kernels.h

Go to the documentation of this file.

#ifndef _DLRCOMPUTERVISION_KERNELS_H_
#define _DLRCOMPUTERVISION_KERNELS_H_

#include <dlrComputerVision/kernel.h>

namespace dlr {

  namespace computerVision {
    
    template<class KERNEL_TYPE>
    Kernel<KERNEL_TYPE>
    getGaussianKernel(double rowSigma, double columnSigma);


    template<class KERNEL_TYPE>
    Kernel<KERNEL_TYPE>
    getGaussianKernel(size_t rows, size_t columns,
                      double rowSigma=-1.0, double columnSigma=-1.0);


  } // namespace computerVision

} // namespace dlr


/* =============== Implementation follows =============== */

#include <dlrNumeric/utilities.h>

namespace dlr {

  namespace computerVision {

    namespace privateCode {
      
      template <class TYPE>
      Array1D<TYPE>
      getGaussian1D(size_t size, double sigma, bool normalize=false)
      {
        const double myPi = 3.14159265359;
        Array1D<TYPE> result(size);
        double x = (1.0 - static_cast<double>(size))/2.0;
        double twoSigmaSq = 2.0 * sigma * sigma;
        double k = 1.0 / (std::sqrt(2.0 * myPi) * sigma);
        for(size_t index0 = 0; index0 < size; ++index0) {
          result[index0] = k * exp(-x * x / twoSigmaSq);
          x += 1.0;
        }
        if(normalize) {
          result /= sum(result);
        }
        return result;
      }
      
    } // namespace privateCode


    // This function generates and returns a separable Gaussian kernel.
    template<class KERNEL_TYPE>
    Kernel<KERNEL_TYPE>
    getGaussianKernel(double rowSigma, double columnSigma)
    {
      size_t rows = static_cast<size_t>(6.0 * rowSigma + 1.0);
      size_t columns = static_cast<size_t>(6.0 * columnSigma + 1.0);
      if(rows % 2 == 0) {
        ++rows;
      }
      if(columns % 2 == 0) {
        ++columns;
      }
      return getGaussianKernel<KERNEL_TYPE>(
        rows, columns, rowSigma, columnSigma);
    }

    
    // This function generates and returns a separable Gaussian kernel.
    template<class KERNEL_TYPE>
    Kernel<KERNEL_TYPE>
    getGaussianKernel(size_t rows, size_t columns,
                      double rowSigma, double columnSigma)
    {
      // Argument checking.
      if(rows == 0 || columns == 0) {
        DLR_THROW(ValueException, "getGaussianKernel()",
                  "Arguments rows and columns may not have value of zero.");
      }
      if(rowSigma < 0.0) {
        rowSigma = rows / 6.0;
      }
      if(columnSigma < 0.0) {
        columnSigma = columns / 6.0;
      }

      Array1D<KERNEL_TYPE> rowComponent =
        privateCode::getGaussian1D<KERNEL_TYPE>(columns, columnSigma, true);
      Array1D<KERNEL_TYPE> columnComponent =
        privateCode::getGaussian1D<KERNEL_TYPE>(rows, rowSigma, true);
      return Kernel<KERNEL_TYPE>(rowComponent, columnComponent);
    }


  } // namespace computerVision
  
} // namespace dlr

#endif /* #ifndef _DLRCOMPUTERVISION_KERNEL_H_ */