bool bayesian_nnetwork<T>::calculate(const math::vertex<T>& input,
math::vertex<T>& mean,
math::matrix<T>& covariance)
{
if(nnets.size() <= 0) return false;
const unsigned int D = nnets[0]->output_size();
mean.resize(D);
covariance.resize(D,D);
mean.zero();
covariance.zero();
T ninv = T(1.0f/nnets.size());
T ninv2 = T(1.0f/(nnets.size() - 1));
if(nnets.size() <= D)
covariance.identity(); // regularizer term for small datasize
for(unsigned int i=0;i<nnets.size();i++){
nnets[i]->input() = input;
nnets[i]->calculate();
math::vertex<T> out = nnets[i]->output();
mean += ninv*out;
covariance += ninv2*out.outerproduct();
}
covariance -= mean.outerproduct();
return true;
}
/** Returns reflected version of a matrix */
const math::matrix<float> Convolution::reflection(const math::matrix<float> A)
{
int size = A.rowno();
math::matrix<float> C(size, size);
int col = A.colno();
int row = A.rowno();
for (int x = 0; x < row; x++){
for (int y = 0; y < col; y++){
C((size - 1 - x), (size - 1 - y)) = A(x, y);
}
}
return C;
}
/** Sums all of the matrixes elements */
const float Convolution::sum(const math::matrix<float> A)
{
float sum = 0.0;
int col = A.colno();
int row = A.rowno();
for(int x=0; x<row; x++){
for(int y=0; y<col; y++){
sum += A(x,y);
}
}
return sum;
}
double BayesianClassifier::calculate_recognition_error(const math::matrix<double> & x,
const QVector<Distribution> & distributions,
const math::matrix<double> & regretMatrix)
{
double R = 0.0;
bool first = true;
for (size_t i = 0; i < regretMatrix.RowNo(); ++i)
{
double tR = 0.0;
for (int j = 0; j < distributions.size(); ++j)
tR += regretMatrix(i, j)*g(x, distributions[j]);
if (first)
{
R = tR;
first = false;
}
if (R > tR)
R = tR;
}
return -R;
}
const int MorphDilate::morph(math::matrix<float> window, math::matrix<bool> se)
{
float min = PIXEL_VAL_MAX+1;
int tempx,tempy;
tempx = window.rowno();
tempy = window.colno();
for(int i=0;i<tempx;i++){
for(int j=0;j<tempy;j++){
if(se(i,j)==true){
if(window(i,j)<min){
min = window(i,j);
}
}
}
}
return (int) min;
}
void bind(bool force = false) const
{
if ((flags_ & TRANSFORM_BIND_DISABLED) && !force)
return;
if ((flags_ & TRANSFORM_CACHE) == 0)
{
transform_matrix_.set_identity();
transform_matrix_.set_scale(scale_);
transform_matrix_ *= orientation_.to_matrix();
transform_matrix_.apply_translation(location_);
flags_ |= TRANSFORM_CACHE;
}
}
const int MorphErode::morph(math::matrix<float> window, math::matrix<bool> se)
{
float max=0.0;
int row = se.rowno();
int col = se.colno();
for(int i=0; i<row; i++)
{
for(int j=0; j<col; j++)
{
if(se(i,j)==true && window(i,j)>max)
{
max = window(i,j);
}
}
}
return max;
}
/** Joins to matrices by multiplying the A[i,j] with B[i,j].
* Warning! Both Matrices must be squares with the same size!
*/
const math::matrix<float> Convolution::join(math::matrix<float> A, math::matrix<float> B)
{
int size = A.rowno();
math::matrix<float> C(size, size);
for(int x=0; x<size; x++){
for(int y=0; y<size; y++) {
C(x, y) = A(x, y) * B(x, y);
}
}
return C;
}
/** Joins to matrices by multiplying the A[i,j] with B[i,j].
* Warning! Both Matrices must be squares with the same size!
*/
const math::matrix<double> Convolution::join(math::matrix<double> A, math::matrix<double> B)
{
int size = A.RowNo();
math::matrix<double> C(size, size);
for(int i = 0; i < size; i++)
for(int j = 0; j < size; j++)
{
C(i, j) = A(i, j) * B(i, j);
}
return C;
}
/** Sums all of the matrixes elements */
const double Convolution::sum(const math::matrix<double> A)
{
double sum = 0.0;
int size = A.RowNo();
for(int i = 0; i < size; i++)
for(int j = 0; j < size; j++)
{
sum += A(i, j);
}
return sum;
}
const int MorphDilate::morph(math::matrix<double> window, math::matrix<bool> se)
{
double min = PIXEL_VAL_MAX+1;
int size = window.ColNo();
for (int i=0;i<size;i++){
for (int j=0;j<size;j++){
if (se(i,j)) {
if (window(i,j)<min) {
min=window(i,j);
}
}
}
}
return min;
}