git.sesse.net Git - ffmpeg/blob - libavfilter/vf_unsharp.c

   1 /*
   2  * Original copyright (c) 2002 Remi Guyomarch <rguyom@pobox.com>
   3  * Port copyright (c) 2010 Daniel G. Taylor <dan@programmer-art.org>
   4  * Relicensed to the LGPL with permission from Remi Guyomarch.
   5  *
   6  * This file is part of FFmpeg.
   7  *
   8  * FFmpeg is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * FFmpeg is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with FFmpeg; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 /**
  24  * @file
  25  * blur / sharpen filter, ported to FFmpeg from MPlayer
  26  * libmpcodecs/unsharp.c.
  27  *
  28  * This code is based on:
  29  *
  30  * An Efficient algorithm for Gaussian blur using finite-state machines
  31  * Frederick M. Waltz and John W. V. Miller
  32  *
  33  * SPIE Conf. on Machine Vision Systems for Inspection and Metrology VII
  34  * Originally published Boston, Nov 98
  35  *
  36  * http://www.engin.umd.umich.edu/~jwvm/ece581/21_GBlur.pdf
  37  */
  38
  39 #include "avfilter.h"
  40 #include "formats.h"
  41 #include "internal.h"
  42 #include "video.h"
  43 #include "libavutil/common.h"
  44 #include "libavutil/imgutils.h"
  45 #include "libavutil/mem.h"
  46 #include "libavutil/opt.h"
  47 #include "libavutil/pixdesc.h"
  48 #include "unsharp.h"
  49
  50 typedef struct TheadData {
  51     UnsharpFilterParam *fp;
  52     uint8_t       *dst;
  53     const uint8_t *src;
  54     int dst_stride;
  55     int src_stride;
  56     int width;
  57     int height;
  58 } ThreadData;
  59
  60 #define DEF_UNSHARP_SLICE_FUNC(name, nbits)                                                           \
  61 static int name##_##nbits(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)                    \
  62 {                                                                                                     \
  63     ThreadData *td = arg;                                                                             \
  64     UnsharpFilterParam *fp = td->fp;                                                                  \
  65     UnsharpContext *s = ctx->priv;                                                                    \
  66     uint32_t **sc = fp->sc;                                                                           \
  67     uint32_t *sr = fp->sr;                                                                            \
  68     const uint##nbits##_t *src2 = NULL;                                                               \
  69     const int amount = fp->amount;                                                                    \
  70     const int steps_x = fp->steps_x;                                                                  \
  71     const int steps_y = fp->steps_y;                                                                  \
  72     const int scalebits = fp->scalebits;                                                              \
  73     const int32_t halfscale = fp->halfscale;                                                          \
  74                                                                                                       \
  75     uint##nbits##_t *dst = (uint##nbits##_t*)td->dst;                                                 \
  76     const uint##nbits##_t *src = (const uint##nbits##_t *)td->src;                                    \
  77     int dst_stride = td->dst_stride;                                                                  \
  78     int src_stride = td->src_stride;                                                                  \
  79     const int width = td->width;                                                                      \
  80     const int height = td->height;                                                                    \
  81     const int sc_offset = jobnr * 2 * steps_y;                                                        \
  82     const int sr_offset = jobnr * (MAX_MATRIX_SIZE - 1);                                              \
  83     const int slice_start = (height * jobnr) / nb_jobs;                                               \
  84     const int slice_end = (height * (jobnr+1)) / nb_jobs;                                             \
  85                                                                                                       \
  86     int32_t res;                                                                                      \
  87     int x, y, z;                                                                                      \
  88     uint32_t tmp1, tmp2;                                                                              \
  89                                                                                                       \
  90     if (!amount) {                                                                                    \
  91         av_image_copy_plane(td->dst + slice_start * dst_stride, dst_stride,                           \
  92                             td->src + slice_start * src_stride, src_stride,                           \
  93                             width * s->bps, slice_end - slice_start);                                 \
  94         return 0;                                                                                     \
  95     }                                                                                                 \
  96                                                                                                       \
  97     for (y = 0; y < 2 * steps_y; y++)                                                                 \
  98         memset(sc[sc_offset + y], 0, sizeof(sc[y][0]) * (width + 2 * steps_x));                       \
  99                                                                                                       \
 100     dst_stride = dst_stride / s->bps;                                                                 \
 101     src_stride = src_stride / s->bps;                                                                 \
 102     /* if this is not the first tile, we start from (slice_start - steps_y) */                        \
 103     /* so we can get smooth result at slice boundary */                                               \
 104     if (slice_start > steps_y) {                                                                      \
 105         src += (slice_start - steps_y) * src_stride;                                                  \
 106         dst += (slice_start - steps_y) * dst_stride;                                                  \
 107     }                                                                                                 \
 108                                                                                                       \
 109     for (y = -steps_y + slice_start; y < steps_y + slice_end; y++) {                                  \
 110         if (y < height)                                                                               \
 111             src2 = src;                                                                               \
 112                                                                                                       \
 113         memset(sr + sr_offset, 0, sizeof(sr[0]) * (2 * steps_x - 1));                                 \
 114         for (x = -steps_x; x < width + steps_x; x++) {                                                \
 115             tmp1 = x <= 0 ? src2[0] : x >= width ? src2[width-1] : src2[x];                           \
 116             for (z = 0; z < steps_x * 2; z += 2) {                                                    \
 117                 tmp2 = sr[sr_offset + z + 0] + tmp1; sr[sr_offset + z + 0] = tmp1;                    \
 118                 tmp1 = sr[sr_offset + z + 1] + tmp2; sr[sr_offset + z + 1] = tmp2;                    \
 119             }                                                                                         \
 120             for (z = 0; z < steps_y * 2; z += 2) {                                                    \
 121                 tmp2 = sc[sc_offset + z + 0][x + steps_x] + tmp1;                                     \
 122                 sc[sc_offset + z + 0][x + steps_x] = tmp1;                                            \
 123                 tmp1 = sc[sc_offset + z + 1][x + steps_x] + tmp2;                                     \
 124                 sc[sc_offset + z + 1][x + steps_x] = tmp2;                                            \
 125             }                                                                                         \
 126             if (x >= steps_x && y >= (steps_y + slice_start)) {                                       \
 127                 const uint##nbits##_t *srx = src - steps_y * src_stride + x - steps_x;                \
 128                 uint##nbits##_t *dsx       = dst - steps_y * dst_stride + x - steps_x;                \
 129                                                                                                       \
 130                 res = (int32_t)*srx + ((((int32_t) * srx -                                            \
 131                       (int32_t)((tmp1 + halfscale) >> scalebits)) * amount) >> (8+nbits));            \
 132                 *dsx = av_clip_uint##nbits(res);                                                      \
 133             }                                                                                         \
 134         }                                                                                             \
 135         if (y >= 0) {                                                                                 \
 136             dst += dst_stride;                                                                        \
 137             src += src_stride;                                                                        \
 138         }                                                                                             \
 139     }                                                                                                 \
 140     return 0;                                                                                         \
 141 }
 142 DEF_UNSHARP_SLICE_FUNC(unsharp_slice, 16)
 143 DEF_UNSHARP_SLICE_FUNC(unsharp_slice, 8)
 144
 145 static int apply_unsharp_c(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
 146 {
 147     AVFilterLink *inlink = ctx->inputs[0];
 148     UnsharpContext *s = ctx->priv;
 149     int i, plane_w[3], plane_h[3];
 150     UnsharpFilterParam *fp[3];
 151     ThreadData td;
 152
 153     plane_w[0] = inlink->w;
 154     plane_w[1] = plane_w[2] = AV_CEIL_RSHIFT(inlink->w, s->hsub);
 155     plane_h[0] = inlink->h;
 156     plane_h[1] = plane_h[2] = AV_CEIL_RSHIFT(inlink->h, s->vsub);
 157     fp[0] = &s->luma;
 158     fp[1] = fp[2] = &s->chroma;
 159     for (i = 0; i < 3; i++) {
 160         td.fp = fp[i];
 161         td.dst = out->data[i];
 162         td.src = in->data[i];
 163         td.width = plane_w[i];
 164         td.height = plane_h[i];
 165         td.dst_stride = out->linesize[i];
 166         td.src_stride = in->linesize[i];
 167         ctx->internal->execute(ctx, s->unsharp_slice, &td, NULL, FFMIN(plane_h[i], s->nb_threads));
 168     }
 169     return 0;
 170 }
 171
 172 static void set_filter_param(UnsharpFilterParam *fp, int msize_x, int msize_y, float amount)
 173 {
 174     fp->msize_x = msize_x;
 175     fp->msize_y = msize_y;
 176     fp->amount = amount * 65536.0;
 177
 178     fp->steps_x = msize_x / 2;
 179     fp->steps_y = msize_y / 2;
 180     fp->scalebits = (fp->steps_x + fp->steps_y) * 2;
 181     fp->halfscale = 1 << (fp->scalebits - 1);
 182 }
 183
 184 static av_cold int init(AVFilterContext *ctx)
 185 {
 186     UnsharpContext *s = ctx->priv;
 187
 188     set_filter_param(&s->luma,   s->lmsize_x, s->lmsize_y, s->lamount);
 189     set_filter_param(&s->chroma, s->cmsize_x, s->cmsize_y, s->camount);
 190
 191     if (s->luma.scalebits >= 26 || s->chroma.scalebits >= 26) {
 192         av_log(ctx, AV_LOG_ERROR, "luma or chroma matrix size too big\n");
 193         return AVERROR(EINVAL);
 194     }
 195     s->apply_unsharp = apply_unsharp_c;
 196     return 0;
 197 }
 198
 199 static int query_formats(AVFilterContext *ctx)
 200 {
 201     static const enum AVPixelFormat pix_fmts[] = {
 202         AV_PIX_FMT_YUV420P,  AV_PIX_FMT_YUV422P,  AV_PIX_FMT_YUV444P,  AV_PIX_FMT_YUV410P,
 203         AV_PIX_FMT_YUV411P,  AV_PIX_FMT_YUV440P,  AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
 204         AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
 205         AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV440P10,
 206         AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12,
 207         AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
 208         AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_NONE
 209     };
 210
 211     AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
 212     if (!fmts_list)
 213         return AVERROR(ENOMEM);
 214     return ff_set_common_formats(ctx, fmts_list);
 215 }
 216
 217 static int init_filter_param(AVFilterContext *ctx, UnsharpFilterParam *fp, const char *effect_type, int width)
 218 {
 219     int z;
 220     UnsharpContext *s = ctx->priv;
 221     const char *effect = fp->amount == 0 ? "none" : fp->amount < 0 ? "blur" : "sharpen";
 222
 223     if  (!(fp->msize_x & fp->msize_y & 1)) {
 224         av_log(ctx, AV_LOG_ERROR,
 225                "Invalid even size for %s matrix size %dx%d\n",
 226                effect_type, fp->msize_x, fp->msize_y);
 227         return AVERROR(EINVAL);
 228     }
 229
 230     av_log(ctx, AV_LOG_VERBOSE, "effect:%s type:%s msize_x:%d msize_y:%d amount:%0.2f\n",
 231            effect, effect_type, fp->msize_x, fp->msize_y, fp->amount / 65535.0);
 232
 233     fp->sr = av_malloc_array((MAX_MATRIX_SIZE - 1) * s->nb_threads, sizeof(uint32_t));
 234     fp->sc = av_mallocz_array(2 * fp->steps_y * s->nb_threads, sizeof(uint32_t *));
 235     if (!fp->sr || !fp->sc)
 236         return AVERROR(ENOMEM);
 237
 238     for (z = 0; z < 2 * fp->steps_y * s->nb_threads; z++)
 239         if (!(fp->sc[z] = av_malloc_array(width + 2 * fp->steps_x,
 240                                           sizeof(*(fp->sc[z])))))
 241             return AVERROR(ENOMEM);
 242
 243     return 0;
 244 }
 245
 246 static int config_input(AVFilterLink *inlink)
 247 {
 248     UnsharpContext *s = inlink->dst->priv;
 249     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
 250     int ret;
 251
 252     s->hsub = desc->log2_chroma_w;
 253     s->vsub = desc->log2_chroma_h;
 254     s->bitdepth = desc->comp[0].depth;
 255     s->bps = s->bitdepth > 8 ? 2 : 1;
 256     s->unsharp_slice = s->bitdepth > 8 ? unsharp_slice_16 : unsharp_slice_8;
 257
 258     // ensure (height / nb_threads) > 4 * steps_y,
 259     // so that we don't have too much overlap between two threads
 260     s->nb_threads = FFMIN(ff_filter_get_nb_threads(inlink->dst),
 261                           inlink->h / (4 * s->luma.steps_y));
 262
 263     ret = init_filter_param(inlink->dst, &s->luma,   "luma",   inlink->w);
 264     if (ret < 0)
 265         return ret;
 266     ret = init_filter_param(inlink->dst, &s->chroma, "chroma", AV_CEIL_RSHIFT(inlink->w, s->hsub));
 267     if (ret < 0)
 268         return ret;
 269
 270     return 0;
 271 }
 272
 273 static void free_filter_param(UnsharpFilterParam *fp, int nb_threads)
 274 {
 275     int z;
 276
 277     if (fp->sc) {
 278         for (z = 0; z < 2 * fp->steps_y * nb_threads; z++)
 279             av_freep(&fp->sc[z]);
 280         av_freep(&fp->sc);
 281     }
 282     av_freep(&fp->sr);
 283 }
 284
 285 static av_cold void uninit(AVFilterContext *ctx)
 286 {
 287     UnsharpContext *s = ctx->priv;
 288
 289     free_filter_param(&s->luma, s->nb_threads);
 290     free_filter_param(&s->chroma, s->nb_threads);
 291 }
 292
 293 static int filter_frame(AVFilterLink *link, AVFrame *in)
 294 {
 295     UnsharpContext *s = link->dst->priv;
 296     AVFilterLink *outlink   = link->dst->outputs[0];
 297     AVFrame *out;
 298     int ret = 0;
 299
 300     out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
 301     if (!out) {
 302         av_frame_free(&in);
 303         return AVERROR(ENOMEM);
 304     }
 305     av_frame_copy_props(out, in);
 306
 307     ret = s->apply_unsharp(link->dst, in, out);
 308
 309     av_frame_free(&in);
 310
 311     if (ret < 0) {
 312         av_frame_free(&out);
 313         return ret;
 314     }
 315     return ff_filter_frame(outlink, out);
 316 }
 317
 318 #define OFFSET(x) offsetof(UnsharpContext, x)
 319 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
 320 #define MIN_SIZE 3
 321 #define MAX_SIZE 23
 322 static const AVOption unsharp_options[] = {
 323     { "luma_msize_x",   "set luma matrix horizontal size",   OFFSET(lmsize_x), AV_OPT_TYPE_INT,   { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS },
 324     { "lx",             "set luma matrix horizontal size",   OFFSET(lmsize_x), AV_OPT_TYPE_INT,   { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS },
 325     { "luma_msize_y",   "set luma matrix vertical size",     OFFSET(lmsize_y), AV_OPT_TYPE_INT,   { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS },
 326     { "ly",             "set luma matrix vertical size",     OFFSET(lmsize_y), AV_OPT_TYPE_INT,   { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS },
 327     { "luma_amount",    "set luma effect strength",          OFFSET(lamount),  AV_OPT_TYPE_FLOAT, { .dbl = 1 },       -2,        5, FLAGS },
 328     { "la",             "set luma effect strength",          OFFSET(lamount),  AV_OPT_TYPE_FLOAT, { .dbl = 1 },       -2,        5, FLAGS },
 329     { "chroma_msize_x", "set chroma matrix horizontal size", OFFSET(cmsize_x), AV_OPT_TYPE_INT,   { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS },
 330     { "cx",             "set chroma matrix horizontal size", OFFSET(cmsize_x), AV_OPT_TYPE_INT,   { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS },
 331     { "chroma_msize_y", "set chroma matrix vertical size",   OFFSET(cmsize_y), AV_OPT_TYPE_INT,   { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS },
 332     { "cy",             "set chroma matrix vertical size",   OFFSET(cmsize_y), AV_OPT_TYPE_INT,   { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS },
 333     { "chroma_amount",  "set chroma effect strength",        OFFSET(camount),  AV_OPT_TYPE_FLOAT, { .dbl = 0 },       -2,        5, FLAGS },
 334     { "ca",             "set chroma effect strength",        OFFSET(camount),  AV_OPT_TYPE_FLOAT, { .dbl = 0 },       -2,        5, FLAGS },
 335     { "opencl",         "ignored",                           OFFSET(opencl),   AV_OPT_TYPE_BOOL,  { .i64 = 0 },        0,        1, FLAGS },
 336     { NULL }
 337 };
 338
 339 AVFILTER_DEFINE_CLASS(unsharp);
 340
 341 static const AVFilterPad avfilter_vf_unsharp_inputs[] = {
 342     {
 343         .name         = "default",
 344         .type         = AVMEDIA_TYPE_VIDEO,
 345         .filter_frame = filter_frame,
 346         .config_props = config_input,
 347     },
 348     { NULL }
 349 };
 350
 351 static const AVFilterPad avfilter_vf_unsharp_outputs[] = {
 352     {
 353         .name = "default",
 354         .type = AVMEDIA_TYPE_VIDEO,
 355     },
 356     { NULL }
 357 };
 358
 359 const AVFilter ff_vf_unsharp = {
 360     .name          = "unsharp",
 361     .description   = NULL_IF_CONFIG_SMALL("Sharpen or blur the input video."),
 362     .priv_size     = sizeof(UnsharpContext),
 363     .priv_class    = &unsharp_class,
 364     .init          = init,
 365     .uninit        = uninit,
 366     .query_formats = query_formats,
 367     .inputs        = avfilter_vf_unsharp_inputs,
 368     .outputs       = avfilter_vf_unsharp_outputs,
 369     .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
 370 };