FFmpeg: libavfilter/vf_hqdn3d.c Source File

00001 /*
00002  * Copyright (c) 2003 Daniel Moreno <comac AT comac DOT darktech DOT org>
00003  * Copyright (c) 2010 Baptiste Coudurier
00004  * Copyright (c) 2012 Loren Merritt
00005  *
00006  * This file is part of FFmpeg, ported from MPlayer.
00007  *
00008  * FFmpeg is free software; you can redistribute it and/or modify
00009  * it under the terms of the GNU General Public License as published by
00010  * the Free Software Foundation; either version 2 of the License, or
00011  * (at your option) any later version.
00012  *
00013  * FFmpeg is distributed in the hope that it will be useful,
00014  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00015  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
00016  * GNU General Public License for more details.
00017  *
00018  * You should have received a copy of the GNU General Public License along
00019  * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
00020  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
00021  */
00022 
00029 #include "libavutil/common.h"
00030 #include "libavutil/pixdesc.h"
00031 #include "libavutil/intreadwrite.h"
00032 #include "avfilter.h"
00033 #include "formats.h"
00034 #include "internal.h"
00035 #include "video.h"
00036 
00037 typedef struct {
00038     int16_t *coefs[4];
00039     uint16_t *line;
00040     uint16_t *frame_prev[3];
00041     double strength[4];
00042     int hsub, vsub;
00043     int depth;
00044     void (*denoise_row[17])(uint8_t *src, uint8_t *dst, uint16_t *line_ant, uint16_t *frame_ant, ptrdiff_t w, int16_t *spatial, int16_t *temporal);
00045 } HQDN3DContext;
00046 
00047 void ff_hqdn3d_row_8_x86(uint8_t *src, uint8_t *dst, uint16_t *line_ant, uint16_t *frame_ant, ptrdiff_t w, int16_t *spatial, int16_t *temporal);
00048 void ff_hqdn3d_row_9_x86(uint8_t *src, uint8_t *dst, uint16_t *line_ant, uint16_t *frame_ant, ptrdiff_t w, int16_t *spatial, int16_t *temporal);
00049 void ff_hqdn3d_row_10_x86(uint8_t *src, uint8_t *dst, uint16_t *line_ant, uint16_t *frame_ant, ptrdiff_t w, int16_t *spatial, int16_t *temporal);
00050 void ff_hqdn3d_row_16_x86(uint8_t *src, uint8_t *dst, uint16_t *line_ant, uint16_t *frame_ant, ptrdiff_t w, int16_t *spatial, int16_t *temporal);
00051 
00052 #define LUT_BITS (depth==16 ? 8 : 4)
00053 #define LOAD(x) (((depth==8 ? src[x] : AV_RN16A(src+(x)*2)) << (16-depth)) + (((1<<(16-depth))-1)>>1))
00054 #define STORE(x,val) (depth==8 ? dst[x] = (val) >> (16-depth)\
00055                     : AV_WN16A(dst+(x)*2, (val) >> (16-depth)))
00056 
00057 av_always_inline
00058 static uint32_t lowpass(int prev, int cur, int16_t *coef, int depth)
00059 {
00060     int d = (prev - cur) >> (8 - LUT_BITS);
00061     return cur + coef[d];
00062 }
00063 
00064 av_always_inline
00065 static void denoise_temporal(uint8_t *src, uint8_t *dst,
00066                              uint16_t *frame_ant,
00067                              int w, int h, int sstride, int dstride,
00068                              int16_t *temporal, int depth)
00069 {
00070     long x, y;
00071     uint32_t tmp;
00072 
00073     temporal += 256 << LUT_BITS;
00074 
00075     for (y = 0; y < h; y++) {
00076         for (x = 0; x < w; x++) {
00077             frame_ant[x] = tmp = lowpass(frame_ant[x], LOAD(x), temporal, depth);
00078             STORE(x, tmp);
00079         }
00080         src += sstride;
00081         dst += dstride;
00082         frame_ant += w;
00083     }
00084 }
00085 
00086 av_always_inline
00087 static void denoise_spatial(HQDN3DContext *hqdn3d,
00088                             uint8_t *src, uint8_t *dst,
00089                             uint16_t *line_ant, uint16_t *frame_ant,
00090                             int w, int h, int sstride, int dstride,
00091                             int16_t *spatial, int16_t *temporal, int depth)
00092 {
00093     long x, y;
00094     uint32_t pixel_ant;
00095     uint32_t tmp;
00096 
00097     spatial  += 256 << LUT_BITS;
00098     temporal += 256 << LUT_BITS;
00099 
00100     /* First line has no top neighbor. Only left one for each tmp and
00101      * last frame */
00102     pixel_ant = LOAD(0);
00103     for (x = 0; x < w; x++) {
00104         line_ant[x] = tmp = pixel_ant = lowpass(pixel_ant, LOAD(x), spatial, depth);
00105         frame_ant[x] = tmp = lowpass(frame_ant[x], tmp, temporal, depth);
00106         STORE(x, tmp);
00107     }
00108 
00109     for (y = 1; y < h; y++) {
00110         src += sstride;
00111         dst += dstride;
00112         frame_ant += w;
00113         if (hqdn3d->denoise_row[depth]) {
00114             hqdn3d->denoise_row[depth](src, dst, line_ant, frame_ant, w, spatial, temporal);
00115             continue;
00116         }
00117         pixel_ant = LOAD(0);
00118         for (x = 0; x < w-1; x++) {
00119             line_ant[x] = tmp = lowpass(line_ant[x], pixel_ant, spatial, depth);
00120             pixel_ant = lowpass(pixel_ant, LOAD(x+1), spatial, depth);
00121             frame_ant[x] = tmp = lowpass(frame_ant[x], tmp, temporal, depth);
00122             STORE(x, tmp);
00123         }
00124         line_ant[x] = tmp = lowpass(line_ant[x], pixel_ant, spatial, depth);
00125         frame_ant[x] = tmp = lowpass(frame_ant[x], tmp, temporal, depth);
00126         STORE(x, tmp);
00127     }
00128 }
00129 
00130 av_always_inline
00131 static void denoise_depth(HQDN3DContext *hqdn3d,
00132                           uint8_t *src, uint8_t *dst,
00133                           uint16_t *line_ant, uint16_t **frame_ant_ptr,
00134                           int w, int h, int sstride, int dstride,
00135                           int16_t *spatial, int16_t *temporal, int depth)
00136 {
00137     // FIXME: For 16bit depth, frame_ant could be a pointer to the previous
00138     // filtered frame rather than a separate buffer.
00139     long x, y;
00140     uint16_t *frame_ant = *frame_ant_ptr;
00141     if (!frame_ant) {
00142         uint8_t *frame_src = src;
00143         *frame_ant_ptr = frame_ant = av_malloc(w*h*sizeof(uint16_t));
00144         for (y = 0; y < h; y++, src += sstride, frame_ant += w)
00145             for (x = 0; x < w; x++)
00146                 frame_ant[x] = LOAD(x);
00147         src = frame_src;
00148         frame_ant = *frame_ant_ptr;
00149     }
00150 
00151     if (spatial[0])
00152         denoise_spatial(hqdn3d, src, dst, line_ant, frame_ant,
00153                         w, h, sstride, dstride, spatial, temporal, depth);
00154     else
00155         denoise_temporal(src, dst, frame_ant,
00156                          w, h, sstride, dstride, temporal, depth);
00157 }
00158 
00159 #define denoise(...) \
00160     switch (hqdn3d->depth) {\
00161         case  8: denoise_depth(__VA_ARGS__,  8); break;\
00162         case  9: denoise_depth(__VA_ARGS__,  9); break;\
00163         case 10: denoise_depth(__VA_ARGS__, 10); break;\
00164         case 16: denoise_depth(__VA_ARGS__, 16); break;\
00165     }
00166 
00167 static int16_t *precalc_coefs(double dist25, int depth)
00168 {
00169     int i;
00170     double gamma, simil, C;
00171     int16_t *ct = av_malloc((512<<LUT_BITS)*sizeof(int16_t));
00172     if (!ct)
00173         return NULL;
00174 
00175     gamma = log(0.25) / log(1.0 - FFMIN(dist25,252.0)/255.0 - 0.00001);
00176 
00177     for (i = -255<<LUT_BITS; i <= 255<<LUT_BITS; i++) {
00178         double f = ((i<<(9-LUT_BITS)) + (1<<(8-LUT_BITS)) - 1) / 512.0; // midpoint of the bin
00179         simil = 1.0 - FFABS(f) / 255.0;
00180         C = pow(simil, gamma) * 256.0 * f;
00181         ct[(256<<LUT_BITS)+i] = lrint(C);
00182     }
00183 
00184     ct[0] = !!dist25;
00185     return ct;
00186 }
00187 
00188 #define PARAM1_DEFAULT 4.0
00189 #define PARAM2_DEFAULT 3.0
00190 #define PARAM3_DEFAULT 6.0
00191 
00192 static int init(AVFilterContext *ctx, const char *args)
00193 {
00194     HQDN3DContext *hqdn3d = ctx->priv;
00195     double lum_spac, lum_tmp, chrom_spac, chrom_tmp;
00196     double param1, param2, param3, param4;
00197 
00198     lum_spac   = PARAM1_DEFAULT;
00199     chrom_spac = PARAM2_DEFAULT;
00200     lum_tmp    = PARAM3_DEFAULT;
00201     chrom_tmp  = lum_tmp * chrom_spac / lum_spac;
00202 
00203     if (args) {
00204         switch (sscanf(args, "%lf:%lf:%lf:%lf",
00205                        &param1, &param2, &param3, &param4)) {
00206         case 1:
00207             lum_spac   = param1;
00208             chrom_spac = PARAM2_DEFAULT * param1 / PARAM1_DEFAULT;
00209             lum_tmp    = PARAM3_DEFAULT * param1 / PARAM1_DEFAULT;
00210             chrom_tmp  = lum_tmp * chrom_spac / lum_spac;
00211             break;
00212         case 2:
00213             lum_spac   = param1;
00214             chrom_spac = param2;
00215             lum_tmp    = PARAM3_DEFAULT * param1 / PARAM1_DEFAULT;
00216             chrom_tmp  = lum_tmp * chrom_spac / lum_spac;
00217             break;
00218         case 3:
00219             lum_spac   = param1;
00220             chrom_spac = param2;
00221             lum_tmp    = param3;
00222             chrom_tmp  = lum_tmp * chrom_spac / lum_spac;
00223             break;
00224         case 4:
00225             lum_spac   = param1;
00226             chrom_spac = param2;
00227             lum_tmp    = param3;
00228             chrom_tmp  = param4;
00229             break;
00230         }
00231     }
00232 
00233     hqdn3d->strength[0] = lum_spac;
00234     hqdn3d->strength[1] = lum_tmp;
00235     hqdn3d->strength[2] = chrom_spac;
00236     hqdn3d->strength[3] = chrom_tmp;
00237 
00238     av_log(ctx, AV_LOG_VERBOSE, "ls:%lf cs:%lf lt:%lf ct:%lf\n",
00239            lum_spac, chrom_spac, lum_tmp, chrom_tmp);
00240     if (lum_spac < 0 || chrom_spac < 0 || isnan(chrom_tmp)) {
00241         av_log(ctx, AV_LOG_ERROR,
00242                "Invalid negative value for luma or chroma spatial strength, "
00243                "or resulting value for chroma temporal strength is nan.\n");
00244         return AVERROR(EINVAL);
00245     }
00246 
00247     return 0;
00248 }
00249 
00250 static void uninit(AVFilterContext *ctx)
00251 {
00252     HQDN3DContext *hqdn3d = ctx->priv;
00253 
00254     av_freep(&hqdn3d->coefs[0]);
00255     av_freep(&hqdn3d->coefs[1]);
00256     av_freep(&hqdn3d->coefs[2]);
00257     av_freep(&hqdn3d->coefs[3]);
00258     av_freep(&hqdn3d->line);
00259     av_freep(&hqdn3d->frame_prev[0]);
00260     av_freep(&hqdn3d->frame_prev[1]);
00261     av_freep(&hqdn3d->frame_prev[2]);
00262 }
00263 
00264 static int query_formats(AVFilterContext *ctx)
00265 {
00266     static const enum PixelFormat pix_fmts[] = {
00267         PIX_FMT_YUV420P,
00268         PIX_FMT_YUV422P,
00269         PIX_FMT_YUV444P,
00270         PIX_FMT_YUV410P,
00271         PIX_FMT_YUV411P,
00272         PIX_FMT_YUV440P,
00273         PIX_FMT_YUVJ420P,
00274         PIX_FMT_YUVJ422P,
00275         PIX_FMT_YUVJ444P,
00276         PIX_FMT_YUVJ440P,
00277         AV_NE( PIX_FMT_YUV420P9BE, PIX_FMT_YUV420P9LE ),
00278         AV_NE( PIX_FMT_YUV422P9BE, PIX_FMT_YUV422P9LE ),
00279         AV_NE( PIX_FMT_YUV444P9BE, PIX_FMT_YUV444P9LE ),
00280         AV_NE( PIX_FMT_YUV420P10BE, PIX_FMT_YUV420P10LE ),
00281         AV_NE( PIX_FMT_YUV422P10BE, PIX_FMT_YUV422P10LE ),
00282         AV_NE( PIX_FMT_YUV444P10BE, PIX_FMT_YUV444P10LE ),
00283         AV_NE( PIX_FMT_YUV420P16BE, PIX_FMT_YUV420P16LE ),
00284         AV_NE( PIX_FMT_YUV422P16BE, PIX_FMT_YUV422P16LE ),
00285         AV_NE( PIX_FMT_YUV444P16BE, PIX_FMT_YUV444P16LE ),
00286         PIX_FMT_NONE
00287     };
00288 
00289     ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
00290 
00291     return 0;
00292 }
00293 
00294 static int config_input(AVFilterLink *inlink)
00295 {
00296     HQDN3DContext *hqdn3d = inlink->dst->priv;
00297     int i;
00298 
00299     hqdn3d->hsub = av_pix_fmt_descriptors[inlink->format].log2_chroma_w;
00300     hqdn3d->vsub = av_pix_fmt_descriptors[inlink->format].log2_chroma_h;
00301     hqdn3d->depth = av_pix_fmt_descriptors[inlink->format].comp[0].depth_minus1+1;
00302 
00303     hqdn3d->line = av_malloc(inlink->w * sizeof(*hqdn3d->line));
00304     if (!hqdn3d->line)
00305         return AVERROR(ENOMEM);
00306 
00307     for (i = 0; i < 4; i++) {
00308         hqdn3d->coefs[i] = precalc_coefs(hqdn3d->strength[i], hqdn3d->depth);
00309         if (!hqdn3d->coefs[i])
00310             return AVERROR(ENOMEM);
00311     }
00312 
00313 #if HAVE_YASM
00314     hqdn3d->denoise_row[ 8] = ff_hqdn3d_row_8_x86;
00315     hqdn3d->denoise_row[ 9] = ff_hqdn3d_row_9_x86;
00316     hqdn3d->denoise_row[10] = ff_hqdn3d_row_10_x86;
00317     hqdn3d->denoise_row[16] = ff_hqdn3d_row_16_x86;
00318 #endif
00319 
00320     return 0;
00321 }
00322 
00323 static int null_draw_slice(AVFilterLink *link, int y, int h, int slice_dir)
00324 {
00325     return 0;
00326 }
00327 
00328 static int end_frame(AVFilterLink *inlink)
00329 {
00330     HQDN3DContext *hqdn3d = inlink->dst->priv;
00331     AVFilterLink *outlink = inlink->dst->outputs[0];
00332     AVFilterBufferRef *inpic  = inlink ->cur_buf;
00333     AVFilterBufferRef *outpic = outlink->out_buf;
00334     int ret, c;
00335 
00336     for (c = 0; c < 3; c++) {
00337         denoise(hqdn3d, inpic->data[c], outpic->data[c],
00338                 hqdn3d->line, &hqdn3d->frame_prev[c],
00339                 inpic->video->w >> (!!c * hqdn3d->hsub),
00340                 inpic->video->h >> (!!c * hqdn3d->vsub),
00341                 inpic->linesize[c], outpic->linesize[c],
00342                 hqdn3d->coefs[c?2:0], hqdn3d->coefs[c?3:1]);
00343     }
00344 
00345     if ((ret = ff_draw_slice(outlink, 0, inpic->video->h, 1)) < 0 ||
00346         (ret = ff_end_frame(outlink)) < 0)
00347         return ret;
00348     return 0;
00349 }
00350 
00351 AVFilter avfilter_vf_hqdn3d = {
00352     .name          = "hqdn3d",
00353     .description   = NULL_IF_CONFIG_SMALL("Apply a High Quality 3D Denoiser."),
00354 
00355     .priv_size     = sizeof(HQDN3DContext),
00356     .init          = init,
00357     .uninit        = uninit,
00358     .query_formats = query_formats,
00359 
00360     .inputs    = (const AVFilterPad[]) {{ .name             = "default",
00361                                           .type             = AVMEDIA_TYPE_VIDEO,
00362                                           .start_frame      = ff_inplace_start_frame,
00363                                           .draw_slice       = null_draw_slice,
00364                                           .config_props     = config_input,
00365                                           .end_frame        = end_frame },
00366                                         { .name = NULL}},
00367 
00368     .outputs   = (const AVFilterPad[]) {{ .name             = "default",
00369                                           .type             = AVMEDIA_TYPE_VIDEO },
00370                                         { .name = NULL}},
00371 };