yading@10: /*
yading@10:  * Copyright (c) 2003 Daniel Moreno <comac AT comac DOT darktech DOT org>
yading@10:  * Copyright (c) 2010 Baptiste Coudurier
yading@10:  * Copyright (c) 2012 Loren Merritt
yading@10:  *
yading@10:  * This file is part of FFmpeg, ported from MPlayer.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or modify
yading@10:  * it under the terms of the GNU General Public License as published by
yading@10:  * the Free Software Foundation; either version 2 of the License, or
yading@10:  * (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
yading@10:  * GNU General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU General Public License along
yading@10:  * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
yading@10:  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * high quality 3d video denoiser, ported from MPlayer
yading@10:  * libmpcodecs/vf_hqdn3d.c.
yading@10:  */
yading@10: 
yading@10: #include <float.h>
yading@10: 
yading@10: #include "config.h"
yading@10: #include "libavutil/common.h"
yading@10: #include "libavutil/pixdesc.h"
yading@10: #include "libavutil/intreadwrite.h"
yading@10: #include "libavutil/opt.h"
yading@10: 
yading@10: #include "avfilter.h"
yading@10: #include "formats.h"
yading@10: #include "internal.h"
yading@10: #include "video.h"
yading@10: #include "vf_hqdn3d.h"
yading@10: 
yading@10: #define LUT_BITS (depth==16 ? 8 : 4)
yading@10: #define LOAD(x) (((depth == 8 ? src[x] : AV_RN16A(src + (x) * 2)) << (16 - depth))\
yading@10:                  + (((1 << (16 - depth)) - 1) >> 1))
yading@10: #define STORE(x,val) (depth == 8 ? dst[x] = (val) >> (16 - depth) : \
yading@10:                                    AV_WN16A(dst + (x) * 2, (val) >> (16 - depth)))
yading@10: 
yading@10: av_always_inline
yading@10: static uint32_t lowpass(int prev, int cur, int16_t *coef, int depth)
yading@10: {
yading@10:     int d = (prev - cur) >> (8 - LUT_BITS);
yading@10:     return cur + coef[d];
yading@10: }
yading@10: 
yading@10: av_always_inline
yading@10: static void denoise_temporal(uint8_t *src, uint8_t *dst,
yading@10:                              uint16_t *frame_ant,
yading@10:                              int w, int h, int sstride, int dstride,
yading@10:                              int16_t *temporal, int depth)
yading@10: {
yading@10:     long x, y;
yading@10:     uint32_t tmp;
yading@10: 
yading@10:     temporal += 256 << LUT_BITS;
yading@10: 
yading@10:     for (y = 0; y < h; y++) {
yading@10:         for (x = 0; x < w; x++) {
yading@10:             frame_ant[x] = tmp = lowpass(frame_ant[x], LOAD(x), temporal, depth);
yading@10:             STORE(x, tmp);
yading@10:         }
yading@10:         src += sstride;
yading@10:         dst += dstride;
yading@10:         frame_ant += w;
yading@10:     }
yading@10: }
yading@10: 
yading@10: av_always_inline
yading@10: static void denoise_spatial(HQDN3DContext *hqdn3d,
yading@10:                             uint8_t *src, uint8_t *dst,
yading@10:                             uint16_t *line_ant, uint16_t *frame_ant,
yading@10:                             int w, int h, int sstride, int dstride,
yading@10:                             int16_t *spatial, int16_t *temporal, int depth)
yading@10: {
yading@10:     long x, y;
yading@10:     uint32_t pixel_ant;
yading@10:     uint32_t tmp;
yading@10: 
yading@10:     spatial  += 256 << LUT_BITS;
yading@10:     temporal += 256 << LUT_BITS;
yading@10: 
yading@10:     /* First line has no top neighbor. Only left one for each tmp and
yading@10:      * last frame */
yading@10:     pixel_ant = LOAD(0);
yading@10:     for (x = 0; x < w; x++) {
yading@10:         line_ant[x] = tmp = pixel_ant = lowpass(pixel_ant, LOAD(x), spatial, depth);
yading@10:         frame_ant[x] = tmp = lowpass(frame_ant[x], tmp, temporal, depth);
yading@10:         STORE(x, tmp);
yading@10:     }
yading@10: 
yading@10:     for (y = 1; y < h; y++) {
yading@10:         src += sstride;
yading@10:         dst += dstride;
yading@10:         frame_ant += w;
yading@10:         if (hqdn3d->denoise_row[depth]) {
yading@10:             hqdn3d->denoise_row[depth](src, dst, line_ant, frame_ant, w, spatial, temporal);
yading@10:             continue;
yading@10:         }
yading@10:         pixel_ant = LOAD(0);
yading@10:         for (x = 0; x < w-1; x++) {
yading@10:             line_ant[x] = tmp = lowpass(line_ant[x], pixel_ant, spatial, depth);
yading@10:             pixel_ant = lowpass(pixel_ant, LOAD(x+1), spatial, depth);
yading@10:             frame_ant[x] = tmp = lowpass(frame_ant[x], tmp, temporal, depth);
yading@10:             STORE(x, tmp);
yading@10:         }
yading@10:         line_ant[x] = tmp = lowpass(line_ant[x], pixel_ant, spatial, depth);
yading@10:         frame_ant[x] = tmp = lowpass(frame_ant[x], tmp, temporal, depth);
yading@10:         STORE(x, tmp);
yading@10:     }
yading@10: }
yading@10: 
yading@10: av_always_inline
yading@10: static void denoise_depth(HQDN3DContext *hqdn3d,
yading@10:                           uint8_t *src, uint8_t *dst,
yading@10:                           uint16_t *line_ant, uint16_t **frame_ant_ptr,
yading@10:                           int w, int h, int sstride, int dstride,
yading@10:                           int16_t *spatial, int16_t *temporal, int depth)
yading@10: {
yading@10:     // FIXME: For 16bit depth, frame_ant could be a pointer to the previous
yading@10:     // filtered frame rather than a separate buffer.
yading@10:     long x, y;
yading@10:     uint16_t *frame_ant = *frame_ant_ptr;
yading@10:     if (!frame_ant) {
yading@10:         uint8_t *frame_src = src;
yading@10:         *frame_ant_ptr = frame_ant = av_malloc(w*h*sizeof(uint16_t));
yading@10:         for (y = 0; y < h; y++, src += sstride, frame_ant += w)
yading@10:             for (x = 0; x < w; x++)
yading@10:                 frame_ant[x] = LOAD(x);
yading@10:         src = frame_src;
yading@10:         frame_ant = *frame_ant_ptr;
yading@10:     }
yading@10: 
yading@10:     if (spatial[0])
yading@10:         denoise_spatial(hqdn3d, src, dst, line_ant, frame_ant,
yading@10:                         w, h, sstride, dstride, spatial, temporal, depth);
yading@10:     else
yading@10:         denoise_temporal(src, dst, frame_ant,
yading@10:                          w, h, sstride, dstride, temporal, depth);
yading@10: }
yading@10: 
yading@10: #define denoise(...) \
yading@10:     switch (hqdn3d->depth) {\
yading@10:         case  8: denoise_depth(__VA_ARGS__,  8); break;\
yading@10:         case  9: denoise_depth(__VA_ARGS__,  9); break;\
yading@10:         case 10: denoise_depth(__VA_ARGS__, 10); break;\
yading@10:         case 16: denoise_depth(__VA_ARGS__, 16); break;\
yading@10:     }
yading@10: 
yading@10: static int16_t *precalc_coefs(double dist25, int depth)
yading@10: {
yading@10:     int i;
yading@10:     double gamma, simil, C;
yading@10:     int16_t *ct = av_malloc((512<<LUT_BITS)*sizeof(int16_t));
yading@10:     if (!ct)
yading@10:         return NULL;
yading@10: 
yading@10:     gamma = log(0.25) / log(1.0 - FFMIN(dist25,252.0)/255.0 - 0.00001);
yading@10: 
yading@10:     for (i = -255<<LUT_BITS; i <= 255<<LUT_BITS; i++) {
yading@10:         double f = ((i<<(9-LUT_BITS)) + (1<<(8-LUT_BITS)) - 1) / 512.0; // midpoint of the bin
yading@10:         simil = 1.0 - FFABS(f) / 255.0;
yading@10:         C = pow(simil, gamma) * 256.0 * f;
yading@10:         ct[(256<<LUT_BITS)+i] = lrint(C);
yading@10:     }
yading@10: 
yading@10:     ct[0] = !!dist25;
yading@10:     return ct;
yading@10: }
yading@10: 
yading@10: #define PARAM1_DEFAULT 4.0
yading@10: #define PARAM2_DEFAULT 3.0
yading@10: #define PARAM3_DEFAULT 6.0
yading@10: 
yading@10: static int init(AVFilterContext *ctx)
yading@10: {
yading@10:     HQDN3DContext *hqdn3d = ctx->priv;
yading@10: 
yading@10:     if (!hqdn3d->strength[LUMA_SPATIAL])
yading@10:         hqdn3d->strength[LUMA_SPATIAL] = PARAM1_DEFAULT;
yading@10:     if (!hqdn3d->strength[CHROMA_SPATIAL])
yading@10:         hqdn3d->strength[CHROMA_SPATIAL] = PARAM2_DEFAULT * hqdn3d->strength[LUMA_SPATIAL] / PARAM1_DEFAULT;
yading@10:     if (!hqdn3d->strength[LUMA_TMP])
yading@10:         hqdn3d->strength[LUMA_TMP]   = PARAM3_DEFAULT * hqdn3d->strength[LUMA_SPATIAL] / PARAM1_DEFAULT;
yading@10:     if (!hqdn3d->strength[CHROMA_TMP])
yading@10:         hqdn3d->strength[CHROMA_TMP] = hqdn3d->strength[LUMA_TMP] * hqdn3d->strength[CHROMA_SPATIAL] / hqdn3d->strength[LUMA_SPATIAL];
yading@10: 
yading@10:     av_log(ctx, AV_LOG_VERBOSE, "ls:%f cs:%f lt:%f ct:%f\n",
yading@10:            hqdn3d->strength[LUMA_SPATIAL], hqdn3d->strength[CHROMA_SPATIAL],
yading@10:            hqdn3d->strength[LUMA_TMP], hqdn3d->strength[CHROMA_TMP]);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static void uninit(AVFilterContext *ctx)
yading@10: {
yading@10:     HQDN3DContext *hqdn3d = ctx->priv;
yading@10: 
yading@10:     av_freep(&hqdn3d->coefs[0]);
yading@10:     av_freep(&hqdn3d->coefs[1]);
yading@10:     av_freep(&hqdn3d->coefs[2]);
yading@10:     av_freep(&hqdn3d->coefs[3]);
yading@10:     av_freep(&hqdn3d->line);
yading@10:     av_freep(&hqdn3d->frame_prev[0]);
yading@10:     av_freep(&hqdn3d->frame_prev[1]);
yading@10:     av_freep(&hqdn3d->frame_prev[2]);
yading@10: }
yading@10: 
yading@10: static int query_formats(AVFilterContext *ctx)
yading@10: {
yading@10:     static const enum AVPixelFormat pix_fmts[] = {
yading@10:         AV_PIX_FMT_YUV420P,
yading@10:         AV_PIX_FMT_YUV422P,
yading@10:         AV_PIX_FMT_YUV444P,
yading@10:         AV_PIX_FMT_YUV410P,
yading@10:         AV_PIX_FMT_YUV411P,
yading@10:         AV_PIX_FMT_YUV440P,
yading@10:         AV_PIX_FMT_YUVJ420P,
yading@10:         AV_PIX_FMT_YUVJ422P,
yading@10:         AV_PIX_FMT_YUVJ444P,
yading@10:         AV_PIX_FMT_YUVJ440P,
yading@10:         AV_NE( AV_PIX_FMT_YUV420P9BE, AV_PIX_FMT_YUV420P9LE ),
yading@10:         AV_NE( AV_PIX_FMT_YUV422P9BE, AV_PIX_FMT_YUV422P9LE ),
yading@10:         AV_NE( AV_PIX_FMT_YUV444P9BE, AV_PIX_FMT_YUV444P9LE ),
yading@10:         AV_NE( AV_PIX_FMT_YUV420P10BE, AV_PIX_FMT_YUV420P10LE ),
yading@10:         AV_NE( AV_PIX_FMT_YUV422P10BE, AV_PIX_FMT_YUV422P10LE ),
yading@10:         AV_NE( AV_PIX_FMT_YUV444P10BE, AV_PIX_FMT_YUV444P10LE ),
yading@10:         AV_NE( AV_PIX_FMT_YUV420P16BE, AV_PIX_FMT_YUV420P16LE ),
yading@10:         AV_NE( AV_PIX_FMT_YUV422P16BE, AV_PIX_FMT_YUV422P16LE ),
yading@10:         AV_NE( AV_PIX_FMT_YUV444P16BE, AV_PIX_FMT_YUV444P16LE ),
yading@10:         AV_PIX_FMT_NONE
yading@10:     };
yading@10: 
yading@10:     ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static int config_input(AVFilterLink *inlink)
yading@10: {
yading@10:     HQDN3DContext *hqdn3d = inlink->dst->priv;
yading@10:     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
yading@10:     int i;
yading@10: 
yading@10:     hqdn3d->hsub  = desc->log2_chroma_w;
yading@10:     hqdn3d->vsub  = desc->log2_chroma_h;
yading@10:     hqdn3d->depth = desc->comp[0].depth_minus1+1;
yading@10: 
yading@10:     hqdn3d->line = av_malloc(inlink->w * sizeof(*hqdn3d->line));
yading@10:     if (!hqdn3d->line)
yading@10:         return AVERROR(ENOMEM);
yading@10: 
yading@10:     for (i = 0; i < 4; i++) {
yading@10:         hqdn3d->coefs[i] = precalc_coefs(hqdn3d->strength[i], hqdn3d->depth);
yading@10:         if (!hqdn3d->coefs[i])
yading@10:             return AVERROR(ENOMEM);
yading@10:     }
yading@10: 
yading@10:     if (ARCH_X86)
yading@10:         ff_hqdn3d_init_x86(hqdn3d);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static int filter_frame(AVFilterLink *inlink, AVFrame *in)
yading@10: {
yading@10:     HQDN3DContext *hqdn3d = inlink->dst->priv;
yading@10:     AVFilterLink *outlink = inlink->dst->outputs[0];
yading@10: 
yading@10:     AVFrame *out;
yading@10:     int direct, c;
yading@10: 
yading@10:     if (av_frame_is_writable(in)) {
yading@10:         direct = 1;
yading@10:         out = in;
yading@10:     } else {
yading@10:         direct = 0;
yading@10:         out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
yading@10:         if (!out) {
yading@10:             av_frame_free(&in);
yading@10:             return AVERROR(ENOMEM);
yading@10:         }
yading@10: 
yading@10:         av_frame_copy_props(out, in);
yading@10:     }
yading@10: 
yading@10:     for (c = 0; c < 3; c++) {
yading@10:         denoise(hqdn3d, in->data[c], out->data[c],
yading@10:                 hqdn3d->line, &hqdn3d->frame_prev[c],
yading@10:                 in->width  >> (!!c * hqdn3d->hsub),
yading@10:                 in->height >> (!!c * hqdn3d->vsub),
yading@10:                 in->linesize[c], out->linesize[c],
yading@10:                 hqdn3d->coefs[c?2:0], hqdn3d->coefs[c?3:1]);
yading@10:     }
yading@10: 
yading@10:     if (!direct)
yading@10:         av_frame_free(&in);
yading@10: 
yading@10:     return ff_filter_frame(outlink, out);
yading@10: }
yading@10: 
yading@10: #define OFFSET(x) offsetof(HQDN3DContext, x)
yading@10: #define FLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_FILTERING_PARAM
yading@10: static const AVOption options[] = {
yading@10:     { "luma_spatial",   "spatial luma strength",    OFFSET(strength[LUMA_SPATIAL]),   AV_OPT_TYPE_DOUBLE, { .dbl = 0.0 }, 0, DBL_MAX, FLAGS },
yading@10:     { "chroma_spatial", "spatial chroma strength",  OFFSET(strength[CHROMA_SPATIAL]), AV_OPT_TYPE_DOUBLE, { .dbl = 0.0 }, 0, DBL_MAX, FLAGS },
yading@10:     { "luma_tmp",       "temporal luma strength",   OFFSET(strength[LUMA_TMP]),       AV_OPT_TYPE_DOUBLE, { .dbl = 0.0 }, 0, DBL_MAX, FLAGS },
yading@10:     { "chroma_tmp",     "temporal chroma strength", OFFSET(strength[CHROMA_TMP]),     AV_OPT_TYPE_DOUBLE, { .dbl = 0.0 }, 0, DBL_MAX, FLAGS },
yading@10:     { NULL },
yading@10: };
yading@10: 
yading@10: static const AVClass hqdn3d_class = {
yading@10:     .class_name = "hqdn3d",
yading@10:     .item_name  = av_default_item_name,
yading@10:     .option     = options,
yading@10:     .version    = LIBAVUTIL_VERSION_INT,
yading@10: };
yading@10: 
yading@10: static const AVFilterPad avfilter_vf_hqdn3d_inputs[] = {
yading@10:     {
yading@10:         .name         = "default",
yading@10:         .type         = AVMEDIA_TYPE_VIDEO,
yading@10:         .config_props = config_input,
yading@10:         .filter_frame = filter_frame,
yading@10:     },
yading@10:     { NULL }
yading@10: };
yading@10: 
yading@10: 
yading@10: static const AVFilterPad avfilter_vf_hqdn3d_outputs[] = {
yading@10:     {
yading@10:         .name = "default",
yading@10:         .type = AVMEDIA_TYPE_VIDEO
yading@10:     },
yading@10:     { NULL }
yading@10: };
yading@10: 
yading@10: AVFilter avfilter_vf_hqdn3d = {
yading@10:     .name          = "hqdn3d",
yading@10:     .description   = NULL_IF_CONFIG_SMALL("Apply a High Quality 3D Denoiser."),
yading@10: 
yading@10:     .priv_size     = sizeof(HQDN3DContext),
yading@10:     .priv_class    = &hqdn3d_class,
yading@10:     .init          = init,
yading@10:     .uninit        = uninit,
yading@10:     .query_formats = query_formats,
yading@10: 
yading@10:     .inputs    = avfilter_vf_hqdn3d_inputs,
yading@10: 
yading@10:     .outputs   = avfilter_vf_hqdn3d_outputs,
yading@10: };