libmpcodecs/vf_ow.c

   1 /*
   2     Copyright (C) 2007 Michael Niedermayer <michaelni@gmx.at>
   3
   4     This program is free software; you can redistribute it and/or modify
   5     it under the terms of the GNU General Public License as published by
   6     the Free Software Foundation; either version 2 of the License, or
   7     (at your option) any later version.
   8
   9     This program is distributed in the hope that it will be useful,
  10     but WITHOUT ANY WARRANTY; without even the implied warranty of
  11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12     GNU General Public License for more details.
  13
  14     You should have received a copy of the GNU General Public License
  15     along with this program; if not, write to the Free Software
  16     Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17 */
  18
  19 /**
  20  * @todo try to change to int
  21  * @todo try lifting based implementation
  22  * @todo optimize optimize optimize
  23  * @todo hard tresholding
  24  * @todo use QP to decide filter strength
  25  * @todo wavelet normalization / least squares optimal signal vs. noise thresholds
  26  */
  27
  28 #include <stdio.h>
  29 #include <string.h>
  30 #include <inttypes.h>
  31 #include <math.h>
  32
  33 #include "config.h"
  34
  35 #include "mp_msg.h"
  36
  37 #ifdef HAVE_MALLOC_H
  38 #include <malloc.h>
  39 #endif
  40
  41 #include "img_format.h"
  42 #include "mp_image.h"
  43 #include "vf.h"
  44
  45 //===========================================================================//
  46 static const uint8_t  __attribute__((aligned(8))) dither[8][8]={
  47 {  0,  48,  12,  60,   3,  51,  15,  63, },
  48 { 32,  16,  44,  28,  35,  19,  47,  31, },
  49 {  8,  56,   4,  52,  11,  59,   7,  55, },
  50 { 40,  24,  36,  20,  43,  27,  39,  23, },
  51 {  2,  50,  14,  62,   1,  49,  13,  61, },
  52 { 34,  18,  46,  30,  33,  17,  45,  29, },
  53 { 10,  58,   6,  54,   9,  57,   5,  53, },
  54 { 42,  26,  38,  22,  41,  25,  37,  21, },
  55 };
  56 //FIXME the above is duplicated in many filters
  57
  58 struct vf_priv_s {
  59     float strength[2];
  60     float delta;
  61     int mode;
  62     int depth;
  63     float *plane[16][4];
  64     int stride;
  65 };
  66
  67 #define S 1.41421356237 //sqrt(2)
  68
  69 static const double coeff[2][5]={
  70     {
  71          0.6029490182363579  *S,
  72          0.2668641184428723  *S,
  73         -0.07822326652898785 *S,
  74         -0.01686411844287495 *S,
  75          0.02674875741080976 *S
  76     },{
  77          1.115087052456994   /S,
  78         -0.5912717631142470  /S,
  79         -0.05754352622849957 /S,
  80          0.09127176311424948 /S
  81     }
  82 };
  83
  84 static const double icoeff[2][5]={
  85     {
  86          1.115087052456994   /S,
  87          0.5912717631142470  /S,
  88         -0.05754352622849957 /S,
  89         -0.09127176311424948 /S
  90     },{
  91          0.6029490182363579  *S,
  92         -0.2668641184428723  *S,
  93         -0.07822326652898785 *S,
  94          0.01686411844287495 *S,
  95          0.02674875741080976 *S
  96     }
  97 };
  98 #undef S
  99
 100 static inline int mirror(int x, int w){
 101     while((unsigned)x > (unsigned)w){
 102         x=-x;
 103         if(x<0) x+= 2*w;
 104     }
 105     return x;
 106 }
 107
 108 static inline void decompose(float *dstL, float *dstH, float *src, int stride, int w){
 109     int x, i;
 110     for(x=0; x<w; x++){
 111         double sumL= src[x*stride] * coeff[0][0];
 112         double sumH= src[x*stride] * coeff[1][0];
 113         for(i=1; i<=4; i++){
 114             double s= (src[mirror(x-i, w-1)*stride] + src[mirror(x+i, w-1)*stride]);
 115
 116             sumL+= coeff[0][i]*s;
 117             sumH+= coeff[1][i]*s;
 118         }
 119         dstL[x*stride]= sumL;
 120         dstH[x*stride]= sumH;
 121     }
 122 }
 123
 124 static inline void compose(float *dst, float *srcL, float *srcH, int stride, int w){
 125     int x, i;
 126     for(x=0; x<w; x++){
 127         double sumL= srcL[x*stride] * icoeff[0][0];
 128         double sumH= srcH[x*stride] * icoeff[1][0];
 129         for(i=1; i<=4; i++){
 130             int x0= mirror(x-i, w-1)*stride;
 131             int x1= mirror(x+i, w-1)*stride;
 132
 133             sumL+= icoeff[0][i]*(srcL[x0] + srcL[x1]);
 134             sumH+= icoeff[1][i]*(srcH[x0] + srcH[x1]);
 135         }
 136         dst[x*stride]= (sumL + sumH)*0.5;
 137     }
 138 }
 139
 140 static inline void decompose2D(float *dstL, float *dstH, float *src, int xstride, int ystride, int step, int w, int h){
 141     int y, x;
 142     for(y=0; y<h; y++)
 143         for(x=0; x<step; x++)
 144             decompose(dstL + ystride*y + xstride*x, dstH + ystride*y + xstride*x, src + ystride*y + xstride*x, step*xstride, (w-x+step-1)/step);
 145 }
 146
 147 static inline void compose2D(float *dst, float *srcL, float *srcH, int xstride, int ystride, int step, int w, int h){
 148     int y, x;
 149     for(y=0; y<h; y++)
 150         for(x=0; x<step; x++)
 151             compose(dst + ystride*y + xstride*x, srcL + ystride*y + xstride*x, srcH + ystride*y + xstride*x, step*xstride, (w-x+step-1)/step);
 152 }
 153
 154 static void decompose2D2(float *dst[4], float *src, float *temp[2], int stride, int step, int w, int h){
 155     decompose2D(temp[0], temp[1], src    , 1, stride, step  , w, h);
 156     decompose2D( dst[0],  dst[1], temp[0], stride, 1, step  , h, w);
 157     decompose2D( dst[2],  dst[3], temp[1], stride, 1, step  , h, w);
 158 }
 159
 160 static void compose2D2(float *dst, float *src[4], float *temp[2], int stride, int step, int w, int h){
 161     compose2D(temp[0],  src[0],  src[1], stride, 1, step  , h, w);
 162     compose2D(temp[1],  src[2],  src[3], stride, 1, step  , h, w);
 163     compose2D(dst    , temp[0], temp[1], 1, stride, step  , w, h);
 164 }
 165
 166 static void filter(struct vf_priv_s *p, uint8_t *dst, uint8_t *src, int dst_stride, int src_stride, int width, int height, int is_luma){
 167     int x,y, i, j;
 168 //    double sum=0;
 169     double s= p->strength[!is_luma];
 170     int depth= p->depth;
 171
 172     while(1<<depth > width || 1<<depth > height)
 173         depth--;
 174
 175     for(y=0; y<height; y++)
 176         for(x=0; x<width; x++)
 177             p->plane[0][0][x + y*p->stride]= src[x + y*src_stride];
 178
 179     for(i=0; i<depth; i++){
 180         decompose2D2(p->plane[i+1], p->plane[i][0], p->plane[0]+1,p->stride, 1<<i, width, height);
 181     }
 182     for(i=0; i<depth; i++){
 183         for(j=1; j<4; j++){
 184             for(y=0; y<height; y++){
 185                 for(x=0; x<width; x++){
 186                     double v= p->plane[i+1][j][x + y*p->stride];
 187                     if     (v> s) v-=s;
 188                     else if(v<-s) v+=s;
 189                     else          v =0;
 190                     p->plane[i+1][j][x + y*p->stride]= v;
 191                 }
 192             }
 193         }
 194     }
 195     for(i=depth-1; i>=0; i--){
 196         compose2D2(p->plane[i][0], p->plane[i+1], p->plane[0]+1, p->stride, 1<<i, width, height);
 197     }
 198
 199     for(y=0; y<height; y++)
 200         for(x=0; x<width; x++){
 201             i= p->plane[0][0][x + y*p->stride] + dither[x&7][y&7]*(1.0/64) + 1.0/128; //yes the rounding is insane but optimal :)
 202 //            double e= i - src[x + y*src_stride];
 203 //            sum += e*e;
 204             if((unsigned)i > 255U) i= ~(i>>31);
 205             dst[x + y*dst_stride]= i;
 206         }
 207
 208 //    printf("%f\n", sum/height/width);
 209 }
 210
 211 static int config(struct vf_instance_s* vf, int width, int height, int d_width, int d_height, unsigned int flags, unsigned int outfmt){
 212     int h= (height+15)&(~15);
 213     int i,j;
 214
 215     vf->priv->stride= (width+15)&(~15);
 216     for(j=0; j<4; j++){
 217         for(i=0; i<=vf->priv->depth; i++)
 218             vf->priv->plane[i][j]= malloc(vf->priv->stride*h*sizeof(vf->priv->plane[0][0][0]));
 219     }
 220
 221     return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
 222 }
 223
 224 static void get_image(struct vf_instance_s* vf, mp_image_t *mpi){
 225     if(mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change
 226     // ok, we can do pp in-place (or pp disabled):
 227     vf->dmpi=vf_get_image(vf->next,mpi->imgfmt,
 228         mpi->type, mpi->flags | MP_IMGFLAG_READABLE, mpi->width, mpi->height);
 229     mpi->planes[0]=vf->dmpi->planes[0];
 230     mpi->stride[0]=vf->dmpi->stride[0];
 231     mpi->width=vf->dmpi->width;
 232     if(mpi->flags&MP_IMGFLAG_PLANAR){
 233         mpi->planes[1]=vf->dmpi->planes[1];
 234         mpi->planes[2]=vf->dmpi->planes[2];
 235         mpi->stride[1]=vf->dmpi->stride[1];
 236         mpi->stride[2]=vf->dmpi->stride[2];
 237     }
 238     mpi->flags|=MP_IMGFLAG_DIRECT;
 239 }
 240
 241 static int put_image(struct vf_instance_s* vf, mp_image_t *mpi, double pts){
 242     mp_image_t *dmpi;
 243
 244     if(!(mpi->flags&MP_IMGFLAG_DIRECT)){
 245         // no DR, so get a new image! hope we'll get DR buffer:
 246         dmpi=vf_get_image(vf->next,mpi->imgfmt,
 247             MP_IMGTYPE_TEMP,
 248             MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
 249             mpi->width,mpi->height);
 250         vf_clone_mpi_attributes(dmpi, mpi);
 251     }else{
 252         dmpi=vf->dmpi;
 253     }
 254
 255     filter(vf->priv, dmpi->planes[0], mpi->planes[0], dmpi->stride[0], mpi->stride[0], mpi->w, mpi->h, 1);
 256     filter(vf->priv, dmpi->planes[1], mpi->planes[1], dmpi->stride[1], mpi->stride[1], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, 0);
 257     filter(vf->priv, dmpi->planes[2], mpi->planes[2], dmpi->stride[2], mpi->stride[2], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, 0);
 258
 259     return vf_next_put_image(vf,dmpi, pts);
 260 }
 261
 262 static void uninit(struct vf_instance_s* vf){
 263     int i,j;
 264     if(!vf->priv) return;
 265
 266     for(j=0; j<4; j++){
 267         for(i=0; i<16; i++){
 268             free(vf->priv->plane[i][j]);
 269             vf->priv->plane[i][j]= NULL;
 270         }
 271     }
 272
 273     free(vf->priv);
 274     vf->priv=NULL;
 275 }
 276
 277 //===========================================================================//
 278 static int query_format(struct vf_instance_s* vf, unsigned int fmt){
 279     switch(fmt){
 280         case IMGFMT_YVU9:
 281         case IMGFMT_IF09:
 282         case IMGFMT_YV12:
 283         case IMGFMT_I420:
 284         case IMGFMT_IYUV:
 285         case IMGFMT_CLPL:
 286         case IMGFMT_Y800:
 287         case IMGFMT_Y8:
 288         case IMGFMT_444P:
 289         case IMGFMT_422P:
 290         case IMGFMT_411P:
 291             return vf_next_query_format(vf,fmt);
 292     }
 293     return 0;
 294 }
 295
 296
 297 static int open(vf_instance_t *vf, char* args){
 298     vf->config=config;
 299     vf->put_image=put_image;
 300     vf->get_image=get_image;
 301     vf->query_format=query_format;
 302     vf->uninit=uninit;
 303     vf->priv=malloc(sizeof(struct vf_priv_s));
 304     memset(vf->priv, 0, sizeof(struct vf_priv_s));
 305
 306     vf->priv->depth= 8;
 307     vf->priv->strength[0]= 1.0;
 308     vf->priv->strength[1]= 1.0;
 309     vf->priv->delta= 1.0;
 310
 311     if (args) sscanf(args, "%d:%f:%f:%d:%f", &vf->priv->depth,
 312                      &vf->priv->strength[0],
 313                      &vf->priv->strength[1],
 314                      &vf->priv->mode,
 315                      &vf->priv->delta);
 316
 317     return 1;
 318 }
 319
 320 const vf_info_t vf_info_ow = {
 321     "overcomplete wavelet denoiser",
 322     "ow",
 323     "Michael Niedermayer",
 324     "",
 325     open,
 326     NULL
 327 };