2 * Copyright (C) 2007 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of MPlayer.
6 * MPlayer is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * MPlayer is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License along
17 * with MPlayer; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
22 * @todo try to change to int
23 * @todo try lifting based implementation
24 * @todo optimize optimize optimize
25 * @todo hard tresholding
26 * @todo use QP to decide filter strength
27 * @todo wavelet normalization / least squares optimal signal vs. noise thresholds
36 #include "img_format.h"
40 //===========================================================================//
41 static const uint8_t __attribute__((aligned(8))) dither[8][8]={
42 { 0, 48, 12, 60, 3, 51, 15, 63, },
43 { 32, 16, 44, 28, 35, 19, 47, 31, },
44 { 8, 56, 4, 52, 11, 59, 7, 55, },
45 { 40, 24, 36, 20, 43, 27, 39, 23, },
46 { 2, 50, 14, 62, 1, 49, 13, 61, },
47 { 34, 18, 46, 30, 33, 17, 45, 29, },
48 { 10, 58, 6, 54, 9, 57, 5, 53, },
49 { 42, 26, 38, 22, 41, 25, 37, 21, },
51 //FIXME the above is duplicated in many filters
62 #define S 1.41421356237 //sqrt(2)
64 static const double coeff[2][5]={
66 0.6029490182363579 *S,
67 0.2668641184428723 *S,
68 -0.07822326652898785 *S,
69 -0.01686411844287495 *S,
70 0.02674875741080976 *S
73 -0.5912717631142470 /S,
74 -0.05754352622849957 /S,
75 0.09127176311424948 /S
79 static const double icoeff[2][5]={
82 0.5912717631142470 /S,
83 -0.05754352622849957 /S,
84 -0.09127176311424948 /S
86 0.6029490182363579 *S,
87 -0.2668641184428723 *S,
88 -0.07822326652898785 *S,
89 0.01686411844287495 *S,
90 0.02674875741080976 *S
95 static inline int mirror(int x, int w){
96 while((unsigned)x > (unsigned)w){
103 static inline void decompose(float *dstL, float *dstH, float *src, int stride, int w){
106 double sumL= src[x*stride] * coeff[0][0];
107 double sumH= src[x*stride] * coeff[1][0];
109 double s= (src[mirror(x-i, w-1)*stride] + src[mirror(x+i, w-1)*stride]);
111 sumL+= coeff[0][i]*s;
112 sumH+= coeff[1][i]*s;
114 dstL[x*stride]= sumL;
115 dstH[x*stride]= sumH;
119 static inline void compose(float *dst, float *srcL, float *srcH, int stride, int w){
122 double sumL= srcL[x*stride] * icoeff[0][0];
123 double sumH= srcH[x*stride] * icoeff[1][0];
125 int x0= mirror(x-i, w-1)*stride;
126 int x1= mirror(x+i, w-1)*stride;
128 sumL+= icoeff[0][i]*(srcL[x0] + srcL[x1]);
129 sumH+= icoeff[1][i]*(srcH[x0] + srcH[x1]);
131 dst[x*stride]= (sumL + sumH)*0.5;
135 static inline void decompose2D(float *dstL, float *dstH, float *src, int xstride, int ystride, int step, int w, int h){
138 for(x=0; x<step; x++)
139 decompose(dstL + ystride*y + xstride*x, dstH + ystride*y + xstride*x, src + ystride*y + xstride*x, step*xstride, (w-x+step-1)/step);
142 static inline void compose2D(float *dst, float *srcL, float *srcH, int xstride, int ystride, int step, int w, int h){
145 for(x=0; x<step; x++)
146 compose(dst + ystride*y + xstride*x, srcL + ystride*y + xstride*x, srcH + ystride*y + xstride*x, step*xstride, (w-x+step-1)/step);
149 static void decompose2D2(float *dst[4], float *src, float *temp[2], int stride, int step, int w, int h){
150 decompose2D(temp[0], temp[1], src , 1, stride, step , w, h);
151 decompose2D( dst[0], dst[1], temp[0], stride, 1, step , h, w);
152 decompose2D( dst[2], dst[3], temp[1], stride, 1, step , h, w);
155 static void compose2D2(float *dst, float *src[4], float *temp[2], int stride, int step, int w, int h){
156 compose2D(temp[0], src[0], src[1], stride, 1, step , h, w);
157 compose2D(temp[1], src[2], src[3], stride, 1, step , h, w);
158 compose2D(dst , temp[0], temp[1], 1, stride, step , w, h);
161 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){
164 double s= p->strength[!is_luma];
167 while(1<<depth > width || 1<<depth > height)
170 for(y=0; y<height; y++)
171 for(x=0; x<width; x++)
172 p->plane[0][0][x + y*p->stride]= src[x + y*src_stride];
174 for(i=0; i<depth; i++){
175 decompose2D2(p->plane[i+1], p->plane[i][0], p->plane[0]+1,p->stride, 1<<i, width, height);
177 for(i=0; i<depth; i++){
179 for(y=0; y<height; y++){
180 for(x=0; x<width; x++){
181 double v= p->plane[i+1][j][x + y*p->stride];
185 p->plane[i+1][j][x + y*p->stride]= v;
190 for(i=depth-1; i>=0; i--){
191 compose2D2(p->plane[i][0], p->plane[i+1], p->plane[0]+1, p->stride, 1<<i, width, height);
194 for(y=0; y<height; y++)
195 for(x=0; x<width; x++){
196 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 :)
197 // double e= i - src[x + y*src_stride];
199 if((unsigned)i > 255U) i= ~(i>>31);
200 dst[x + y*dst_stride]= i;
203 // printf("%f\n", sum/height/width);
206 static int config(struct vf_instance *vf, int width, int height, int d_width, int d_height, unsigned int flags, unsigned int outfmt){
207 int h= (height+15)&(~15);
210 vf->priv->stride= (width+15)&(~15);
212 for(i=0; i<=vf->priv->depth; i++)
213 vf->priv->plane[i][j]= malloc(vf->priv->stride*h*sizeof(vf->priv->plane[0][0][0]));
216 return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
219 static void get_image(struct vf_instance *vf, mp_image_t *mpi){
220 if(mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change
221 // ok, we can do pp in-place (or pp disabled):
222 vf->dmpi=vf_get_image(vf->next,mpi->imgfmt,
223 mpi->type, mpi->flags | MP_IMGFLAG_READABLE, mpi->width, mpi->height);
224 mpi->planes[0]=vf->dmpi->planes[0];
225 mpi->stride[0]=vf->dmpi->stride[0];
226 mpi->width=vf->dmpi->width;
227 if(mpi->flags&MP_IMGFLAG_PLANAR){
228 mpi->planes[1]=vf->dmpi->planes[1];
229 mpi->planes[2]=vf->dmpi->planes[2];
230 mpi->stride[1]=vf->dmpi->stride[1];
231 mpi->stride[2]=vf->dmpi->stride[2];
233 mpi->flags|=MP_IMGFLAG_DIRECT;
236 static int put_image(struct vf_instance *vf, mp_image_t *mpi, double pts){
239 if(!(mpi->flags&MP_IMGFLAG_DIRECT)){
240 // no DR, so get a new image! hope we'll get DR buffer:
241 dmpi=vf_get_image(vf->next,mpi->imgfmt,
243 MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
244 mpi->width,mpi->height);
245 vf_clone_mpi_attributes(dmpi, mpi);
250 filter(vf->priv, dmpi->planes[0], mpi->planes[0], dmpi->stride[0], mpi->stride[0], mpi->w, mpi->h, 1);
251 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);
252 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);
254 return vf_next_put_image(vf,dmpi, pts);
257 static void uninit(struct vf_instance *vf){
259 if(!vf->priv) return;
263 free(vf->priv->plane[i][j]);
264 vf->priv->plane[i][j]= NULL;
272 //===========================================================================//
273 static int query_format(struct vf_instance *vf, unsigned int fmt){
286 return vf_next_query_format(vf,fmt);
292 static int vf_open(vf_instance_t *vf, char *args){
294 vf->put_image=put_image;
295 vf->get_image=get_image;
296 vf->query_format=query_format;
298 vf->priv=malloc(sizeof(struct vf_priv_s));
299 memset(vf->priv, 0, sizeof(struct vf_priv_s));
302 vf->priv->strength[0]= 1.0;
303 vf->priv->strength[1]= 1.0;
304 vf->priv->delta= 1.0;
306 if (args) sscanf(args, "%d:%f:%f:%d:%f", &vf->priv->depth,
307 &vf->priv->strength[0],
308 &vf->priv->strength[1],
315 const vf_info_t vf_info_ow = {
316 "overcomplete wavelet denoiser",
318 "Michael Niedermayer",