// bug? delete2d<int> would crash?!
void test_LemotionComp(){
int h = 3;
int w = 6;
int mb = 3;
const double LE[8][18] =
{{0 , 1 , 0 , 0 , 1 , 1 ,0, 0 , 1 ,1 ,1 ,1 ,0, 0 , 1 ,1 ,1 ,1},{0, 0 , 1 ,1 ,1 ,1, 0 , 1 , 1 ,0, 0 , 1, 0 , 1 , 1 ,0, 0 , 1},
{0 , 1 , 0 , 0 , 1 , 1 ,0, 0 , 1 ,1 ,1 ,1 ,0, 0 , 1 ,1 ,1 ,1},{0, 0 , 1 ,1 ,1 ,1, 0 , 1 , 1 ,0, 0 , 1, 0 , 1 , 1 ,0, 0 , 1},
{0 , 1 , 0 , 0 , 1 , 1 ,0, 0 , 1 ,1 ,1 ,1 ,0, 0 , 1 ,1 ,1 ,1},{0, 0 , 1 ,1 ,1 ,1, 0 , 1 , 1 ,0, 0 , 1, 0 , 1 , 1 ,0, 0 , 1},
{0 , 1 , 0 , 0 , 1 , 1 ,0, 0 , 1 ,1 ,1 ,1 ,0, 0 , 1 ,1 ,1 ,1},{0, 0 , 1 ,1 ,1 ,1, 0 , 1 , 1 ,0, 0 , 1, 0 , 1 , 1 ,0, 0 , 1}};
const int MV[2][3] = {{0,0,0},{0,0,0}};
double ** le = new2d<double>(PXL,w*h,0);
double ** comp = new2d<double>(PXL,w*h,0);
int ** mv = new2d<int>(2,h*w/mb/mb);
for(int i = 0 ; i < PXL ; ++i)
for(int j = 0 ; j < w*h ; ++j){
le[i][j] = LE[i][j];
}
for(int i = 0 ; i<h*w/mb/mb ; ++i){
mv[0][i] = MV[0][i];
mv[1][i] = MV[1][i];
}
printf("le:\n");
for(int i = 0 ; i < PXL ; ++i){
for(int j = 0 ; j < w*h ; ++j)
printf("%lf ",le[i][j]);
printf("\n");
}
motionComp(le,mv,h,w,mb,comp);
printf("\ncomp:\n");
for(int i = 0 ; i < PXL ; ++i){
for(int j = 0 ; j < w*h ; ++j)
printf("%lf ",comp[i][j]);
printf("\n");
}
printf("delete comp\n");
delete2d<double>(comp);
printf("delete le\n");
delete2d<double>(le);
printf("delete mv\n");
delete2d<int>(mv);
printf("done\n");
}
// bug? delete2d<int> would crash?!
void test_motionComp(){
int h = 3;
int w = 6;
int mb = 3;
const int imgI[3][6] = {{0 , 1 , 0 , 0 , 1 , 1},{0, 0 , 1 ,1 ,1 ,1},{0, 0 , 1 ,1 ,0 ,1}};
const int MV[2][3] = {{0,0,0},{0,0,0}};
int ** im = new2d<int>(h,w,0);
int ** comp = new2d<int>(h,w,0);
int ** mv = new2d<int>(2,h*w/mb/mb);
for(int i = 0 ; i < h ; ++i)
for(int j = 0 ; j < w ; ++j){
im[i][j] = imgI[i][j];
mv[0][i*w+j] = MV[0][i*w+j];
mv[1][i*w+j] = MV[1][i*w+j];
}
printf("imgI:\n");
for(int i = 0 ; i < h ; ++i){
for(int j = 0 ; j < w ; ++j)
printf("%d ",im[i][j]);
printf("\n");
}
motionComp(im,mv,h,w,mb,comp);
printf("\ncomp:\n");
for(int i = 0 ; i < h ; ++i){
for(int j = 0 ; j < w ; ++j)
printf("%d ",comp[i][j]);
printf("\n");
}
printf("delete comp\n");
delete2d<int>(comp);
printf("delete im\n");
delete2d<int>(im);
printf("delete mv\n");
delete2d<int>(mv);
printf("done\n");
}
//**************************POCS superresolution restoration algorithm************************************//
//function: 实现POCS超分辨率复原
//input:
//imgIn: 输入图像的指针，理想的HR图
//imgWidth: 输入图像的宽度
//imgHeight: 输入图像的高度
//LRNum: 产生的低分辨率图像的数目
//iterNum: 复原迭代次数
//output:
//imgOut: 输出图像的指针
void pocs(unsigned char* imgIn,int imgWidth,int imgHeight,int LRNum,int iterNum, unsigned char* imgOut, bool noiseAuto, double noiseVar)
{
//生成LRNum数目的LR图像
//LRNum 数目的LR图像指针
unsigned char *p[MAXLRNUM];
int newHeight = imgHeight/2;
int newWidth = imgWidth/2;
int masksize = 5;
int k;
/* blur =[.25 0 1 0 .25;...
0 1 2 1 0;...
1 2 4 2 1;...
0 1 2 1 0;...
.25 0 1 0 .25];
double blur[25] = {0.25,0,1.0,0,0.25,
0,1.0,2.0,1.0,
1.0,2.0,4.0,2.0,1.0,
0,1.0,2.0,1.0,0,
0.25,0,1.0,0,0.25};//21
for(k=0;k<masksize*masksize;k++)
{
blur[k]=blur[k]/21;
}
*/
double blur[25] = {0.00296901674395050, 0.0133062098910137, 0.0219382312797146, 0.0133062098910137, 0.00296901674395050,
0.0133062098910137, 0.0596342954361801, 0.0983203313488458, 0.0596342954361801, 0.0133062098910137,
0.0219382312797146, 0.0983203313488458, 0.162102821637127, 0.0983203313488458, 0.0219382312797146,
0.0133062098910137, 0.0596342954361801, 0.0983203313488458, 0.0596342954361801, 0.0133062098910137,
0.00296901674395050, 0.0133062098910137, 0.0219382312797146, 0.0133062098910137, 0.00296901674395050
};
int seqNum[9] = {1,5,7,9,2,3,6,4,8};
double sigmaHtr=0;
for(k=0;k<masksize*masksize;k++)
{
sigmaHtr=sigmaHtr+pow(*(blur+k),2);
}
for(k=0;k<LRNum;k++)
{
int s1_rand, s2_rand, s3_rand, s4_rand;
s1_rand = seqNum[k];
s2_rand = seqNum[k];
s3_rand = seqNum[k];
s4_rand = seqNum[k];
unsigned char *LRtmp[MAXLRNUM];
LRtmp[k] = new unsigned char [newHeight*newWidth];
p[k] = new unsigned char [imgWidth*imgHeight];
//产生低分辨率图像
GenLR(imgIn,LRtmp[k],imgWidth,imgHeight,noiseAuto,noiseVar,s1_rand,s2_rand,s3_rand,s4_rand);
//在此处直接进行双线性插值
zoomGray2(LRtmp[k], newWidth, newHeight, 2, 2,p[k], imgWidth, imgHeight);
delete []LRtmp[k];
}
//生成参考HR，p[0]第一幅图为参考图像---------------------------------到此为止算法正常实现
//开始迭代
for(int iter=0;iter<iterNum;iter++)
{
int err_acc;
for(int picNum=1;picNum<LRNum;picNum++)
{
int i,j;
//读入下一幅图，并与参考帧进行运动估计,通过运动补偿获得估计的运动图像f^
int motionVect[2][MAX];
double DScomputations[1];
//分配内存
unsigned char* imgComp = new unsigned char[imgWidth*imgHeight];
unsigned char* residual= new unsigned char[imgWidth*imgHeight];
// unsigned char residual;
motionEstDS( p[0],p[picNum], imgWidth, imgHeight, motionVect, DScomputations);
motionComp(p[picNum],16,imgWidth,imgHeight,motionVect,imgComp);//补偿向p[0]估计
//************************imgComp是对参考图像的估计********************************//
//乘上模糊PSF后算作估计值
Filter(imgComp,imgWidth,imgHeight,1,blur,masksize);
// 循环遍历每一个像素
for(i=0;i<imgHeight;i++)
{
for(j=0;j<imgWidth;j++)
{
int m,n;
unsigned char res;
//得出估计的残差数据块
res = *(imgComp+i*imgWidth+j)-*(p[0]+i*imgWidth+j);
//根据残差反向迭代每一个像素
err_acc += res;
if(res>R)
{
for(m=0;m<5;m++)
for(n=0;n<5;n++)
{
if((i+m-2<0)||(j+n-2)<0||(i+m-2>=imgHeight)||(j+n-2>=imgWidth))
continue;
else
*(residual+(i+m-2)*imgWidth+(j+n-2))+=(unsigned char)((res-R)**(blur+m*5+n)/sigmaHtr);
}
}
else if(res<-R)
{
for(m=0;m<5;m++)
for(n=0;n<5;n++)
{
if((i+m-2<0)||(j+n-2)<0||(i+m-2>=imgHeight)||(j+n-2>=imgWidth))
continue;
else
*(residual+(i+m-2)*imgWidth+(j+n-2))+=(unsigned char)((res+R)**(blur+m*5+n)/sigmaHtr);
}
}
else
*(residual+i*imgWidth+j)=0;
}
}
if(err_acc=255*R)
break;
for(i=0;i<imgHeight;i++)
{
for(j=0;j<imgWidth;j++)
{
*(p[0]+i*imgWidth+j)=*(p[0]+i*imgWidth+j)+*(residual+i*imgWidth+j);//残差的权重对图像复原效果有较大的影响
}
}
delete []imgComp;
delete []residual;
/*
for(i=0;i<imgHeight;i++)
{
for(j=0;j<imgWidth;j++)
{
//初始化误差和
err_acc = 0;
unsigned char res;
//得出估计的残差数据块
res = *(imgComp+i*imgWidth+j)-*(p[0]+i*imgWidth+j);
err_acc += res;
//根据残差反向迭代每一个像素
if(res>R)
*(residual+i*imgWidth+j)=res-R;
else if(res<-R)
*(residual+i*imgWidth+j)=res+R;
else
*(residual+i*imgWidth+j)=0;
}
}
Filter(residual,imgWidth,imgHeight,1,blur,masksize);
//反向迭代，对参考图像做修正，下一次循环使用的是修正后的参考图像
if(err_acc=255*R)
break;
for(i=0;i<imgHeight;i++)
{
for(j=0;j<imgWidth;j++)
{
*(p[0]+i*imgWidth+j)=*(p[0]+i*imgWidth+j)+*(residual+i*imgWidth+j)*1.9649;
}
}
delete []imgComp;
delete []residual;
*/
}
}
//复制图像输出
for(int i=0;i<imgHeight;i++)
{
for(int j=0;j<imgWidth;j++)
{
*(imgOut+i*imgWidth+j)=*(p[0]+i*imgWidth+j);
}
}
//清空内存
for(k=0;k<LRNum;k++)
{
delete []p[k];
}
}
int main(int argc,char* argv[]){
startRandom();
// control SNR step
const double s_snr = -1;
const double step = 0.05;
const int snr_size = 500;
// control the certain Frame and the bp
int frame = 2;
if(argc >= 4)
frame = strtod(argv[3],NULL);
// for video encode
const int puncture = 0; // puncture or not
const double rate = 1/(double)(2-puncture); // code rate
const double a = 1; // Fading amplitude. a=1 -> AWGN channel
double EbN0,L_c,sigma;
// for basical info
char buffer[50];
const int h = __HEIGHT, w = __WIDTH, f = frame+2 ;
const int lm = h*w;
const int lu = lm+(G_L-1);
int*** Y = new3d<int>(f,h,w);
int** MV_prev = new2d<int>(2,lm/64,0);
int** MV = new2d<int>(2,lm/64,0);
double** Lu = new2d<double>(PXL,lu,0);
double** Lu_prev = new2d<double>(PXL,lu,0);
double** Lu_next = new2d<double>(PXL,lu,0);
// frame buffer
int*** imgr_bp = new3d<int>(PXL,h,w);
int*** imgr_bp_prev = new3d<int>(PXL,h,w);
int*** imgr_bp_next = new3d<int>(PXL,h,w);
int*** img = new3d<int>(PXL,h,w);
int*** img_prev = new3d<int>(PXL,h,w);
int*** img_next = new3d<int>(PXL,h,w);
int** imgr = new2d<int>(h,w);
int** imgr_prev = new2d<int>(h,w);
int** imgr_next = new2d<int>(h,w);
double** Lu_c = new2d<double>(PXL,lu); //channel decoder output
// buffer for Ia, Ie
double** Le = new2d<double>(PXL,lm); // source extrinsic informatiom
double** Le_s_inter = new2d<double>(PXL,lm,0); // source extrinsic information
double** Le_s_inter_next = new2d<double>(PXL,lm,0); // source extrinsic information
double** Le_s_intra = new2d<double>(PXL,lm,0); // source extrinsic information
double** Ia = new2d<double>(PXL,snr_size);
double** Ie = new2d<double>(PXL,snr_size);
int idx = 0;
double tmp_sum = 0, tmp ;
double* beta_t = (double*) malloc(sizeof(double)*PXL);
double* beta_s = (double*) malloc(sizeof(double)*PXL);
double* beta_t_prev = (double*) malloc(sizeof(double)*PXL);
// read YUV
sprintf(buffer,"%s_cif.yuv",argv[1]);
yuv_read(buffer,h,w,f,Y,NULL,NULL);
img2bp_frame(Y[frame],h,w,imgr_bp);
img2bp_frame(Y[frame-1],h,w,imgr_bp_prev);
img2bp_frame(Y[frame+1],h,w,imgr_bp_next);
printf("processing ... %2.2f%%",0);
for(double snr = s_snr; idx < snr_size ; snr+=step, idx++ ){
printf("\rprocessing ... %2.2f%%",100*(snr-s_snr)/step/snr_size);
// encode
EbN0 = pow(10,snr/10); // convert Eb/N0[dB] to normal number
L_c = 4*a*EbN0*rate; // reliability value of the channel
sigma = 1/sqrt(2*rate*EbN0); // standard deviation of AWGN noise
for(int t_lvl = 0 ; t_lvl <PXL ; ++t_lvl){
tmp_sum = 0;
for(int i = 0 ; i < lm ; ++i){
tmp = (2*imgr_bp[t_lvl][i/w][i%w] - 1);
Lu[t_lvl][i] = 0.5*L_c*( tmp + sigma*gaussian_noise());
tmp_sum+=log2(1 + exp(-Lu[t_lvl][i]*tmp));
tmp = (2*imgr_bp_prev[t_lvl][i/w][i%w] - 1);
Lu_prev[t_lvl][i] = 0.5*L_c*( tmp + sigma*gaussian_noise());
}
Ia[t_lvl][idx] = 1 - tmp_sum/lm;
}
// decode
for(int i = 0, j=0, t_lvl=0; i <h ; ++i)
for(j = 0 ; j < w ; ++j)
for(t_lvl=0 ; t_lvl < PXL ; ++t_lvl){
img[t_lvl][i][j] = ((Lu[t_lvl][j+i*w]>=0)?1:0);
img_prev[t_lvl][i][j] = ((Lu_prev[t_lvl][j+i*w]>=0)?1:0);
}
bin2dec_img(img,h,w,imgr);
bin2dec_img(img_prev,h,w,imgr_prev);
motionEstES(imgr,imgr_prev,h,w,8,5,MV_prev);
intra_inter_beta_estimation(img,img_prev,MV_prev,beta_s,beta_t,h,w,8);
motionComp(Lu_prev,MV_prev,h,w,8,Lu_c);
mrf_siso_inter(Lu,Lu_c,beta_t,h,w,Le_s_inter,0);
mrf_siso_intra(Lu,beta_s,h,w,Le_s_intra,0);
for(int i = 0, t_lvl = 0 ; i < lm ; ++i)
for(t_lvl = 0 ; t_lvl < PXL ; ++t_lvl)
Le[t_lvl][i] = (Le_s_inter[t_lvl][i] + Le_s_intra[t_lvl][i]); // Le_s_prev in the next frame
// prepare for next frame decode
// encode
for(int t_lvl = 0 ; t_lvl <PXL ; ++t_lvl){
for(int i = 0 ; i < lm ; ++i){
tmp = (2*imgr_bp_next[t_lvl][i/w][i%w] - 1);
Lu_next[t_lvl][i] = 0.5*L_c*( tmp + sigma*gaussian_noise()); // reuse as Lu_next
}
}
for(int i = 0, j=0, t_lvl=0; i <h ; ++i)
for(j = 0 ; j < w ; ++j)
for(t_lvl=0 ; t_lvl < PXL ; ++t_lvl){
img_next[t_lvl][i][j] = ((Lu_next[t_lvl][j+i*w]>=0)?1:0);
}
bin2dec_img(img_next,h,w,imgr_next);
motionEstES(imgr,imgr_next,h,w,8,5,MV);
if(frame==1){
intra_inter_beta_estimation(imgr_bp,imgr_bp_next,MV,beta_s,beta_t,h,w,8);
for(int i = 0 ; i < PXL ; ++i)
beta_t_prev[i] = 0;
}
else
intra_inter2_beta_estimation(imgr_bp_next,imgr_bp,imgr_bp_prev,MV,beta_s,beta_t,beta_t_prev,h,w,8);
motionComp(Lu_next,MV,h,w,8,Lu_c);
mrf_siso_inter(Lu,Lu_c,beta_t,h,w,Le_s_inter_next,0);
mrf_siso_intra(Lu,beta_s,h,w,Le_s_intra,0);
for(int i = 0, t_lvl = 0 ; i < lm ; ++i)
for(t_lvl = 0 ; t_lvl < PXL ; ++t_lvl)
Le[t_lvl][i] = (Le_s_inter_next[t_lvl][i] + Le_s_intra[t_lvl][i] + Le_s_inter[t_lvl][i]*beta_t_prev[t_lvl]);
// compute Ie
for(int t_lvl = 0 , i = 0; t_lvl < PXL ; ++t_lvl){
tmp_sum = 0;
for(i=0; i < lm ; ++i){
tmp_sum+=log2(1 + exp(-Le[t_lvl][i]*(2*imgr_bp[0][0][i+t_lvl*lm]-1)));
}
Ie[t_lvl][idx] = 1 - tmp_sum/lm;
}
}
printf("\r100.00%% completed!\n");
FILE *file = fopen("output/exit_curve_ps.txt","a+");
for(int t_lvl = 0 ; t_lvl < PXL ; ++t_lvl){
fprintf(file,"==== v2 ========================\n%s: frame#%d,bp#%d\nIa=\n",argv[1],frame+1,t_lvl+1);
for(int i = 0 ; i < snr_size ; ++i)
fprintf(file,"%lf,",Ia[t_lvl][i]);
fprintf(file,"\nIe=\n");
for(int i = 0 ; i < snr_size ; ++i)
fprintf(file,"%lf,",Ie[t_lvl][i]);
fprintf(file,"\n\n");
}
fclose(file);
write_ps_for_matlab(Ia,Ie,snr_size);
// free memory
deleteY(Y);
delete3d<int>(imgr_bp);
delete3d<int>(imgr_bp_prev);
delete3d<int>(imgr_bp_next);
delete3d<int>(img_next);
delete3d<int>(img_prev);
delete3d<int>(img);
delete2d<int>(imgr_next);
delete2d<int>(imgr_prev);
delete2d<int>(imgr);
delete2d<double>(Lu_c);
delete2d<double>(Lu_next);
delete2d<double>(Lu);
delete2d<double>(Lu_prev);
delete2d<double>(Ia);
delete2d<double>(Ie);
delete2d<int>(MV_prev);
delete2d<int>(MV);
delete2d<double>(Le);
delete2d<double>(Le_s_inter);
delete2d<double>(Le_s_inter_next);
delete2d<double>(Le_s_intra);
free(beta_s);
free(beta_t);
free(beta_t_prev);
}
