/*=========================================================================
 *
 *  Copyright UMC Utrecht and contributors
 *
 *  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.txt
 *
 *  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.
 *
 *=========================================================================*/
/*=========================================================================

  Program:   Insight Segmentation & Registration Toolkit
  Module:    $RCSfile: itkTransform.h,v $
  Date:      $Date: 2008-06-29 12:58:58 $
  Version:   $Revision: 1.64 $

  Copyright (c) Insight Software Consortium. All rights reserved.
  See ITKCopyright.txt or http://www.itk.org/HTML/Copyright.htm for details.

     This software is distributed WITHOUT ANY WARRANTY; without even
     the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
     PURPOSE.  See the above copyright notices for more information.

=========================================================================*/
#ifndef itkAdvancedTransform_h
#define itkAdvancedTransform_h

#include "itkTransform.h"
#include "itkMatrix.h"
#include "itkFixedArray.h"
#include <cassert>

namespace itk
{

/** \class AdvancedTransform
 * \brief Transform maps points, vectors and covariant vectors from an input
 * space to an output space.
 *
 * This abstract class define the generic interface for a geometrical
 * transformation from one space to another. The class provides methods
 * for mapping points, vectors and covariant vectors from the input space
 * to the output space.
 *
 * Given that transformation are not necessarily invertible, this basic
 * class does not provide the methods for back transformation. Back transform
 * methods are implemented in derived classes where appropriate.
 *
 * \par Registration Framework Support
 * Typically a Transform class has several methods for setting its
 * parameters. For use in the registration framework, the parameters must
 * also be represented by an array of doubles to allow communication
 * with generic optimizers. The Array of transformation parameters is set using
 * the SetParameters() method.
 *
 * Another requirement of the registration framework is the computation
 * of the Jacobian of the transform T. In general, an ImageToImageMetric
 * requires the knowledge of this Jacobian in order to compute the metric
 * derivatives. The Jacobian is a matrix whose element are the partial
 * derivatives of the transformation with respect to the array of parameters
 * mu that defines the transform, evaluated at a point p: dT/dmu(p).
 *
 * If penalty terms are included in the registration, the transforms also
 * need to implement other derivatives of T. Often, penalty terms are functions
 * of the spatial derivatives of T. Therefore, e.g. the SpatialJacobian dT/dx
 * and the SpatialHessian d^2T/dx_idx_j require implementation. The
 * GetValueAndDerivative() requires the d/dmu of those terms. Therefore,
 * we additionally define GetJacobianOfSpatialJacobian() and
 * GetJacobianOfSpatialHessian().
 *
 * \ingroup Transforms
 *
 */
template <class TScalarType, unsigned int NInputDimensions = 3, unsigned int NOutputDimensions = 3>
class ITK_TEMPLATE_EXPORT AdvancedTransform : public Transform<TScalarType, NInputDimensions, NOutputDimensions>
{
public:
  ITK_DISALLOW_COPY_AND_MOVE(AdvancedTransform);

  /** Standard class typedefs. */
  using Self = AdvancedTransform;
  using Superclass = Transform<TScalarType, NInputDimensions, NOutputDimensions>;
  using Pointer = SmartPointer<Self>;
  using ConstPointer = SmartPointer<const Self>;

  /** New method for creating an object using a factory. */
  // itkNewMacro( Self );

  /** Run-time type information (and related methods). */
  itkTypeMacro(AdvancedTransform, Transform);

  /** Dimension of the domain space. */
  itkStaticConstMacro(InputSpaceDimension, unsigned int, NInputDimensions);
  itkStaticConstMacro(OutputSpaceDimension, unsigned int, NOutputDimensions);

  /** Typedefs from the Superclass. */
  using typename Superclass::ScalarType;
  using typename Superclass::ParametersType;
  using typename Superclass::FixedParametersType;
  using typename Superclass::ParametersValueType;
  using typename Superclass::NumberOfParametersType;
  using typename Superclass::DerivativeType;
  using typename Superclass::JacobianType;
  using typename Superclass::InputVectorType;
  using typename Superclass::OutputVectorType;
  using typename Superclass::InputCovariantVectorType;
  using typename Superclass::OutputCovariantVectorType;
  using typename Superclass::InputVnlVectorType;
  using typename Superclass::OutputVnlVectorType;
  using typename Superclass::InputPointType;
  using typename Superclass::OutputPointType;

  using InverseTransformBaseType = typename Superclass::InverseTransformBaseType;
  using typename Superclass::InverseTransformBasePointer;

  /** Transform typedefs for the from Superclass. */
  using TransformType = Transform<TScalarType, NInputDimensions, NOutputDimensions>;
  using TransformTypePointer = typename TransformType::Pointer;
  using TransformTypeConstPointer = typename TransformType::ConstPointer;

  /** Types for the (Spatial)Jacobian/Hessian.
   * Using an itk::FixedArray instead of an std::vector gives a performance
   * gain for the SpatialHessianType.
   */
  using NonZeroJacobianIndicesType = std::vector<unsigned long>;
  using SpatialJacobianType = Matrix<ScalarType, OutputSpaceDimension, InputSpaceDimension>;
  using JacobianOfSpatialJacobianType = std::vector<SpatialJacobianType>;
  // \todo: think about the SpatialHessian type, should be a 3D native type
  using SpatialHessianType =
    FixedArray<Matrix<ScalarType, InputSpaceDimension, InputSpaceDimension>, OutputSpaceDimension>;
  using JacobianOfSpatialHessianType = std::vector<SpatialHessianType>;
  using InternalMatrixType = typename SpatialJacobianType::InternalMatrixType;

  /** Typedef for the moving image gradient type.
   * This type is defined by the B-spline interpolator as
   * typedef CovariantVector< RealType, ImageDimension >
   * As we cannot access this type we simply re-construct it to be identical.
   */
  using MovingImageGradientType = OutputCovariantVectorType;
  using MovingImageGradientValueType = typename MovingImageGradientType::ValueType;

  /** Get the number of nonzero Jacobian indices. By default all. */
  virtual NumberOfParametersType
  GetNumberOfNonZeroJacobianIndices() const;

  /** Whether the advanced transform has nonzero matrices. */
  itkGetConstMacro(HasNonZeroSpatialHessian, bool);
  itkGetConstMacro(HasNonZeroJacobianOfSpatialHessian, bool);

  /** This returns a sparse version of the Jacobian of the transformation.
   *
   * The Jacobian is expressed as a vector of partial derivatives of the
   * transformation components with respect to the parameters \f$\mu\f$ that
   * define the transformation \f$T\f$, evaluated at a point \f$p\f$.
   *
   * \f[
      J=\left[ \begin{array}{cccc}
      \frac{\partial T_{1}}{\partial \mu_{1}}(p) &
      \frac{\partial T_{1}}{\partial \mu_{2}}(p) &
      \cdots &
      \frac{\partial T_{1}}{\partial \mu_{m}}(p) \\
      \frac{\partial T_{2}}{\partial \mu_{1}}(p) &
      \frac{\partial T_{2}}{\partial \mu_{2}}(p) &
      \cdots &
      \frac{\partial T_{2}}{\partial \mu_{m}}(p) \\
      \vdots & \vdots & \ddots & \vdots \\
      \frac{\partial T_{d}}{\partial \mu_{1}}(p) &
      \frac{\partial T_{d}}{\partial \mu_{2}}(p) &
      \cdots &
      \frac{\partial T_{d}}{\partial \mu_{m}}(p)
      \end{array}\right],
   * \f]
   * with \f$m\f$ the number of parameters, i.e. the size of \f$\mu\f$, and \f$d\f$
   * the dimension of the image.
   */
  virtual void
  GetJacobian(const InputPointType &       inputPoint,
              JacobianType &               j,
              NonZeroJacobianIndicesType & nonZeroJacobianIndices) const = 0;

  /** Compute the inner product of the Jacobian with the moving image gradient.
   * The Jacobian is (partially) constructed inside this function, but not returned.
   */
  virtual void
  EvaluateJacobianWithImageGradientProduct(const InputPointType &          inputPoint,
                                           const MovingImageGradientType & movingImageGradient,
                                           DerivativeType &                imageJacobian,
                                           NonZeroJacobianIndicesType &    nonZeroJacobianIndices) const;

  /** Compute the spatial Jacobian of the transformation.
   *
   * The spatial Jacobian is expressed as a vector of partial derivatives of the
   * transformation components with respect to the spatial position \f$x\f$,
   * evaluated at a point \f$p\f$.
   *
   * \f[
      sJ=\left[ \begin{array}{cccc}
      \frac{\partial T_{1}}{\partial x_{1}}(p) &
      \frac{\partial T_{1}}{\partial x_{2}}(p) &
      \cdots &
      \frac{\partial T_{1}}{\partial x_{m}}(p) \\
      \frac{\partial T_{2}}{\partial x_{1}}(p) &
      \frac{\partial T_{2}}{\partial x_{2}}(p) &
      \cdots &
      \frac{\partial T_{2}}{\partial x_{m}}(p) \\
      \vdots & \vdots & \ddots & \vdots \\
      \frac{\partial T_{d}}{\partial x_{1}}(p) &
      \frac{\partial T_{d}}{\partial x_{2}}(p) &
      \cdots &
      \frac{\partial T_{d}}{\partial x_{m}}(p)
      \end{array}\right],
   * \f]
   * with \f$m\f$ the number of parameters, i.e. the size of \f$\mu\f$, and \f$d\f$
   * the dimension of the image.
   */
  virtual void
  GetSpatialJacobian(const InputPointType & inputPoint, SpatialJacobianType & sj) const = 0;

  /** Override some pure virtual ITK4 functions. */
  void
  ComputeJacobianWithRespectToParameters(const InputPointType & itkNotUsed(p),
                                         JacobianType &         itkNotUsed(j)) const override
  {
    itkExceptionMacro("This ITK4 function is currently not used in elastix.");
  }


  /** Compute the spatial Hessian of the transformation.
   *
   * The spatial Hessian is the vector of matrices of partial second order
   * derivatives of the transformation components with respect to the spatial
   * position \f$x\f$, evaluated at a point \f$p\f$.
   *
   * \f[
      sH=\left[ \begin{array}{cc}
      \frac{\partial^2 T_{i}}{\partial x_{1} \partial x_{1}}(p) &
      \frac{\partial^2 T_{i}}{\partial x_{1} \partial x_{2}}(p) \\
      \frac{\partial^2 T_{i}}{\partial x_{1} \partial x_{2}}(p) &
      \frac{\partial^2 T_{i}}{\partial x_{2} \partial x_{2}}(p) \\
      \end{array}\right],
   * \f]
   * with i the i-th component of the transformation.
   */
  virtual void
  GetSpatialHessian(const InputPointType & inputPoint, SpatialHessianType & sh) const = 0;

  /** Compute the Jacobian of the spatial Jacobian of the transformation.
   *
   * The Jacobian of the spatial Jacobian is the derivative of the spatial
   * Jacobian to the transformation parameters \f$\mu\f$, evaluated at
   * a point \f$p\f$.
   */
  virtual void
  GetJacobianOfSpatialJacobian(const InputPointType &          inputPoint,
                               JacobianOfSpatialJacobianType & jsj,
                               NonZeroJacobianIndicesType &    nonZeroJacobianIndices) const = 0;

  /** Compute both the spatial Jacobian and the Jacobian of the
   * spatial Jacobian of the transformation.
   */
  virtual void
  GetJacobianOfSpatialJacobian(const InputPointType &          inputPoint,
                               SpatialJacobianType &           sj,
                               JacobianOfSpatialJacobianType & jsj,
                               NonZeroJacobianIndicesType &    nonZeroJacobianIndices) const = 0;

  /** Compute the Jacobian of the spatial Hessian of the transformation.
   *
   * The Jacobian of the spatial Hessian is the derivative of the spatial
   * Hessian to the transformation parameters \f$\mu\f$, evaluated at
   * a point \f$p\f$.
   */
  virtual void
  GetJacobianOfSpatialHessian(const InputPointType &         inputPoint,
                              JacobianOfSpatialHessianType & jsh,
                              NonZeroJacobianIndicesType &   nonZeroJacobianIndices) const = 0;

  /** Compute both the spatial Hessian and the Jacobian of the
   * spatial Hessian of the transformation.
   */
  virtual void
  GetJacobianOfSpatialHessian(const InputPointType &         inputPoint,
                              SpatialHessianType &           sh,
                              JacobianOfSpatialHessianType & jsh,
                              NonZeroJacobianIndicesType &   nonZeroJacobianIndices) const = 0;

protected:
  AdvancedTransform() = default;

  // Inherit the other (non-default) constructor from itk::Transform.
  using Superclass::Superclass;

  ~AdvancedTransform() override = default;

  bool m_HasNonZeroSpatialHessian{ true };
  bool m_HasNonZeroJacobianOfSpatialHessian{ true };
};

namespace ImplementationDetails
{
/** Multiplies the input matrix and the input vector. */
template <class TScalarType, unsigned int VInputVectorSize>
void
EvaluateInnerProduct(const vnl_matrix<TScalarType> &                        inputMatrix,
                     const CovariantVector<TScalarType, VInputVectorSize> & inputVector,
                     vnl_vector<TScalarType> &                              outputVector)
{
  assert(inputMatrix.rows() == inputVector.size());
  assert(inputMatrix.columns() == outputVector.size());

  auto inputMatrixIterator = inputMatrix.begin();

  outputVector.fill(0.0);

  for (const double inputVectorElement : inputVector)
  {
    for (auto & outputVectorElement : outputVector)
    {
      outputVectorElement += (*inputMatrixIterator) * inputVectorElement;
      ++inputMatrixIterator;
    }
  }
}
} // namespace ImplementationDetails

} // end namespace itk

#ifndef ITK_MANUAL_INSTANTIATION
#  include "itkAdvancedTransform.hxx"
#endif

#endif