->src_linesize / sizeof(type); \
108 const int sat_linesize = s->sat_linesize; \
109 \
110 const int starty = height * jobnr / nb_jobs; \
111 const int endy = height * (jobnr+1) / nb_jobs; \
112 \
113 uint64_t *sat = s->sat + (starty + 1) * sat_linesize; \
114 uint64_t *square_sat = s->square_sat + (starty + 1) * sat_linesize; \
115 const type *src = (const type *)td->src + starty * linesize; \
116 \
117 int x, y; \
118 \
119 for (y = starty; y < endy; y++) { \
120 for (x = 0; x < width; x++) { \
121 sat[x+1] = sat[x] + src[x]; \
122 square_sat[x+1] = square_sat[x] + (uint64_t)src[x] * src[x]; \
123 } \
124 sat += sat_linesize; \
125 square_sat += sat_linesize; \
126 src += linesize; \
127 } \
128 \
129 return 0; \
130 }
131
PRE_CALCULATE_ROW(uint8_t,byte)132 PRE_CALCULATE_ROW(uint8_t, byte)
133 PRE_CALCULATE_ROW(uint16_t, word)
134
135 static int pre_calculate_col(AVFilterContext *ctx, void *arg,
136 int jobnr, int nb_jobs)
137 {
138 ThreadData *td = arg;
139 YAEPContext *s = ctx->priv;
140
141 const int width = td->width;
142 const int height = td->height;
143 const int sat_linesize = s->sat_linesize;
144
145 const int startx = width * jobnr / nb_jobs;
146 const int endx = width * (jobnr + 1) / nb_jobs;
147
148 uint64_t *sat, *square_sat;
149 int x, y;
150
151 for (x = startx; x < endx; x++) {
152 sat = s->sat + x + 1;
153 square_sat = s->square_sat + x + 1;
154 for (y = 0; y < height; y++) {
155 *(sat+sat_linesize) += *sat;
156 *(square_sat+sat_linesize) += *square_sat;
157 sat += sat_linesize;
158 square_sat += sat_linesize;
159 }
160 }
161
162 return 0;
163 }
164
165 #define FILTER_SLICE(type, name) \
166 static int filter_slice_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
167 { \
168 ThreadData *td = arg; \
169 YAEPContext *s = ctx->priv; \
170 \
171 const int width = td->width; \
172 const int height = td->height; \
173 const int src_linesize = td->src_linesize / sizeof(type); \
174 const int dst_linesize = td->dst_linesize / sizeof(type); \
175 const int sat_linesize = s->sat_linesize; \
176 const int sigma = s->sigma; \
177 const int radius = s->radius; \
178 \
179 uint64_t *sat = s->sat; \
180 uint64_t *square_sat = s->square_sat; \
181 const type *src = (const type *)td->src; \
182 type *dst = (type *)td->dst; \
183 \
184 const int starty = height * jobnr / nb_jobs; \
185 const int endy = height * (jobnr + 1) / nb_jobs; \
186 \
187 int x, y; \
188 int lower_x, higher_x; \
189 int lower_y, higher_y; \
190 int dist_y, count; \
191 uint64_t sum, square_sum, mean, var; \
192 \
193 for (y = starty; y < endy; y++) { \
194 lower_y = y - radius < 0 ? 0 : y - radius; \
195 higher_y = y + radius + 1 > height ? height : y + radius + 1; \
196 dist_y = higher_y - lower_y; \
197 for (x = 0; x < width; x++) { \
198 lower_x = x - radius < 0 ? 0 : x - radius; \
199 higher_x = x + radius + 1 > width ? width : x + radius + 1; \
200 count = dist_y * (higher_x - lower_x); \
201 sum = sat[higher_y * sat_linesize + higher_x] \
202 - sat[higher_y * sat_linesize + lower_x] \
203 - sat[lower_y * sat_linesize + higher_x] \
204 + sat[lower_y * sat_linesize + lower_x]; \
205 square_sum = square_sat[higher_y * sat_linesize + higher_x] \
206 - square_sat[higher_y * sat_linesize + lower_x] \
207 - square_sat[lower_y * sat_linesize + higher_x] \
208 + square_sat[lower_y * sat_linesize + lower_x]; \
209 mean = sum / count; \
210 var = (square_sum - sum * sum / count) / count; \
211 dst[y * dst_linesize + x] = (sigma * mean + var * src[y * src_linesize + x]) / (sigma + var); \
212 } \
213 } \
214 return 0; \
215 }
216
FILTER_SLICE(uint8_t,byte)217 FILTER_SLICE(uint8_t, byte)
218 FILTER_SLICE(uint16_t, word)
219
220 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
221 {
222 AVFilterContext *ctx = inlink->dst;
223 YAEPContext *s = ctx->priv;
224 AVFilterLink *outlink = ctx->outputs[0];
225 AVFrame *out;
226 int plane;
227 const int nb_threads = ff_filter_get_nb_threads(ctx);
228 ThreadData td;
229
230 if (av_frame_is_writable(in)) {
231 out = in;
232 } else {
233 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
234 if (!out) {
235 av_frame_free(&in);
236 return AVERROR(ENOMEM);
237 }
238 av_frame_copy_props(out, in);
239 }
240
241 for (plane = 0; plane < s->nb_planes; plane++) {
242 if (!s->radius || !(s->planes & (1<<plane))) {
243 if (out != in) {
244 av_image_copy_plane(out->data[plane], out->linesize[plane],
245 in->data[plane], in->linesize[plane],
246 s->planewidth[plane] * ((s->depth + 7) / 8),
247 s->planeheight[plane]);
248 }
249 continue;
250 }
251
252 td.width = s->planewidth[plane];
253 td.height = s->planeheight[plane];
254 td.src = in->data[plane];
255 td.src_linesize = in->linesize[plane];
256 ctx->internal->execute(ctx, s->pre_calculate_row, &td, NULL, FFMIN(td.height, nb_threads));
257 ctx->internal->execute(ctx, pre_calculate_col, &td, NULL, FFMIN(td.width, nb_threads));
258
259 td.dst = out->data[plane];
260 td.dst_linesize = out->linesize[plane];
261 ctx->internal->execute(ctx, s->filter_slice, &td, NULL, FFMIN(td.height, nb_threads));
262 }
263
264 if (out != in)
265 av_frame_free(&in);
266
267 return ff_filter_frame(outlink, out);
268 }
269
config_input(AVFilterLink * inlink)270 static int config_input(AVFilterLink *inlink)
271 {
272 YAEPContext *s = inlink->dst->priv;
273 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
274
275 s->depth = desc->comp[0].depth;
276 s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
277 s->planewidth[0] = s->planewidth[3] = inlink->w;
278 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
279 s->planeheight[0] = s->planeheight[3] = inlink->h;
280 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
281
282 s->radius = FFMIN(s->radius, AV_CEIL_RSHIFT(FFMIN(inlink->w, inlink->h), 1));
283
284 if (s->depth <= 8) {
285 s->pre_calculate_row = pre_calculate_row_byte;
286 s->filter_slice = filter_slice_byte;
287 } else {
288 s->pre_calculate_row = pre_calculate_row_word;
289 s->filter_slice = filter_slice_word;
290 }
291
292 // padding one row on the top, and padding one col on the left, that is why + 1 below
293 s->sat_linesize = inlink->w + 1;
294 s->sat = av_mallocz_array(inlink->h + 1, s->sat_linesize*sizeof(*s->sat));
295 if (!s->sat)
296 return AVERROR(ENOMEM);
297
298 s->square_sat = av_mallocz_array(inlink->h + 1, s->sat_linesize*sizeof(*s->square_sat));
299 if (!s->square_sat)
300 return AVERROR(ENOMEM);
301
302 return 0;
303 }
304
305 static const AVFilterPad yaep_inputs[] = {
306 {
307 .name = "default",
308 .type = AVMEDIA_TYPE_VIDEO,
309 .config_props = config_input,
310 .filter_frame = filter_frame,
311 },
312 { NULL }
313 };
314
315 static const AVFilterPad yaep_outputs[] = {
316 {
317 .name = "default",
318 .type = AVMEDIA_TYPE_VIDEO,
319 },
320 { NULL }
321 };
322
323 #define OFFSET(x) offsetof(YAEPContext, x)
324 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
325
326 static const AVOption yaepblur_options[] = {
327 { "radius", "set window radius", OFFSET(radius), AV_OPT_TYPE_INT, {.i64=3}, 0, INT_MAX, .flags=FLAGS },
328 { "r" , "set window radius", OFFSET(radius), AV_OPT_TYPE_INT, {.i64=3}, 0, INT_MAX, .flags=FLAGS },
329 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=1}, 0, 0xF, .flags=FLAGS },
330 { "p", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=1}, 0, 0xF, .flags=FLAGS },
331 { "sigma", "set blur strength", OFFSET(sigma), AV_OPT_TYPE_INT, {.i64=128}, 1, INT_MAX, .flags=FLAGS },
332 { "s", "set blur strength", OFFSET(sigma), AV_OPT_TYPE_INT, {.i64=128}, 1, INT_MAX, .flags=FLAGS },
333 { NULL }
334 };
335
336 AVFILTER_DEFINE_CLASS(yaepblur);
337
338 AVFilter ff_vf_yaepblur = {
339 .name = "yaepblur",
340 .description = NULL_IF_CONFIG_SMALL("Yet another edge preserving blur filter."),
341 .priv_size = sizeof(YAEPContext),
342 .priv_class = &yaepblur_class,
343 .uninit = uninit,
344 .query_formats = query_formats,
345 .inputs = yaep_inputs,
346 .outputs = yaep_outputs,
347 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
348 .process_command = ff_filter_process_command,
349 };
350