Vector<T> IM_Solver::Gauss_Seidel_Iteration(const Base_Matrix<T>& A,
const Vector<T>& b,
Vector<T>& XO,
const T TOL,
const unsigned int Limit)
{
unsigned int k = 1, n = b.getSize();
Vector<T> x(n);
if(A.numCols() != n || A.numRows() != n || XO.getSize() != n)
throw Incompatible_Vector_Err();
while(k <= Limit)
{
for(unsigned int i = 0; i < n; i++)
{
T sum = 0;
for(unsigned int j = 0; j < n; j++)
{
if(j < i)
{
sum += -A(i,j)*x[j];
}
else if(j > i)
{
sum += -A(i,j)*XO[j];
}
}
x[i] = (sum+b[i])/A(i,i);
}
if(L2_norm(x-XO) < TOL)
{
return x;
}
k++;
XO = x;
}
return Vector<T>(0);
}
//Main function
void *Sub_denoise_mt(void *arguments)
{
nl_mean_mt arg;
arg=*(nl_mean_mt *)arguments;
int PID=0;
PID=arg.thread_num;
int ret;
int x_min,x_max,y_min,y_max, z_min, z_max;
int NbElement2 = (2*arg.neighborhoodsize[0]+1)*(2*arg.neighborhoodsize[1]+1)*(2*arg.neighborhoodsize[2]+1);
typedef float Voisinages[NbElement2];
Voisinages V1,V2;
float w_max = 0;
double epsilon= 0.0001;
int offset_k = 0;
int offset_j = 0;
int offset_kk = 0;
int offset_jj = 0;
for (int k=arg.debut; k < arg.fin ; k++)
{
double count = 0;
offset_k = k*(arg.vol_size[0]*arg.vol_size[1]);
for (int j=0; j < arg.vol_size[1] ; j++)
{
offset_j = j*arg.vol_size[0];
for (int i=0; i < arg.vol_size[0] ; i++)
{
float global_sum = 0;
float average = 0;
x_min = MAX(0,i-arg.search[0]); x_max = MIN(arg.vol_size[0]-1,i+arg.search[0]);
y_min = MAX(0,j-arg.search[1]); y_max = MIN(arg.vol_size[1]-1,j+arg.search[1]);
z_min = MAX(0,k-arg.search[2]); z_max = MIN(arg.vol_size[2]-1,k+arg.search[2]);
// Only the voxel with a local mean and variance different to zero are processed.
// By this way the computational time is reduced since the background is not taken into account (or the B-scan mask for US image)
if ((arg.mean_map[offset_k + offset_j + i] > epsilon) && (arg.var_map[offset_k + offset_j + i] > epsilon))
{
if ((arg.weight_method == 0) || (arg.weight_method == 2))
Neiborghood(arg.in,i,j,k,arg.neighborhoodsize,V1,arg.vol_size,arg.weight_method);
//
if (arg.weight_method == 1)
Neiborghood(arg.in,i,j,k,arg.neighborhoodsize,V1,arg.vol_size,arg.weight_method);
w_max = 0;
for (int kk = z_min; kk <= z_max; kk++)
{
offset_kk = kk*(arg.vol_size[0]*arg.vol_size[1]);
for (int jj = y_min; jj <= y_max; jj++)
{
offset_jj = jj*arg.vol_size[0];
for (int ii = x_min; ii <= x_max; ii++)
{
// To avoid the compute the weigh with null patch and the division by zero in ratio and ratio 2
if ((arg.mean_map[offset_kk + offset_jj + ii] > epsilon) && (arg.var_map[offset_kk + offset_jj + ii] > epsilon))
{
float ratio = arg.mean_map[offset_k + offset_j + i]/arg.mean_map[offset_kk + offset_jj + ii];
float ratio2 = arg.var_map[offset_k + offset_j + i]/arg.var_map[offset_kk + offset_jj + ii];
float weight = 0;
if ((arg.weight_method == 0) || (arg.weight_method == 2))
{
if ( (arg.m_min <= ratio) && (ratio <= (1/arg.m_min))
&& (arg.v_min <= ratio2) && (ratio2 <= (1/arg.v_min)) )
{
if ((ii != i) || (jj != j) || (kk != k))
{
float val_in;
count = count + 1;
Neiborghood(arg.in,ii,jj,kk,arg.neighborhoodsize,V2,arg.vol_size,arg.weight_method);
weight = Weight(L2_norm(V1,V2,arg.neighborhoodsize),arg.beta,arg.filtering_param);
global_sum = global_sum + weight;
val_in=arg.hallucinate?arg.hallucinate[offset_kk + offset_jj + ii]:arg.in[offset_kk + offset_jj + ii];//VF
//Rician denoising
if (arg.weight_method == 2)
average = average + val_in * val_in *weight;
else
average = average + val_in * weight;
if (weight > w_max) w_max = weight;
}
}
}
else
{
if ( (arg.m_min <= ratio) && (ratio <= (1/arg.m_min)))
{
if ((ii != i) || (jj != j) || (kk != k))
{
float val_in;
count = count + 1;
//Pearson distance computation
weight = Weight(Pearson_distance(V1,V2,arg.neighborhoodsize),arg.beta,arg.filtering_param);
global_sum = global_sum + weight;
val_in=arg.hallucinate?arg.hallucinate[offset_kk + offset_jj + ii]:arg.in[offset_kk + offset_jj + ii]; //VF
average = average + val_in * weight;
if (weight > w_max) w_max = weight;
}
}
}
}
}
}
}
global_sum = global_sum + w_max;
double denoised_value = 0.0;
float val_in=arg.hallucinate?arg.hallucinate[offset_k + offset_j + i]:arg.in[offset_k + offset_j + i];//VF
// Rician denoising
if (arg.weight_method == 2)
{
average = average + val_in * val_in * w_max;
if (global_sum != 0.0)
{
denoised_value = (average/global_sum) - (2*arg.filtering_param * arg.filtering_param);
if (denoised_value > 0.0)
denoised_value = sqrt (denoised_value);
else
denoised_value = 0.0;
arg.out[offset_k + offset_j + i] = denoised_value;
}
else
{
arg.out[offset_k + offset_j + i] = val_in;
}
}
// Gaussian and Speckle denoising
else
{
average = average + val_in*w_max;
if (global_sum != 0.0)
arg.out[offset_k + offset_j + i] = average/global_sum;
else
arg.out[offset_k + offset_j + i] = val_in;
}
}
}
}
if(debug)
std::cout<<"Thread ( "<< std::setw(2) <<PID+1<< " ) has finished the slice :"<<std::setw(3)<< k <<std::endl;
ret = pthread_mutex_lock(arg.mp);
*arg.mean_neighboors = *arg.mean_neighboors + count/(arg.vol_size[0]*arg.vol_size[1]);
ret = pthread_mutex_unlock(arg.mp);
}
if(debug)
std::cout<<"End of the thread : "<<std::setw(2)<<PID+1<<std::endl;
pthread_exit(0);
}
