crossCorrelation.hpp

Go to the documentation of this file.

/** \file crossCorrelation.hpp
 * \brief Declares direct discrete cross-correlation utilities.
 * \author Jared R. Males (jaredmales@gmail.com)
 */
 
#ifndef __crossCorrelation_hpp__
#define __crossCorrelation_hpp__
 
#include "../math/fit/fitGaussian.hpp"
 
namespace mx
{
 
// make these _worker
// change to pre-allocate native arrays
// add x and y lag differences
 
// template<class eigenT, class eigenTin1, class eigenTin2>
template <typename floatT, typename sizeT = size_t>
void calcDiscreteCrossCorrelation( floatT *cc,
                                   floatT *m1,
                                   floatT *m2,
                                   sizeT dim1,
                                   sizeT dim2,
                                   sizeT min_x_lag,
                                   sizeT max_x_lag,
                                   sizeT min_y_lag,
                                   sizeT max_y_lag )
{
 
    sizeT cc_dim1, cc_dim2;
    //   size_t dim1 = m1.rows();
    //   size_t dim2 = m2.cols();
 
    //   if( dim1 != m2.rows() || dim2 != m2.cols())
    //  {
    // error
    //     return;
    //  }
 
    //    if( min_x_lag <= 0 )
    //    {
    //       max_x_lag = dim1; //std::min(dim1,dim2);
    //       min_x_lag = -max_x_lag;
    //    }
    //
    //    if( min_y_lag <= 0 )
    //    {
    //       max_y_lag = dim2; //std::min(dim1,dim2);
    //       min_y_lag = -max_y_lag;
    //    }
 
    // cc.resize(maxlag-minlag+1, maxlag-minlag+1);
 
    cc_dim1 = max_x_lag - min_x_lag + 1;
    cc_dim2 = max_y_lag - min_y_lag + 1;
 
    // pout("cc_dims ", cc_dim1, cc_dim2);
    for( sizeT i = min_x_lag; i < max_x_lag + 1; ++i )
    {
        for( sizeT j = min_y_lag; j < max_y_lag + 1; ++j )
        {
            // cc(i-minlag, j-minlag) = 0;
            cc[( j - min_y_lag ) * cc_dim1 + ( i - min_x_lag )] = 0;
 
            // pout((j-min_y_lag)*cc_dim1 + (i-min_x_lag));
 
            for( sizeT k = 0; k < dim1; ++k )
            {
                if( k + i < 0 )
                    continue;
                if( k + i >= dim1 )
                    continue;
 
                for( sizeT l = 0; l < dim2; ++l )
                {
                    if( l + j < 0 )
                        continue;
                    if( l + j >= dim2 )
                        continue;
                    cc[( j - min_y_lag ) * cc_dim1 + ( i - min_x_lag )] +=
                        m1[l * dim1 + k] * m2[( l + j ) * dim1 + ( k + i )];
                }
            }
        }
    }
}
 
template <class eigenT, class eigenTin1, class eigenTin2, class eigenTmask>
void xdiscreteCrossCorrelation(
    eigenT &cc, const eigenTin1 &m1, const eigenTin2 &m2, eigenTmask &mask, int minlag = 0, int maxlag = 0 )
{
 
    size_t dim1 = m1.rows();
    size_t dim2 = m2.cols();
 
    if( dim1 != m2.rows() || dim2 != m2.cols() || dim1 != mask.rows() || dim2 != mask.cols() )
    {
        // error
        return;
    }
 
    if( minlag <= 0 )
    {
        maxlag = std::min( dim1, dim2 );
        minlag = -maxlag;
    }
 
    cc.resize( maxlag - minlag + 1, maxlag - minlag + 1 );
 
    for( int i = minlag; i < maxlag + 1; i++ )
    {
        for( int j = minlag; j < maxlag + 1; j++ )
        {
            cc( i - minlag, j - minlag ) = 0;
 
            for( int k = 0; k < dim1; k++ )
            {
                if( k + i < 0 )
                    continue;
                if( k + i >= dim1 )
                    continue;
 
                for( int l = 0; l < dim2; l++ )
                {
                    if( l + j < 0 )
                        continue;
                    if( l + j >= dim2 )
                        continue;
                    cc( i - minlag, j - minlag ) +=
                        m1( k, l ) * mask( k, l ) * m2( k + i, l + j ) * mask( k + i, l + j );
                }
            }
        }
    }
}
 
template <class eigenT, class eigenTin1, class eigenTin2>
void discreteCrossCorrelation( typename eigenT::Scalar &xlag,
                               typename eigenT::Scalar &ylag,
                               eigenT &cc,
                               const eigenTin1 &m1,
                               const eigenTin2 &m2,
                               int min_x_lag = 0,
                               int max_x_lag = 0,
                               int min_y_lag = 0,
                               int max_y_lag = 0 )
{
    math::fit::fitGaussian2D<math::fit::gaussian2D_gen_fitter<float>> fitG;
 
    size_t dim1 = m1.rows();
    size_t dim2 = m1.cols();
 
    if( dim1 != m2.rows() || dim2 != m2.cols() )
    {
        // error
        return;
    }
 
    //    if( min_x_lag == 0 )
    //    {
    //       max_x_lag = dim1;//std::min(dim1,dim2);
    //       min_x_lag = -max_x_lag;
    //    }
    //
    //    if( min_y_lag <= 0 )
    //    {
    //       max_y_lag = dim2;//std::min(dim1,dim2);
    //       min_y_lag = -max_y_lag;
    //    }
 
    cc.resize( max_x_lag - min_x_lag + 1, max_y_lag - min_y_lag + 1 );
 
    calcDiscreteCrossCorrelation<float, int>( (float *)cc.data(),
                                              (float *)m1.data(),
                                              (float *)m2.data(),
                                              dim1,
                                              dim2,
                                              min_x_lag,
                                              max_x_lag,
                                              min_y_lag,
                                              max_y_lag );
 
    fitG.setArray( cc.data(), cc.rows(), cc.cols() );
 
    typename eigenT::Index row, col;
    typename eigenT::Scalar maxval;
    maxval = cc.maxCoeff( &row, &col );
    cc /= maxval; // Normalize to avoid crazy nans in fit, etc.
    // pout(maxval, row, col);
 
    fitG.setGuess( 0., 1., row, col, 4., 4., 0. );
    fitG.fit();
 
    // fitG.dump_report();
 
    xlag = min_x_lag + fitG.get_params()[2];
    ylag = min_y_lag + fitG.get_params()[3];
}
 
} // namespace mx
 
#endif //__crossCorrelation_hpp__