libavfilter/vf_hqdn3d.c

   1 /*
   2  * Copyright (c) 2003 Daniel Moreno <comac AT comac DOT darktech DOT org>
   3  * Copyright (c) 2010 Baptiste Coudurier
   4  * Copyright (c) 2012 Loren Merritt
   5  *
   6  * This file is part of Libav, ported from MPlayer.
   7  *
   8  * Libav is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 2 of the License, or
  11  * (at your option) any later version.
  12  *
  13  * Libav 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
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License along
  19  * with Libav; if not, write to the Free Software Foundation, Inc.,
  20  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  21  */
  22
  23 /**
  24  * @file
  25  * high quality 3d video denoiser, ported from MPlayer
  26  * libmpcodecs/vf_hqdn3d.c.
  27  */
  28
  29 #include "config.h"
  30 #include "libavutil/common.h"
  31 #include "libavutil/pixdesc.h"
  32 #include "libavutil/intreadwrite.h"
  33 #include "avfilter.h"
  34 #include "formats.h"
  35 #include "internal.h"
  36 #include "video.h"
  37 #include "vf_hqdn3d.h"
  38
  39 #define LUT_BITS (depth==16 ? 8 : 4)
  40 #define RIGHTSHIFT(a,b) (((a)+(((1<<(b))-1)>>1))>>(b))
  41 #define LOAD(x) ((depth==8 ? src[x] : AV_RN16A(src+(x)*2)) << (16-depth))
  42 #define STORE(x,val) (depth==8 ? dst[x] = RIGHTSHIFT(val, 16-depth)\
  43                     : AV_WN16A(dst+(x)*2, RIGHTSHIFT(val, 16-depth)))
  44
  45 av_always_inline
  46 static uint32_t lowpass(int prev, int cur, int16_t *coef, int depth)
  47 {
  48     int d = (prev - cur) >> (8 - LUT_BITS);
  49     return cur + coef[d];
  50 }
  51
  52 av_always_inline
  53 static void denoise_temporal(uint8_t *src, uint8_t *dst,
  54                              uint16_t *frame_ant,
  55                              int w, int h, int sstride, int dstride,
  56                              int16_t *temporal, int depth)
  57 {
  58     long x, y;
  59     uint32_t tmp;
  60
  61     temporal += 256 << LUT_BITS;
  62
  63     for (y = 0; y < h; y++) {
  64         for (x = 0; x < w; x++) {
  65             frame_ant[x] = tmp = lowpass(frame_ant[x], LOAD(x), temporal, depth);
  66             STORE(x, tmp);
  67         }
  68         src += sstride;
  69         dst += dstride;
  70         frame_ant += w;
  71     }
  72 }
  73
  74 av_always_inline
  75 static void denoise_spatial(HQDN3DContext *hqdn3d,
  76                             uint8_t *src, uint8_t *dst,
  77                             uint16_t *line_ant, uint16_t *frame_ant,
  78                             int w, int h, int sstride, int dstride,
  79                             int16_t *spatial, int16_t *temporal, int depth)
  80 {
  81     long x, y;
  82     uint32_t pixel_ant;
  83     uint32_t tmp;
  84
  85     spatial  += 256 << LUT_BITS;
  86     temporal += 256 << LUT_BITS;
  87
  88     /* First line has no top neighbor. Only left one for each tmp and
  89      * last frame */
  90     pixel_ant = LOAD(0);
  91     for (x = 0; x < w; x++) {
  92         line_ant[x] = tmp = pixel_ant = lowpass(pixel_ant, LOAD(x), spatial, depth);
  93         frame_ant[x] = tmp = lowpass(frame_ant[x], tmp, temporal, depth);
  94         STORE(x, tmp);
  95     }
  96
  97     for (y = 1; y < h; y++) {
  98         src += sstride;
  99         dst += dstride;
 100         frame_ant += w;
 101         if (hqdn3d->denoise_row[depth]) {
 102             hqdn3d->denoise_row[depth](src, dst, line_ant, frame_ant, w, spatial, temporal);
 103             continue;
 104         }
 105         pixel_ant = LOAD(0);
 106         for (x = 0; x < w-1; x++) {
 107             line_ant[x] = tmp = lowpass(line_ant[x], pixel_ant, spatial, depth);
 108             pixel_ant = lowpass(pixel_ant, LOAD(x+1), spatial, depth);
 109             frame_ant[x] = tmp = lowpass(frame_ant[x], tmp, temporal, depth);
 110             STORE(x, tmp);
 111         }
 112         line_ant[x] = tmp = lowpass(line_ant[x], pixel_ant, spatial, depth);
 113         frame_ant[x] = tmp = lowpass(frame_ant[x], tmp, temporal, depth);
 114         STORE(x, tmp);
 115     }
 116 }
 117
 118 av_always_inline
 119 static void denoise_depth(HQDN3DContext *hqdn3d,
 120                           uint8_t *src, uint8_t *dst,
 121                           uint16_t *line_ant, uint16_t **frame_ant_ptr,
 122                           int w, int h, int sstride, int dstride,
 123                           int16_t *spatial, int16_t *temporal, int depth)
 124 {
 125     // FIXME: For 16bit depth, frame_ant could be a pointer to the previous
 126     // filtered frame rather than a separate buffer.
 127     long x, y;
 128     uint16_t *frame_ant = *frame_ant_ptr;
 129     if (!frame_ant) {
 130         uint8_t *frame_src = src;
 131         *frame_ant_ptr = frame_ant = av_malloc(w*h*sizeof(uint16_t));
 132         for (y = 0; y < h; y++, src += sstride, frame_ant += w)
 133             for (x = 0; x < w; x++)
 134                 frame_ant[x] = LOAD(x);
 135         src = frame_src;
 136         frame_ant = *frame_ant_ptr;
 137     }
 138
 139     if (spatial[0])
 140         denoise_spatial(hqdn3d, src, dst, line_ant, frame_ant,
 141                         w, h, sstride, dstride, spatial, temporal, depth);
 142     else
 143         denoise_temporal(src, dst, frame_ant,
 144                          w, h, sstride, dstride, temporal, depth);
 145 }
 146
 147 #define denoise(...) \
 148     switch (hqdn3d->depth) {\
 149         case  8: denoise_depth(__VA_ARGS__,  8); break;\
 150         case  9: denoise_depth(__VA_ARGS__,  9); break;\
 151         case 10: denoise_depth(__VA_ARGS__, 10); break;\
 152         case 16: denoise_depth(__VA_ARGS__, 16); break;\
 153     }
 154
 155 static int16_t *precalc_coefs(double dist25, int depth)
 156 {
 157     int i;
 158     double gamma, simil, C;
 159     int16_t *ct = av_malloc((512<<LUT_BITS)*sizeof(int16_t));
 160     if (!ct)
 161         return NULL;
 162
 163     gamma = log(0.25) / log(1.0 - FFMIN(dist25,252.0)/255.0 - 0.00001);
 164
 165     for (i = -255<<LUT_BITS; i <= 255<<LUT_BITS; i++) {
 166         double f = ((i<<(9-LUT_BITS)) + (1<<(8-LUT_BITS)) - 1) / 512.0; // midpoint of the bin
 167         simil = 1.0 - FFABS(f) / 255.0;
 168         C = pow(simil, gamma) * 256.0 * f;
 169         ct[(256<<LUT_BITS)+i] = lrint(C);
 170     }
 171
 172     ct[0] = !!dist25;
 173     return ct;
 174 }
 175
 176 #define PARAM1_DEFAULT 4.0
 177 #define PARAM2_DEFAULT 3.0
 178 #define PARAM3_DEFAULT 6.0
 179
 180 static int init(AVFilterContext *ctx, const char *args)
 181 {
 182     HQDN3DContext *hqdn3d = ctx->priv;
 183     double lum_spac, lum_tmp, chrom_spac, chrom_tmp;
 184     double param1, param2, param3, param4;
 185
 186     lum_spac   = PARAM1_DEFAULT;
 187     chrom_spac = PARAM2_DEFAULT;
 188     lum_tmp    = PARAM3_DEFAULT;
 189     chrom_tmp  = lum_tmp * chrom_spac / lum_spac;
 190
 191     if (args) {
 192         switch (sscanf(args, "%lf:%lf:%lf:%lf",
 193                        &param1, &param2, &param3, &param4)) {
 194         case 1:
 195             lum_spac   = param1;
 196             chrom_spac = PARAM2_DEFAULT * param1 / PARAM1_DEFAULT;
 197             lum_tmp    = PARAM3_DEFAULT * param1 / PARAM1_DEFAULT;
 198             chrom_tmp  = lum_tmp * chrom_spac / lum_spac;
 199             break;
 200         case 2:
 201             lum_spac   = param1;
 202             chrom_spac = param2;
 203             lum_tmp    = PARAM3_DEFAULT * param1 / PARAM1_DEFAULT;
 204             chrom_tmp  = lum_tmp * chrom_spac / lum_spac;
 205             break;
 206         case 3:
 207             lum_spac   = param1;
 208             chrom_spac = param2;
 209             lum_tmp    = param3;
 210             chrom_tmp  = lum_tmp * chrom_spac / lum_spac;
 211             break;
 212         case 4:
 213             lum_spac   = param1;
 214             chrom_spac = param2;
 215             lum_tmp    = param3;
 216             chrom_tmp  = param4;
 217             break;
 218         }
 219     }
 220
 221     hqdn3d->strength[0] = lum_spac;
 222     hqdn3d->strength[1] = lum_tmp;
 223     hqdn3d->strength[2] = chrom_spac;
 224     hqdn3d->strength[3] = chrom_tmp;
 225
 226     av_log(ctx, AV_LOG_VERBOSE, "ls:%f cs:%f lt:%f ct:%f\n",
 227            lum_spac, chrom_spac, lum_tmp, chrom_tmp);
 228     if (lum_spac < 0 || chrom_spac < 0 || isnan(chrom_tmp)) {
 229         av_log(ctx, AV_LOG_ERROR,
 230                "Invalid negative value for luma or chroma spatial strength, "
 231                "or resulting value for chroma temporal strength is nan.\n");
 232         return AVERROR(EINVAL);
 233     }
 234
 235     return 0;
 236 }
 237
 238 static void uninit(AVFilterContext *ctx)
 239 {
 240     HQDN3DContext *hqdn3d = ctx->priv;
 241
 242     av_freep(&hqdn3d->coefs[0]);
 243     av_freep(&hqdn3d->coefs[1]);
 244     av_freep(&hqdn3d->coefs[2]);
 245     av_freep(&hqdn3d->coefs[3]);
 246     av_freep(&hqdn3d->line);
 247     av_freep(&hqdn3d->frame_prev[0]);
 248     av_freep(&hqdn3d->frame_prev[1]);
 249     av_freep(&hqdn3d->frame_prev[2]);
 250 }
 251
 252 static int query_formats(AVFilterContext *ctx)
 253 {
 254     static const enum AVPixelFormat pix_fmts[] = {
 255         AV_PIX_FMT_YUV420P,
 256         AV_PIX_FMT_YUV422P,
 257         AV_PIX_FMT_YUV444P,
 258         AV_PIX_FMT_YUV410P,
 259         AV_PIX_FMT_YUV411P,
 260         AV_PIX_FMT_YUV440P,
 261         AV_PIX_FMT_YUVJ420P,
 262         AV_PIX_FMT_YUVJ422P,
 263         AV_PIX_FMT_YUVJ444P,
 264         AV_PIX_FMT_YUVJ440P,
 265         AV_NE( AV_PIX_FMT_YUV420P9BE, AV_PIX_FMT_YUV420P9LE ),
 266         AV_NE( AV_PIX_FMT_YUV422P9BE, AV_PIX_FMT_YUV422P9LE ),
 267         AV_NE( AV_PIX_FMT_YUV444P9BE, AV_PIX_FMT_YUV444P9LE ),
 268         AV_NE( AV_PIX_FMT_YUV420P10BE, AV_PIX_FMT_YUV420P10LE ),
 269         AV_NE( AV_PIX_FMT_YUV422P10BE, AV_PIX_FMT_YUV422P10LE ),
 270         AV_NE( AV_PIX_FMT_YUV444P10BE, AV_PIX_FMT_YUV444P10LE ),
 271         AV_NE( AV_PIX_FMT_YUV420P16BE, AV_PIX_FMT_YUV420P16LE ),
 272         AV_NE( AV_PIX_FMT_YUV422P16BE, AV_PIX_FMT_YUV422P16LE ),
 273         AV_NE( AV_PIX_FMT_YUV444P16BE, AV_PIX_FMT_YUV444P16LE ),
 274         AV_PIX_FMT_NONE
 275     };
 276
 277     ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
 278
 279     return 0;
 280 }
 281
 282 static int config_input(AVFilterLink *inlink)
 283 {
 284     HQDN3DContext *hqdn3d = inlink->dst->priv;
 285     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
 286     int i;
 287
 288     hqdn3d->hsub  = desc->log2_chroma_w;
 289     hqdn3d->vsub  = desc->log2_chroma_h;
 290     hqdn3d->depth = desc->comp[0].depth_minus1+1;
 291
 292     hqdn3d->line = av_malloc(inlink->w * sizeof(*hqdn3d->line));
 293     if (!hqdn3d->line)
 294         return AVERROR(ENOMEM);
 295
 296     for (i = 0; i < 4; i++) {
 297         hqdn3d->coefs[i] = precalc_coefs(hqdn3d->strength[i], hqdn3d->depth);
 298         if (!hqdn3d->coefs[i])
 299             return AVERROR(ENOMEM);
 300     }
 301
 302     if (ARCH_X86)
 303         ff_hqdn3d_init_x86(hqdn3d);
 304
 305     return 0;
 306 }
 307
 308 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
 309 {
 310     HQDN3DContext *hqdn3d = inlink->dst->priv;
 311     AVFilterLink *outlink = inlink->dst->outputs[0];
 312     AVFrame *out;
 313     int direct, c;
 314
 315     if (av_frame_is_writable(in)) {
 316         direct = 1;
 317         out = in;
 318     } else {
 319         out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
 320         if (!out) {
 321             av_frame_free(&in);
 322             return AVERROR(ENOMEM);
 323         }
 324
 325         av_frame_copy_props(out, in);
 326         out->width  = outlink->w;
 327         out->height = outlink->h;
 328     }
 329
 330     for (c = 0; c < 3; c++) {
 331         denoise(hqdn3d, in->data[c], out->data[c],
 332                 hqdn3d->line, &hqdn3d->frame_prev[c],
 333                 in->width  >> (!!c * hqdn3d->hsub),
 334                 in->height >> (!!c * hqdn3d->vsub),
 335                 in->linesize[c], out->linesize[c],
 336                 hqdn3d->coefs[c?2:0], hqdn3d->coefs[c?3:1]);
 337     }
 338
 339     if (!direct)
 340         av_frame_free(&in);
 341
 342     return ff_filter_frame(outlink, out);
 343 }
 344
 345 static const AVFilterPad avfilter_vf_hqdn3d_inputs[] = {
 346     {
 347         .name         = "default",
 348         .type         = AVMEDIA_TYPE_VIDEO,
 349         .config_props = config_input,
 350         .filter_frame = filter_frame,
 351     },
 352     { NULL }
 353 };
 354
 355 static const AVFilterPad avfilter_vf_hqdn3d_outputs[] = {
 356     {
 357         .name = "default",
 358         .type = AVMEDIA_TYPE_VIDEO
 359     },
 360     { NULL }
 361 };
 362
 363 AVFilter avfilter_vf_hqdn3d = {
 364     .name          = "hqdn3d",
 365     .description   = NULL_IF_CONFIG_SMALL("Apply a High Quality 3D Denoiser."),
 366
 367     .priv_size     = sizeof(HQDN3DContext),
 368     .init          = init,
 369     .uninit        = uninit,
 370     .query_formats = query_formats,
 371
 372     .inputs    = avfilter_vf_hqdn3d_inputs,
 373
 374     .outputs   = avfilter_vf_hqdn3d_outputs,
 375 };