libvigraimpex-dev 1.10.0+git20160211.167be93+dfsg-2+b5 / usr / include / vigra / correlation.hxx

/************************************************************************/
/*                                                                      */
/*               Copyright 2007-2014 by Benjamin Seppke                 */
/*                                                                      */
/*    This file is part of the VIGRA computer vision library.           */
/*    The VIGRA Website is                                              */
/*        http://hci.iwr.uni-heidelberg.de/vigra/                       */
/*    Please direct questions, bug reports, and contributions to        */
/*        ullrich.koethe@iwr.uni-heidelberg.de    or                    */
/*        vigra@informatik.uni-hamburg.de                               */
/*                                                                      */
/*    Permission is hereby granted, free of charge, to any person       */
/*    obtaining a copy of this software and associated documentation    */
/*    files (the "Software"), to deal in the Software without           */
/*    restriction, including without limitation the rights to use,      */
/*    copy, modify, merge, publish, distribute, sublicense, and/or      */
/*    sell copies of the Software, and to permit persons to whom the    */
/*    Software is furnished to do so, subject to the following          */
/*    conditions:                                                       */
/*                                                                      */
/*    The above copyright notice and this permission notice shall be    */
/*    included in all copies or substantial portions of the             */
/*    Software.                                                         */
/*                                                                      */
/*    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND    */
/*    EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES   */
/*    OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND          */
/*    NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT       */
/*    HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,      */
/*    WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING      */
/*    FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR     */
/*    OTHER DEALINGS IN THE SOFTWARE.                                   */
/*                                                                      */
/************************************************************************/

#ifndef VIGRA_CORRELATION_HXX
#define VIGRA_CORRELATION_HXX

#include "stdimage.hxx"
#include "inspectimage.hxx"
#include "functorexpression.hxx"
#include "multi_math.hxx"
#include "multi_fft.hxx"
#include "integral_image.hxx"

// "slow" correlation algorithms are performed using the windowing filter env.
#include "applywindowfunction.hxx"

namespace vigra
{

/** \addtogroup Correlation Image Correlation

    Fast (FFT-based) and slow algorithms to estimate the correlation between images.
*/
//@{

/********************************************************/
/*                                                      */
/*    Fast cross correlation of an image w.r.t a mask   */
/*                                                      */
/********************************************************/

/** \brief This function performes a fast cross-correlation

 This function performes a fast cross-correlation using the Fast Fourier Transform and
 the dependency of the convolution and the correlation in Fourier space.

 The input pixel type <tt>T1</tt> must be a \ref LinearSpace "linear space" over
 the window functions' value_type <tt>T</tt>, i.e. addition of source values, multiplication with functions' values,
 and NumericTraits must be defined. The mask's value_type must be an \ref AlgebraicField "algebraic field",
 i.e. the arithmetic operations (+, -, *, /) and NumericTraits must be defined.

 By default, the borders are filled with zeros. Use the clearBorders switch to change that behavior if you need to.

 <b> Declarations:</b>

 pass 2D array views:
 \code
 namespace vigra {
   template <class T1, class S1,
            class T2, class S2,
            class T3, class S3,
            unsigned int N>
   void
   fastCrossCorrelation(MultiArrayView<N, T1, S1> const & in,
                        MultiArrayView<N, T2, S2> const & mask,
                        MultiArrayView<N, T3, S3> out,
                        bool clearBorders=true);

 }
 \endcode


 <b> Usage:</b>

 <b>\#include</b> \<vigra/correlation.hxx\><br/>
 Namespace: vigra

 \code
 unsigned int m_w=51, m_h=51;
 unsigned int w=1000, h=1000;
 MultiArray<2, float> mask(m_w,m_h), src(w,h), dest(w,h);
 ...

 //compute fast cross correlation of mask and image -> dest
 fastCrossCorrelation(mask, src, dest);
 \endcode

 <b> Preconditions:</b>

 The image must be larger than the size of the mask.
 */
doxygen_overloaded_function(template <...> void fastCrossCorrelation)

template <class T1, class S1,
class T2, class S2,
class T3, class S3,
unsigned int N>
inline void fastCrossCorrelation(MultiArrayView<N, T1, S1> const & in,
                                 MultiArrayView<N, T2, S2> const & mask,
                                 MultiArrayView<N, T3, S3> out,
                                 bool clearBorders=true)
{
    vigra_precondition(in.shape() == out.shape(),
                       "vigra::fastCrossCorrelation(): shape mismatch between input and output.");
    correlateFFT(in, mask, out);

    if(clearBorders)
    {
        typedef typename MultiArrayShape<N>::type Shape;
        Shape maskRadius(floor(mask.shape()/2));
        initMultiArrayBorder(out, maskRadius, maskRadius, T3());
    }
}

namespace detail {

template<class T1, class S1>
inline double integralMultiArrayWindowMean(MultiArrayView<1, T1, S1> const & in,
                                           typename MultiArrayView<1, T1, S1>::difference_type const & left,
                                           typename MultiArrayView<1, T1, S1>::difference_type const & right)
{
    return in[right]-in[left];
}

template<class T1, class S1>
inline double integralMultiArrayWindowMean(MultiArrayView<2, T1, S1> const & in,
                                           typename MultiArrayView<2, T1, S1>::difference_type const & ul,
                                           typename MultiArrayView<2, T1, S1>::difference_type const & lr)
{
    return in[lr] - in(lr[0],ul[1]) - in(ul[0],lr[1]) + in[ul];
}

template<class T1, class S1>
inline double integralMultiArrayWindowMean(MultiArrayView<3, T1, S1> const & in,
                                           typename MultiArrayView<3, T1, S1>::difference_type const & ul,
                                           typename MultiArrayView<3, T1, S1>::difference_type const & lr)
{
    return (in[lr] - in(lr[0],ul[1],lr[2]) - in(ul[0],lr[1],lr[2]) + in(ul[0],ul[1],lr[2]))
         - (in(lr[0],lr[1],ul[2]) - in(lr[0],ul[1],ul[2]) - in(ul[0],lr[1],ul[2]) + in[ul]);
}

} // namespace detail

/********************************************************/
/*                                                      */
/*  Fast normalized cross correlation of mask to image  */
/*                                                      */
/********************************************************/

/** \brief This function performes a fast normalized cross-correlation

    To compute the normalized cross correlation in a fast way, it is using the Fast Fourier Transform     and sum-image look-up-tables as it is suggested by J.P.Lewis (1995):
    <i>"Fast Normalized Cross-Correlation"</i>.

    The input pixel type <tt>T1</tt> must be a \ref LinearSpace "linear space" over
    the window functions' value_type <tt>T</tt>, i.e. addition of source values, multiplication with functions' values,
    and NumericTraits must be defined. The mask's value_type must be an \ref AlgebraicField "algebraic field",
    i.e. the arithmetic operations (+, -, *, /) and NumericTraits must be defined.

    By default, the borders are filled with zeros. Use the clearBorders switch to change that behavior if you need to.

    <b> Declarations:</b>

    pass 2D array views:
    \code
    namespace vigra {
        template <class T1, class S1,
                  class T2, class S2,
                  class T3, class S3>
        void
        fastNormalizedCrossCorrelation(MultiArrayView<2, T1, S1> const & in,
                                       MultiArrayView<2, T2, S2> const & mask,
                                       MultiArrayView<2, T3, S3> out,
                                       bool clearBorders=true);

    }
    \endcode


    <b> Usage:</b>

    <b>\#include</b> \<vigra/correlation.hxx\><br/>
    Namespace: vigra

    \code
    unsigned int m_w=51, m_h=51;
    unsigned int w=1000, h=1000;
    MultiArray<2, float> mask(m_w,m_h), src(w,h), dest(w,h);
    ...

    //compute normalized cross correlation of mask and image -> dest
    fastNormalizedCrossCorrelation(mask, src, dest);
    \endcode

    <b> Preconditions:</b>

    The image must be larger than the size of the mask.
*/
doxygen_overloaded_function(template <...> void fastNormalizedCrossCorrelation)

template <class T1, class S1,
          class T2, class S2,
          class T3, class S3,
          unsigned int N>
inline void fastNormalizedCrossCorrelation(MultiArrayView<N, T1, S1> const & in,
                                           MultiArrayView<N, T2, S2> const & mask,
                                           MultiArrayView<N, T3, S3> out,
                                           bool clearBorders=true)
{
    using namespace vigra::multi_math;
    typedef typename MultiArrayShape<N>::type Shape;

    vigra_precondition(in.shape() == out.shape(),
                       "vigra::fastNormalizedCrossCorrelation(): shape mismatch between input and output.");

    vigra_precondition(N>0 && N<=3,
                       "vigra::fastNormalizedCrossCorrelation(): only implemented for arrays of 1, 2 or 3 dimensions.");

    for(unsigned int dim=0; dim<N; dim++)
    {
        vigra_precondition(mask.shape()[dim] % 2 == 1, "vigra::fastNormalizedCrossCorrelation(): Mask width has to be of odd size!");
        vigra_precondition(in.shape()[dim] >= mask.shape()[dim] , "vigra::fastNormalizedCrossCorrelation(): Mask is larger than image!");
    }

    //find mask mean and variance
    double mask_mean = 0.0,
           mask_var  = 0.0,
           mask_size = prod(mask.shape());
    mask.meanVariance(&mask_mean, &mask_var);

    //calculate the fix part of the denumerator a.k.a. the mask std. deviation
    double fix_denumerator = mask_size*sqrt(mask_var);

    if(fix_denumerator == 0)
    {
        out = 0;
    }
    else
    {
        //pre-normalize the mask
        MultiArray<N, double> norm_mask(mask.shape());
        norm_mask = mask;
        norm_mask -= mask_mean;

        //calculate (semi-normalized) numerator:
        MultiArray<N, double> corr_result(in.shape());

        corr_result=in;
        fastCrossCorrelation(corr_result, norm_mask, corr_result, clearBorders);


        //Create fast sum tables for the variable denumerator
        MultiArray<N, double> sum_table(in.shape()+1),
                              sum_table2(in.shape()+1);

        typename MultiArray<N, double>::difference_type zero_diff;

        // now finally fill the sum tables
        // keep a zero line/coloum at the beginning to avoid border computations and conditionals
        integralMultiArray(in,sum_table.subarray(zero_diff+1, in.shape()+1));
        integralMultiArraySquared(in, sum_table2.subarray(zero_diff+1, in.shape()+1));

        MultiCoordinateIterator<N> i(in.shape()-mask.shape()+1), end = i.getEndIterator();

        Shape maskRadius(floor(mask.shape()/2));
        for(; i != end; ++i)
        {
            //calculate e(window) and e(window^2)
            double window_mean         = detail::integralMultiArrayWindowMean(sum_table,  *i, *i+mask.shape()),
                   window_squared_mean = detail::integralMultiArrayWindowMean(sum_table2, *i, *i+mask.shape()),
                   var_denumerator = sqrt(mask_size*window_squared_mean - window_mean*window_mean);

            //calculate overall result
            if(var_denumerator == 0)
            {
                out[*i+maskRadius] = 0;
            }
            else
            {
                out[*i+maskRadius] = mask_size*corr_result[*i+maskRadius]/(var_denumerator*fix_denumerator);
            }
        }
    }
}

template<class MaskIterator, class MaskAccessor>
class CorrelationFunctor
{
public:
    CorrelationFunctor(MaskIterator m_ul, MaskIterator m_lr, MaskAccessor m_acc)
    : m_mask_ul(m_ul),
      m_mask_lr(m_lr),
      m_mask_acc(m_acc)
    {
    }

    CorrelationFunctor(triple<MaskIterator,MaskIterator,MaskAccessor> m)
    : m_mask_ul(m.first),
      m_mask_lr(m.second),
      m_mask_acc(m.third)
    {
    }

    template <class SrcIterator,  class SrcAccessor, class DestIterator,  class DestAccessor>
    void operator()(SrcIterator s, SrcAccessor s_acc, DestIterator d, DestAccessor d_acc) const
    {
        using namespace vigra;

        SrcIterator s_ul = s - windowShape()/2,
                    s_lr = s + windowShape()/2+Diff2D(1,1);

        //find img2 mean
        FindAverage<double> average;
        inspectImage(srcIterRange(s_ul, s_lr, s_acc), average);

        SrcIterator ys = s_ul;
        SrcIterator xs = ys;

        MaskIterator ym = m_mask_ul;
        MaskIterator xm = ym;

        double  res=0;

        for( ; ys.y != s_lr.y; ys.y++, ym.y++)
        {
            for(xs = ys, xm = ym; xs.x != s_lr.x; xs.x++, xm.x++)
            {
                res += m_mask_acc(xm)*s_acc(xs);
            }
        }
        d_acc.set(res,d);
    }

    Diff2D windowShape() const
    {
        return m_mask_lr - m_mask_ul;
    }

private:
    MaskIterator m_mask_ul;
    MaskIterator m_mask_lr;
    MaskAccessor m_mask_acc;
};

/********************************************************/
/*                                                      */
/*          Cross correlation of mask to image          */
/*                                                      */
/********************************************************/

/** \brief This function performes a (slow) cross-correlation

    This function performes a (slow) cross-correlation using the window function environment
    and comparison of the mask with the underlying image part for each pixel. This may
    however be faster for very few comparisons.

    The input pixel type <tt>T1</tt> must be a \ref LinearSpace "linear space" over
    the window functions' value_type <tt>T</tt>, i.e. addition of source values, multiplication with functions' values,
    and NumericTraits must be defined. The mask's value_type must be an \ref AlgebraicField "algebraic field",
    i.e. the arithmetic operations (+, -, *, /) and NumericTraits must be defined.

    <b> Declarations:</b>

    pass 2D array views:
    \code
    namespace vigra {
        template <class T1, class S1,
                  class T2, class S2,
                  class T3, class S3>
        void
        crossCorrelation(MultiArrayView<2, T1, S1> const & in,
                         MultiArrayView<2, T2, S2> const & mask,
                         MultiArrayView<2, T3, S3> out);

    }
    \endcode

    \deprecatedAPI{crossCorrelation}
    pass \ref ImageIterators and \ref DataAccessors :
    \code
    namespace vigra {
        template <class SrcIterator, class SrcAccessor,
                  class MaskIterator, class MaskAccessor,
                  class DestIterator, class DestAccessor>
        void crossCorrelation(SrcIterator s_ul, SrcIterator s_lr, SrcAccessor s_acc,
                              MaskIterator m_ul, MaskIterator m_lr, MaskAccessor m_acc,
                              DestIterator d_ul, DestAccessor d_acc);
    }
    \endcode
    use argument objects in conjunction with \ref ArgumentObjectFactories :
    \code
    namespace vigra {
        template <class SrcIterator, class SrcAccessor,
                  class MaskIterator, class MaskAccessor,
                  class DestIterator, class DestAccessor>
        void
        crossCorrelation(triple<MaskIterator, MaskIterator, MaskAccessor> src,
                         triple<SrcIterator, SrcIterator, SrcAccessor> mask,
                         pair<DestIterator, DestAccessor> dest);
    }
    \endcode
    \deprecatedEnd

    <b> Usage:</b>

    <b>\#include</b> \<vigra/correlation.hxx\><br/>
    Namespace: vigra

    \code
    unsigned int m_w=51, m_h=51;
    unsigned int w=1000, h=1000;
    MultiArray<2, float> mask(m_w,m_h), src(w,h), dest(w,h);
    ...

    //compute (slow) cross correlation of mask and image -> dest
    crossCorrelation(mask, src, dest);
    \endcode

    <b> Preconditions:</b>

    The image must be larger than the size of the mask.
*/
doxygen_overloaded_function(template <...> void crossCorrelation)

template <class SrcIterator, class SrcAccessor,
          class MaskIterator, class MaskAccessor,
          class DestIterator, class DestAccessor>
inline void crossCorrelation(SrcIterator s_ul,  SrcIterator s_lr,   SrcAccessor s_acc,
                             MaskIterator m_ul,  MaskIterator m_lr,   MaskAccessor m_acc,
                             DestIterator d_ul, DestAccessor d_acc,
                             BorderTreatmentMode border = BORDER_TREATMENT_AVOID)
{
    CorrelationFunctor<MaskIterator, MaskAccessor> func(m_ul, m_lr, m_acc);
    applyWindowFunction(s_ul, s_lr, s_acc, d_ul, d_acc, func, border);
}

template <class SrcIterator, class SrcAccessor,
          class MaskIterator, class MaskAccessor,
          class DestIterator, class DestAccessor>
inline void crossCorrelation(triple<SrcIterator, SrcIterator, SrcAccessor> src,
                             triple<MaskIterator, MaskIterator, MaskAccessor> mask,
                             pair<DestIterator, DestAccessor> dest,
                             BorderTreatmentMode border = BORDER_TREATMENT_AVOID)
{
    crossCorrelation(src.first, src.second, src.third,
                     mask.first, mask.second, mask.third,
                     dest.first, dest.second,
                     border);
}

template <class T1, class S1,
          class T2, class S2,
          class T3, class S3>
inline void crossCorrelation(MultiArrayView<2, T1, S1> const & in,
                             MultiArrayView<2, T2, S2> const & mask,
                             MultiArrayView<2, T3, S3> out)
{
    vigra_precondition(in.shape() == out.shape(),
                        "vigra::crossCorrelation(): shape mismatch between input and output.");
    crossCorrelation(srcImageRange(in),
                     srcImageRange(mask),
                     destImage(out));
}

template<class MaskIterator, class MaskAccessor>
class NormalizedCorrelationFunctor
{
public:

    NormalizedCorrelationFunctor(MaskIterator m_ul, MaskIterator m_lr, MaskAccessor m_acc)
    : m_mask_ul(m_ul),
      m_mask_lr(m_lr),
      m_mask_acc(m_acc),
      m_s11(0.0),
      m_avg1(0.0)
    {
        init_s11();
    }

    NormalizedCorrelationFunctor(triple<MaskIterator,MaskIterator,MaskAccessor> mask)
    : m_mask_ul(mask.first),
      m_mask_lr(mask.second),
      m_mask_acc(mask.third),
      m_s11(0.0),
      m_avg1(0.0)
    {
        init_s11();
    }

    template <class SrcIterator,  class SrcAccessor, class DestIterator,  class DestAccessor>
    void operator()(SrcIterator s, SrcAccessor s_acc, DestIterator d, DestAccessor d_acc) const
    {
        using namespace vigra;

        SrcIterator s_ul = s - windowShape()/2,
        s_lr = s + windowShape()/2+Diff2D(1,1);

        if(m_s11 == 0.0)
        {
            d_acc.set(0,d);
        }
        else
        {
            //find img2 mean
            FindAverage<double> average;
            inspectImage(srcIterRange(s_ul, s_lr, s_acc), average);

            SrcIterator ys = s_ul;
            SrcIterator xs = ys;

            MaskIterator ym = m_mask_ul;
            MaskIterator xm = ym;

            double  s1=0,s2=0, s12=0, s22=0;

            for( ; ys.y != s_lr.y; ys.y++, ym.y++)
            {
                for(xs = ys, xm = ym; xs.x != s_lr.x; xs.x++, xm.x++)
                {
                    s1 = m_mask_acc(xm);
                    s2 = s_acc(xs);
                    s12 += (s1-m_avg1)*(s2-average());
                    s22 += (s2-average())*(s2-average());
                }
            }
            if(s22 == 0.0)
            {
                d_acc.set(0,d);
            }
            else
            {
                d_acc.set(s12/sqrt(m_s11*s22),d);
            }
        }
    }

    Diff2D windowShape() const
    {
        return m_mask_lr - m_mask_ul;
    }

private:
    void init_s11()
    {
        //find mask mean
        FindAverage<double> average;
        inspectImage(srcIterRange(m_mask_ul, m_mask_lr, m_mask_acc), average);

        MaskIterator ym = m_mask_ul;
        MaskIterator xm = ym;

        m_avg1 = average();

        for( ; ym.y != m_mask_lr.y; ym.y++)
        {
            for(xm = ym; xm.x != m_mask_lr.x; xm.x++)
            {
                m_s11 += (m_mask_acc(xm)-m_avg1)*(m_mask_acc(xm)-m_avg1);
            }
        }
    }

    MaskIterator m_mask_ul;
    MaskIterator m_mask_lr;
    MaskAccessor m_mask_acc;

    double m_s11;
    double m_avg1;
};

/********************************************************/
/*                                                      */
/*     Normalized Cross correlation of mask to image    */
/*                                                      */
/********************************************************/

/** \brief This function performes a (slow) normalized cross-correlation

    This function performes a (slow) normalized cross-correlation using the window function
    environment and comparison of the mask with the underlying image part for each pixel.
    This may however be faster for very few comparisons.

    The input pixel type <tt>T1</tt> must be a \ref LinearSpace "linear space" over
    the window functions' value_type <tt>T</tt>, i.e. addition of source values, multiplication with functions' values,
    and NumericTraits must be defined. The mask's value_type must be an \ref AlgebraicField "algebraic field",
    i.e. the arithmetic operations (+, -, *, /) and NumericTraits must be defined.

    <b> Declarations:</b>

    pass 2D array views:
    \code
    namespace vigra {
        template <class T1, class S1,
                  class T2, class S2,
                  class T3, class S3>
        void
        normalizedCrossCorrelation(MultiArrayView<2, T1, S1> const & in,
                                   MultiArrayView<2, T2, S2> const & mask,
                                   MultiArrayView<2, T3, S3> out);

    }
    \endcode

    \deprecatedAPI{normalizedCrossCorrelation}
    pass \ref ImageIterators and \ref DataAccessors :
    \code
    namespace vigra {
        template <class SrcIterator, class SrcAccessor,
                  class MaskIterator, class MaskAccessor,
                  class DestIterator, class DestAccessor>
        void normalizedCrossCorrelation(SrcIterator s_ul, SrcIterator s_lr, SrcAccessor s_acc,
                                        MaskIterator m_ul, MaskIterator m_lr, MaskAccessor m_acc,
                                        DestIterator d_ul, DestAccessor d_acc);
    }
    \endcode
    use argument objects in conjunction with \ref ArgumentObjectFactories :
    \code
    namespace vigra {
        template <class SrcIterator, class SrcAccessor,
                  class MaskIterator, class MaskAccessor,
                  class DestIterator, class DestAccessor>
        void
        normalizedCrossCorrelation(triple<SrcIterator, SrcIterator, SrcAccessor> src,
                                   triple<MaskIterator, MaskIterator, MaskAccessor> mask,
                                   pair<DestIterator, DestAccessor> dest);
    }
    \endcode
    \deprecatedEnd

    <b> Usage:</b>

    <b>\#include</b> \<vigra/correlation.hxx\><br/>
    Namespace: vigra

    \code
    unsigned int m_w=51, m_h=51;
    unsigned int w=1000, h=1000;
    MultiArray<2, float> mask(m_w,m_h), src(w,h), dest(w,h);
    ...

    //compute (slow) normalized cross correlation of mask and image -> dest
    normalizedCrossCorrelation(mask, src, dest);
    \endcode

    <b> Preconditions:</b>

    The image must be larger than the size of the mask.
*/
doxygen_overloaded_function(template <...> void normalizedCrossCorrelation)


template <class SrcIterator, class SrcAccessor,
          class MaskIterator, class MaskAccessor,
          class DestIterator, class DestAccessor>
inline void normalizedCrossCorrelation(SrcIterator s_ul,  SrcIterator s_lr,   SrcAccessor s_acc,
                                       MaskIterator m_ul,  MaskIterator m_lr,   MaskAccessor m_acc,
                                       DestIterator d_ul, DestAccessor d_acc,
                                       BorderTreatmentMode border = BORDER_TREATMENT_AVOID)
{
    NormalizedCorrelationFunctor<MaskIterator, MaskAccessor> func(m_ul, m_lr, m_acc);
    applyWindowFunction(s_ul, s_lr, s_acc, d_ul, d_acc, func, border);
}

template <class SrcIterator, class SrcAccessor,
          class MaskIterator, class MaskAccessor,
          class DestIterator, class DestAccessor>
inline void normalizedCrossCorrelation(triple<SrcIterator, SrcIterator, SrcAccessor> src,
                                       triple<MaskIterator, MaskIterator, MaskAccessor> mask,
                                       pair<DestIterator, DestAccessor> dest,
                                       BorderTreatmentMode border = BORDER_TREATMENT_AVOID)
{
    normalizedCrossCorrelation(src.first, src.second, src.third,
                               mask.first, mask.second, mask.third,
                               dest.first, dest.second,
                               border);
}

template <class T1, class S1,
          class T2, class S2,
          class T3, class S3>
inline void normalizedCrossCorrelation(MultiArrayView<2, T1, S1> const & in,
                                       MultiArrayView<2, T2, S2> const & mask,
                                       MultiArrayView<2, T3, S3> out)
{
    vigra_precondition(in.shape() == out.shape(),
                        "vigra::normalizedCrossCorrelation(): shape mismatch between input and output.");
    normalizedCrossCorrelation(srcImageRange(in),
                               srcImageRange(mask),
                               destImage(out));
}

//@}

} //end of namespace vigra

#endif //VIGRA_CORRELATION_HXX