otbHistogramOfOrientedGradientCovariantImageFunction.hxx

Go to the documentation of this file.

/*
 * Copyright (C) 2005-2022 Centre National d'Etudes Spatiales (CNES)
 *
 * This file is part of Orfeo Toolbox
 *
 *     https://www.orfeo-toolbox.org/
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
 
#ifndef otbHistogramOfOrientedGradientCovariantImageFunction_hxx
#define otbHistogramOfOrientedGradientCovariantImageFunction_hxx
 
#include "otbHistogramOfOrientedGradientCovariantImageFunction.h"
#include "itkConstNeighborhoodIterator.h"
#include "itkNumericTraits.h"
#include "otbMath.h"
 
namespace otb
{
 
template <class TInputImage, class TOutputPrecision, class TCoordRep>
HistogramOfOrientedGradientCovariantImageFunction<TInputImage, TOutputPrecision, TCoordRep>::HistogramOfOrientedGradientCovariantImageFunction()
  : m_NeighborhoodRadius(8), m_NumberOfOrientationBins(18)
{
}
 
template <class TInputImage, class TOutputPrecision, class TCoordRep>
void HistogramOfOrientedGradientCovariantImageFunction<TInputImage, TOutputPrecision, TCoordRep>::PrintSelf(std::ostream& os, itk::Indent indent) const
{
  this->Superclass::PrintSelf(os, indent);
  os << indent << " Neighborhood radius value   : " << m_NeighborhoodRadius << std::endl;
}
 
template <class TInputImage, class TOutputPrecision, class TCoordRep>
typename HistogramOfOrientedGradientCovariantImageFunction<TInputImage, TOutputPrecision, TCoordRep>::OutputType
HistogramOfOrientedGradientCovariantImageFunction<TInputImage, TOutputPrecision, TCoordRep>::EvaluateAtIndex(const IndexType& index) const
{
  // Build output vector
  OutputType hog;
 
  // Check for input image
  if (!this->GetInputImage())
  {
    return hog;
  }
 
  // Check for out of buffer
  if (!this->IsInsideBuffer(index))
  {
    return hog;
  }
 
  // Create an N-d neighborhood kernel, using a zeroflux boundary condition
  typename InputImageType::SizeType kernelSize;
  kernelSize.Fill(m_NeighborhoodRadius);
 
  itk::ConstNeighborhoodIterator<InputImageType> it(kernelSize, this->GetInputImage(), this->GetInputImage()->GetBufferedRegion());
 
  // Set the iterator at the desired location
  it.SetLocation(index);
 
  // Offset to be used in the loops
  typename InputImageType::OffsetType offset;
 
  // Compute the center bin radius
  double centerBinRadius = static_cast<double>(m_NeighborhoodRadius) / 2;
 
  // Define a gaussian kernel around the center location
  double squaredRadius          = m_NeighborhoodRadius * m_NeighborhoodRadius;
  double squaredCenterBinRadius = centerBinRadius * centerBinRadius;
  double squaredSigma           = 0.25 * squaredRadius;
 
  // Build a global orientation histogram
  std::vector<TOutputPrecision> globalOrientationHistogram(m_NumberOfOrientationBins, 0.);
 
  // Compute the orientation bin width
  double orientationBinWidth = otb::CONST_2PI / m_NumberOfOrientationBins;
 
  // Build the global orientation histogram
  for (int i = -(int)m_NeighborhoodRadius; i < (int)m_NeighborhoodRadius; ++i)
  {
    for (int j = -(int)m_NeighborhoodRadius; j < (int)m_NeighborhoodRadius; ++j)
    {
      // Check if the current pixel lies within a disc of radius m_NeighborhoodRadius
      double currentSquaredRadius = i * i + j * j;
      if (currentSquaredRadius < squaredRadius)
      {
        // If so, compute the gaussian weighting (this could be
        // computed once for all for the sake of optimisation)
        double gWeight = (1 / std::sqrt(otb::CONST_2PI * squaredSigma)) * std::exp(-currentSquaredRadius / (2 * squaredSigma));
 
        // Compute pixel location
        offset[0] = i;
        offset[1] = j;
 
        // Get the current gradient covariant value
        InputPixelType gradient = it.GetPixel(offset);
 
        // Then, compute the gradient orientation
        double angle = std::atan2(gradient[1], gradient[0]);
 
        // Also compute its magnitude
        TOutputPrecision magnitude = std::sqrt(gradient[0] * gradient[0] + gradient[1] * gradient[1]);
 
        // Determine the bin index (shift of otb::CONST_PI since atan2 values
        // lies in [-pi, pi]
        unsigned int binIndex = std::floor((otb::CONST_PI + angle) / orientationBinWidth);
 
        // Handle special case where angle = pi, and binIndex is out-of-bound
        if (binIndex == m_NumberOfOrientationBins)
          binIndex = m_NumberOfOrientationBins - 1;
 
        // Cumulate values
        globalOrientationHistogram[binIndex] += magnitude * gWeight;
      }
    }
  }
 
  // Compute principal orientation
  // TODO: Replace this with a stl algorithm
  double       maxOrientationHistogramValue = globalOrientationHistogram[0];
  unsigned int maxOrientationHistogramBin   = 0;
 
  for (unsigned int i = 1; i < m_NumberOfOrientationBins; ++i)
  {
    // Retrieve current value
    double currentValue = globalOrientationHistogram[i];
 
    // Look for new maximum
    if (maxOrientationHistogramValue < currentValue)
    {
      maxOrientationHistogramValue = currentValue;
      maxOrientationHistogramBin   = i;
    }
  }
 
  // Derive principal orientation
  double principalOrientation = maxOrientationHistogramBin * orientationBinWidth - otb::CONST_PI;
 
  // Initialize the five spatial bins
  std::vector<TOutputPrecision> centerHistogram(m_NumberOfOrientationBins, 0.);
  std::vector<TOutputPrecision> upperLeftHistogram(m_NumberOfOrientationBins, 0.);
  std::vector<TOutputPrecision> upperRightHistogram(m_NumberOfOrientationBins, 0.);
  std::vector<TOutputPrecision> lowerLeftHistogram(m_NumberOfOrientationBins, 0.);
  std::vector<TOutputPrecision> lowerRightHistogram(m_NumberOfOrientationBins, 0.);
 
  // Walk the image once more to fill the spatial bins
  for (int i = -(int)m_NeighborhoodRadius; i < (int)m_NeighborhoodRadius; ++i)
  {
    for (int j = -(int)m_NeighborhoodRadius; j < (int)m_NeighborhoodRadius; ++j)
    {
      // Check if the current pixel lies within a disc of radius m_NeighborhoodRadius
      double currentSquaredRadius = i * i + j * j;
      if (currentSquaredRadius < squaredRadius)
      {
        // If so, compute the gaussian weighting (this could be
        // computed once for all for the sake of optimisation)
        double gWeight = (1 / std::sqrt(otb::CONST_2PI * squaredSigma)) * std::exp(-currentSquaredRadius / (2 * squaredSigma));
 
        // Compute pixel location
        offset[0] = i;
        offset[1] = j;
 
        // Get the current gradient covariant value
        InputPixelType gradient = it.GetPixel(offset);
 
        // Then, compute the compensated gradient orientation
        double angle = std::atan2(gradient[1], gradient[0]) - principalOrientation;
 
        // Angle is supposed to lie with [-pi, pi], so we ensure that
        // compenstation did not introduce out-of-range values
        if (angle > otb::CONST_PI)
        {
          angle -= otb::CONST_2PI;
        }
        else if (angle < -otb::CONST_PI)
        {
          angle += otb::CONST_2PI;
        }
 
        // Also compute its magnitude
        TOutputPrecision magnitude = std::sqrt(gradient[0] * gradient[0] + gradient[1] * gradient[1]);
 
        // Determine the bin index (shift of otb::CONST_PI since atan2 values
        // lies in [-pi, pi]
        unsigned int binIndex = std::floor((otb::CONST_PI + angle) / orientationBinWidth);
 
        if (binIndex == m_NumberOfOrientationBins)
          binIndex = m_NumberOfOrientationBins - 1;
 
        // Compute the angular position
        double angularPosition = std::atan2((double)j, (double)i) - principalOrientation;
 
        // Angle is supposed to lie within [-pi, pi], so we ensure that
        // the compensation did not introduce out-of-range values
        if (angularPosition > otb::CONST_PI)
        {
          angularPosition -= otb::CONST_2PI;
        }
        else if (angularPosition < -otb::CONST_PI)
        {
          angularPosition += otb::CONST_2PI;
        }
 
        // Check if we lie in center bin
        if (currentSquaredRadius < squaredCenterBinRadius)
        {
          centerHistogram[binIndex] += magnitude * gWeight;
        }
        else if (angularPosition > 0)
        {
          if (angularPosition < otb::CONST_PI_2)
          {
            upperRightHistogram[binIndex] += magnitude * gWeight;
          }
          else
          {
            upperLeftHistogram[binIndex] += magnitude * gWeight;
          }
        }
        else
        {
          if (angularPosition > -otb::CONST_PI_2)
          {
            lowerRightHistogram[binIndex] += magnitude * gWeight;
          }
          else
          {
            lowerLeftHistogram[binIndex] += magnitude * gWeight;
          }
        }
 
        // Cumulate values
        globalOrientationHistogram[binIndex] += magnitude * gWeight;
      }
    }
  }
 
  // Build the final output
  hog.push_back(centerHistogram);
  hog.push_back(upperLeftHistogram);
  hog.push_back(upperRightHistogram);
  hog.push_back(lowerRightHistogram);
  hog.push_back(lowerLeftHistogram);
 
  // Normalize each histogram
  for (typename OutputType::iterator oIt = hog.begin(); oIt != hog.end(); ++oIt)
  {
    // Compute L2 norm
    double squaredCumul = 1e-10;
    for (typename std::vector<TOutputPrecision>::const_iterator vIt = oIt->begin(); vIt != oIt->end(); ++vIt)
    {
      squaredCumul += (*vIt) * (*vIt);
    }
    double scale = 1 / std::sqrt(squaredCumul);
    // Apply normalisation factor
    for (typename std::vector<TOutputPrecision>::iterator vIt = oIt->begin(); vIt != oIt->end(); ++vIt)
    {
      (*vIt) *= scale;
    }
  }
 
 
  // Return result
  return hog;
}
 
} // namespace otb
 
#endif