977 lines
33 KiB
C++
977 lines
33 KiB
C++
// ____ ______ __
|
|
// / __ \ / ____// /
|
|
// / /_/ // / / /
|
|
// / ____// /___ / /___ PixInsight Class Library
|
|
// /_/ \____//_____/ PCL 2.4.23
|
|
// ----------------------------------------------------------------------------
|
|
// pcl/LanczosInterpolation.h - Released 2022-03-12T18:59:29Z
|
|
// ----------------------------------------------------------------------------
|
|
// This file is part of the PixInsight Class Library (PCL).
|
|
// PCL is a multiplatform C++ framework for development of PixInsight modules.
|
|
//
|
|
// Copyright (c) 2003-2022 Pleiades Astrophoto S.L. All Rights Reserved.
|
|
//
|
|
// Redistribution and use in both source and binary forms, with or without
|
|
// modification, is permitted provided that the following conditions are met:
|
|
//
|
|
// 1. All redistributions of source code must retain the above copyright
|
|
// notice, this list of conditions and the following disclaimer.
|
|
//
|
|
// 2. All redistributions in binary form must reproduce the above copyright
|
|
// notice, this list of conditions and the following disclaimer in the
|
|
// documentation and/or other materials provided with the distribution.
|
|
//
|
|
// 3. Neither the names "PixInsight" and "Pleiades Astrophoto", nor the names
|
|
// of their contributors, may be used to endorse or promote products derived
|
|
// from this software without specific prior written permission. For written
|
|
// permission, please contact info@pixinsight.com.
|
|
//
|
|
// 4. All products derived from this software, in any form whatsoever, must
|
|
// reproduce the following acknowledgment in the end-user documentation
|
|
// and/or other materials provided with the product:
|
|
//
|
|
// "This product is based on software from the PixInsight project, developed
|
|
// by Pleiades Astrophoto and its contributors (https://pixinsight.com/)."
|
|
//
|
|
// Alternatively, if that is where third-party acknowledgments normally
|
|
// appear, this acknowledgment must be reproduced in the product itself.
|
|
//
|
|
// THIS SOFTWARE IS PROVIDED BY PLEIADES ASTROPHOTO AND ITS CONTRIBUTORS
|
|
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
|
|
// TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
|
// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL PLEIADES ASTROPHOTO OR ITS
|
|
// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
|
// EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, BUSINESS
|
|
// INTERRUPTION; PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; AND LOSS OF USE,
|
|
// DATA OR PROFITS) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
|
|
// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
|
|
// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
|
// POSSIBILITY OF SUCH DAMAGE.
|
|
// ----------------------------------------------------------------------------
|
|
|
|
#ifndef __PCL_LanczosInterpolation_h
|
|
#define __PCL_LanczosInterpolation_h
|
|
|
|
/// \file pcl/LanczosInterpolation.h
|
|
|
|
#include <pcl/Defs.h>
|
|
#include <pcl/Diagnostics.h>
|
|
|
|
#include <pcl/BidimensionalInterpolation.h>
|
|
#include <pcl/Math.h>
|
|
#include <pcl/Utility.h>
|
|
#include <pcl/Vector.h>
|
|
|
|
namespace pcl
|
|
{
|
|
|
|
// ----------------------------------------------------------------------------
|
|
|
|
#define m_width this->m_width
|
|
#define m_height this->m_height
|
|
#define m_fillBorder this->m_fillBorder
|
|
#define m_fillValue this->m_fillValue
|
|
#define m_data this->m_data
|
|
|
|
// ----------------------------------------------------------------------------
|
|
|
|
/*
|
|
* Default clamping threshold for Lanczos interpolation. This value has been
|
|
* selected as the best trade-off for a large set of test linear images.
|
|
*/
|
|
#ifndef __PCL_LANCZOS_CLAMPING_THRESHOLD
|
|
#define __PCL_LANCZOS_CLAMPING_THRESHOLD 0.3F
|
|
#endif
|
|
|
|
// ----------------------------------------------------------------------------
|
|
|
|
/*
|
|
* Floating point and integer LUT-based interpolations for 3rd, 4th and 5th
|
|
* order Lanczos functions. LUTs are initialized automatically on-demand by
|
|
* thread-safe internal routines.
|
|
*
|
|
* Real Lanczos LUTs are accurate to +/- 1e-7 DN
|
|
* Integer Lanczos LUTs are accurate to +/- 1 16-bit DN
|
|
*/
|
|
#define __PCL_LANCZOS_LUT_REAL_RESOLUTION 4096
|
|
const double** PCL_FUNC PCL_InitializeLanczosRealLUT( int );
|
|
#define __PCL_LANCZOS_LUT_INT_RESOLUTION 65535
|
|
const float* PCL_FUNC PCL_InitializeLanczosIntLUT( int );
|
|
|
|
// ----------------------------------------------------------------------------
|
|
|
|
#define PCL_LANCZOS_ACC() \
|
|
if ( s < 0 ) \
|
|
sn -= s, wn -= L; \
|
|
else \
|
|
sp += s, wp += L;
|
|
|
|
/*!
|
|
* \class LanczosInterpolation
|
|
* \brief Two dimensional Lanczos interpolation algorithm.
|
|
*
|
|
* This class uses Lanczos filters to interpolate pixel values at arbitrary
|
|
* coordinates within a two-dimensional data matrix. A one-dimensional Lanczos
|
|
* filter of order \e n is defined by the following equations:
|
|
*
|
|
* <pre>
|
|
* L(x;n) = sinc(x)*sinc(x/n) for |x| < n
|
|
* L(x;n) = 0 for |x| >= n
|
|
* </pre>
|
|
*
|
|
* where sinc() is the normalized sinc function:
|
|
*
|
|
* <pre>
|
|
* sinc(x;n) = 1 for x = 0
|
|
* sinc(x;n) = sin(pi*x)/(pi*x) for x != 0
|
|
* </pre>
|
|
*
|
|
* The Lanczos function has alternating positive and negative lobes toward
|
|
* positive and negative infinity. The order \e n defines the number of lobes
|
|
* preserved in the interpolation filter function: n=1 only includes the
|
|
* central, positive lobe; n=2 includes the first two lobes (one positive and
|
|
* one negative), and so on. The default filter order is three.
|
|
*
|
|
* Lanczos interpolation has excellent detail preservation performance with
|
|
* minimal generation of aliasing patterns for noisy data. Its main drawback is
|
|
* generation of strong undershoot (aka ringing) artifacts when negative
|
|
* function lobes fall over bright pixels and edges. This usually happens with
|
|
* linear data. In the current PCL implementation we have included a clamping
|
|
* mechanism that prevents negative interpolated values and ringing problems
|
|
* for most images.
|
|
*
|
|
* \sa BidimensionalInterpolation, NearestNeighborInterpolation,
|
|
* BilinearInterpolation, BicubicSplineInterpolation,
|
|
* BicubicBSplineInterpolation, BicubicFilterInterpolation,
|
|
* Lanczos3LUTInterpolation, Lanczos4LUTInterpolation, Lanczos5LUTInterpolation
|
|
*/
|
|
template <typename T>
|
|
class PCL_CLASS LanczosInterpolation : public BidimensionalInterpolation<T>
|
|
{
|
|
private:
|
|
|
|
struct Default
|
|
{
|
|
template <typename _T> static bool UseLUT( _T* ) { return false; }
|
|
static bool UseLUT( uint8* ) { return true; }
|
|
static bool UseLUT( int8* ) { return true; }
|
|
static bool UseLUT( uint16* ) { return true; }
|
|
static bool UseLUT( int16* ) { return true; }
|
|
static bool UseLUT( float* ) { return true; }
|
|
};
|
|
|
|
public:
|
|
|
|
/*!
|
|
* Constructs a %LanczosInterpolation instance.
|
|
*
|
|
* \param n Filter order (n >= 1). The Lanczos filter interpolates
|
|
* from the nearest (2*n)^2 mapped source pixels for each
|
|
* interpolation point. The default filter order is 3, so the
|
|
* interpolation uses a neighborhood of 36 pixels by default.
|
|
*
|
|
* \param clamp Clamping threshold. Clamping is applied to fix undershoot
|
|
* (aka ringing) artifacts. A value of this parameter within
|
|
* the [0,1] range enables clamping: The lower the clamping
|
|
* threshold, the more aggressive deringing effect is
|
|
* achieved. A negative threshold value disables the clamping
|
|
* feature. The default value is 0.3. For more information,
|
|
* refer to the documentation for the
|
|
* SetClampingThreshold( float ) member function.
|
|
*
|
|
* \param useLUT If true, the interpolation will use a precomputed LUT of
|
|
* function values at discrete intervals. This greatly
|
|
* improves performance, increasing execution speed by about
|
|
* a factor of 2. In current PCL versions, the Lanczos
|
|
* functions are sampled at 0.00025 px resolution, which
|
|
* provides an interpolation accuracy of about 1.0e-07. This
|
|
* is valid for interpolation of 32-bit floating point data,
|
|
* but can be insufficient for 32-bit integers and double
|
|
* precision, depending on the application. If this parameter
|
|
* is false, the interpolation will compute actual function
|
|
* values for each interpolation point. This parameter is
|
|
* true by default for the uint8, int8, uint16, int16, and
|
|
* float template specializations; false by default for other
|
|
* types.
|
|
*/
|
|
LanczosInterpolation( int n = 3, float clamp = __PCL_LANCZOS_CLAMPING_THRESHOLD, bool useLUT = Default::UseLUT( (T*)0 ) )
|
|
: m_n( Max( 1, n ) )
|
|
, m_lut( useLUT ? PCL_InitializeLanczosRealLUT( m_n ) : nullptr )
|
|
, m_Lx( 2*m_n )
|
|
, m_clampTh( Range( clamp, 0.0F, 1.0F ) )
|
|
, m_clampThInv( 1 - m_clampTh )
|
|
, m_clamp( clamp >= 0 )
|
|
{
|
|
PCL_PRECONDITION( n >= 1 )
|
|
PCL_PRECONDITION( clamp < 0 || 0 <= clamp && clamp <= 1 )
|
|
}
|
|
|
|
/*!
|
|
* Copy constructor.
|
|
*/
|
|
LanczosInterpolation( const LanczosInterpolation& ) = default;
|
|
|
|
/*!
|
|
* Virtual destructor.
|
|
*/
|
|
virtual ~LanczosInterpolation()
|
|
{
|
|
}
|
|
|
|
/*!
|
|
* Interpolated value at \a {x,y} location.
|
|
*
|
|
* \param x,y %Coordinates of the interpolation point (horizontal,
|
|
* vertical).
|
|
*/
|
|
double operator()( double x, double y ) const override
|
|
{
|
|
PCL_PRECONDITION( m_data != nullptr )
|
|
PCL_PRECONDITION( m_width > 0 && m_height > 0 )
|
|
PCL_PRECONDITION( x >= 0 && x < m_width )
|
|
PCL_PRECONDITION( y >= 0 && y < m_height )
|
|
PCL_CHECK( m_n >= 1 )
|
|
|
|
// Central grid coordinates
|
|
int x0 = Range( TruncInt( x ), 0, m_width-1 );
|
|
int y0 = Range( TruncInt( y ), 0, m_height-1 );
|
|
|
|
double sp = 0; // positive filter values
|
|
double sn = 0; // negative filter values
|
|
double wp = 0; // positive filter weight
|
|
double wn = 0; // negative filter weight
|
|
int i; // row index
|
|
|
|
if ( m_lut != nullptr )
|
|
{
|
|
// Discrete interpolation increments
|
|
int dx = TruncInt( __PCL_LANCZOS_LUT_REAL_RESOLUTION*(x - x0) );
|
|
int dy = TruncInt( __PCL_LANCZOS_LUT_REAL_RESOLUTION*(y - y0) );
|
|
|
|
// Precalculate horizontal filter values
|
|
for ( int j = -m_n + 1, k = 0; j <= m_n; ++j, ++k )
|
|
m_Lx[k] = m_lut[k][dx];
|
|
|
|
int k; // LUT node index
|
|
|
|
// Clipped rows at top
|
|
for ( i = -m_n + 1, k = 0; i <= m_n; ++i, ++k )
|
|
{
|
|
int y = y0 + i;
|
|
if ( y >= 0 )
|
|
break;
|
|
if ( m_fillBorder )
|
|
FillRow( sp, sn, wp, wn, m_lut[k][dy] );
|
|
else
|
|
InterpolateRow( sp, sn, wp, wn, m_data - 2*int64( y )*m_width, x0, m_lut[k][dy] );
|
|
}
|
|
|
|
// Unclipped rows
|
|
for ( ; i <= m_n; ++i, ++k )
|
|
{
|
|
int y = y0 + i;
|
|
if ( y == m_height )
|
|
break;
|
|
InterpolateRow( sp, sn, wp, wn, m_data + int64( y )*m_width, x0, m_lut[k][dy] );
|
|
}
|
|
|
|
// Clipped rows at bottom
|
|
for ( ; i <= m_n; ++i, ++k )
|
|
{
|
|
if ( m_fillBorder )
|
|
FillRow( sp, sn, wp, wn, m_lut[k][dy] );
|
|
else
|
|
InterpolateRow( sp, sn, wp, wn, m_data + int64( 2*m_height - 2 - y0 - i )*m_width, x0, m_lut[k][dy] );
|
|
}
|
|
}
|
|
else
|
|
{
|
|
// Interpolation increments
|
|
double dx = x - x0;
|
|
double dy = y - y0;
|
|
|
|
// Precalculate horizontal filter values
|
|
for ( int j = -m_n + 1, k = 0; j <= m_n; ++j, ++k )
|
|
m_Lx[k] = Lanczos( j - dx );
|
|
|
|
// Clipped rows at top
|
|
for ( i = -m_n + 1; i <= m_n; ++i )
|
|
{
|
|
int y = y0 + i;
|
|
if ( y >= 0 )
|
|
break;
|
|
if ( m_fillBorder )
|
|
FillRow( sp, sn, wp, wn, Lanczos( i - dy ) );
|
|
else
|
|
InterpolateRow( sp, sn, wp, wn, m_data - 2*int64( y )*m_width, x0, Lanczos( i - dy ) );
|
|
}
|
|
|
|
// Unclipped rows
|
|
for ( ; i <= m_n; ++i )
|
|
{
|
|
int y = y0 + i;
|
|
if ( y == m_height )
|
|
break;
|
|
InterpolateRow( sp, sn, wp, wn, m_data + int64( y )*m_width, x0, Lanczos( i - dy ) );
|
|
}
|
|
|
|
// Clipped rows at bottom
|
|
for ( ; i <= m_n; ++i )
|
|
{
|
|
if ( m_fillBorder )
|
|
FillRow( sp, sn, wp, wn, Lanczos( i - dy ) );
|
|
else
|
|
InterpolateRow( sp, sn, wp, wn, m_data + int64( 2*m_height - 2 - y0 - i )*m_width, x0, Lanczos( i - dy ) );
|
|
}
|
|
}
|
|
|
|
// Clamping
|
|
if ( m_clamp )
|
|
{
|
|
// Empty data?
|
|
if ( sp == 0 )
|
|
return 0;
|
|
|
|
// Clamping ratio: s-/s+
|
|
double r = sn/sp;
|
|
|
|
// Clamp for s- >= s+
|
|
if ( r >= 1 )
|
|
return sp/wp;
|
|
|
|
// Clamp for c < s- < s+
|
|
if ( r > m_clampTh )
|
|
{
|
|
r = (r - m_clampTh)/m_clampThInv;
|
|
double c = 1 - r*r;
|
|
sn *= c, wn *= c;
|
|
}
|
|
}
|
|
|
|
// Weighted convolution
|
|
return (sp - sn)/(wp - wn);
|
|
}
|
|
|
|
/*!
|
|
* Returns true iff the interpolation clamping feature has been enabled for
|
|
* this object.
|
|
*
|
|
* \sa EnableClamping(), ClampingThreshold()
|
|
*/
|
|
bool IsClampingEnabled() const noexcept
|
|
{
|
|
return m_clamp;
|
|
}
|
|
|
|
/*!
|
|
* Enables (or disables) the interpolation clamping feature.
|
|
*
|
|
* \sa IsClampingEnabled(), DisableClamping(), SetClampingThreshold()
|
|
*/
|
|
void EnableClamping( bool enable = true ) noexcept
|
|
{
|
|
m_clamp = enable;
|
|
}
|
|
|
|
/*!
|
|
* Disables (or enables) the interpolation clamping feature.
|
|
*
|
|
* \sa IsClampingEnabled(), EnableClamping(), SetClampingThreshold()
|
|
*/
|
|
void DisableClamping( bool disable = true ) noexcept
|
|
{
|
|
EnableClamping( !disable );
|
|
}
|
|
|
|
/*!
|
|
* Returns the current <em>clamping threshold</em> for this object.
|
|
*
|
|
* See the documentation for SetClampingThreshold( float ) for a detailed
|
|
* description of the clamping mechanism.
|
|
*
|
|
* \sa SetClampingThreshold(), IsClampingEnabled(), EnableClamping()
|
|
*/
|
|
float ClampingThreshold() const noexcept
|
|
{
|
|
return m_clampTh;
|
|
}
|
|
|
|
/*!
|
|
* Defines a threshold to trigger interpolation \e clamping.
|
|
*
|
|
* Lanczos interpolation generates strong undershoot (aka ringing) artifacts
|
|
* when the negative lobes of the interpolation function fall over bright
|
|
* isolated pixels or edges. The clamping mechanism acts by limiting the
|
|
* high-pass component of the interpolation filter selectively to fix these
|
|
* problems.
|
|
*
|
|
* The specified clamping threshold \e clamp must be in the [0,1] range.
|
|
* Lower values cause a more aggressive deringing effect. Too strong of a
|
|
* clamping threshold can degrade performance of the Lanczos filter to some
|
|
* degree, since it tends to block its high-pass behavior.
|
|
*
|
|
* \note The interpolation clamping feature must be enabled for this
|
|
* threshold to have any effect. See the constructor for this class and the
|
|
* documentation for IsClampingEnabled().
|
|
*
|
|
* \sa ClampingThreshold(), IsClampingEnabled(), EnableClamping()
|
|
*/
|
|
void SetClampingThreshold( float clamp ) noexcept
|
|
{
|
|
PCL_PRECONDITION( 0 <= clamp && clamp <= 1 )
|
|
m_clampTh = Range( clamp, 0.0F, 1.0F );
|
|
}
|
|
|
|
private:
|
|
|
|
int m_n; // filter order
|
|
const double** m_lut; // precomputed function values
|
|
mutable DVector m_Lx; // precalculated row of function values
|
|
double m_clampTh; // clamping threshold in [0,1]
|
|
double m_clampThInv; // 1 - m_clampTh
|
|
bool m_clamp; // clamping enabled ?
|
|
|
|
/*
|
|
* Sinc function for x > 0
|
|
*/
|
|
static double Sinc( double x ) noexcept
|
|
{
|
|
x *= Const<double>::pi();
|
|
return (x > 1.0e-07) ? Sin( x )/x : 1.0;
|
|
}
|
|
|
|
/*
|
|
* Evaluate Lanczos function at x.
|
|
*/
|
|
double Lanczos( double x ) const noexcept
|
|
{
|
|
if ( x < 0 )
|
|
x = -x;
|
|
if ( x < m_n )
|
|
return Sinc( x ) * Sinc( x/m_n );
|
|
return 0;
|
|
}
|
|
|
|
/*
|
|
* Interpolate a row of pixels.
|
|
* Can be either an unclipped row or a mirrored border row.
|
|
*/
|
|
void InterpolateRow( double& sp, double& sn, double& wp, double& wn, const T* f, int x0, double Ly ) const noexcept
|
|
{
|
|
int j, k;
|
|
|
|
// Clipped pixels at the left border
|
|
for ( j = -m_n + 1, k = 0; j <= m_n; ++j, ++k )
|
|
{
|
|
int x = x0 + j;
|
|
if ( x >= 0 )
|
|
break;
|
|
double L = m_Lx[k] * Ly;
|
|
double s = (m_fillBorder ? m_fillValue : double( f[-x] )) * L;
|
|
PCL_LANCZOS_ACC()
|
|
}
|
|
|
|
// Unclipped pixels
|
|
for ( ; j <= m_n; ++j, ++k )
|
|
{
|
|
int x = x0 + j;
|
|
if ( x == m_width )
|
|
break;
|
|
double L = m_Lx[k] * Ly;
|
|
double s = f[x] * L;
|
|
PCL_LANCZOS_ACC()
|
|
}
|
|
|
|
// Clipped pixels at the right border
|
|
for ( ; j <= m_n; ++j, ++k )
|
|
{
|
|
int x = x0 + j;
|
|
double L = m_Lx[k] * Ly;
|
|
double s = (m_fillBorder ? m_fillValue : double( f[2*m_width - 2 - x] )) * L;
|
|
PCL_LANCZOS_ACC()
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Interpolate a clipped pixel row with border filling.
|
|
*/
|
|
void FillRow( double& sp, double& sn, double& wp, double& wn, double Ly ) const noexcept
|
|
{
|
|
for ( int j = -m_n + 1, k = 0; j <= m_n; ++j, ++k )
|
|
{
|
|
double L = m_Lx[k] * Ly;
|
|
double s = m_fillValue * L;
|
|
PCL_LANCZOS_ACC()
|
|
}
|
|
}
|
|
};
|
|
|
|
// ----------------------------------------------------------------------------
|
|
|
|
/*!
|
|
* \internal
|
|
* \class LanczosLUTInterpolationBase
|
|
* \brief Base class of two dimensional LUT-based Lanczos interpolation algorithms.
|
|
*
|
|
* This is the base class for fixed-order Lanczos interpolation algorithms
|
|
* implemented through precalculated look-up tables (LUTs). The filter order
|
|
* \a n is specified as the second template class argument. For a description
|
|
* of the Lanczos algorithm and information on its performance and features,
|
|
* refer to the documentation for the LanczosInterpolation class.
|
|
*
|
|
* LUT-based Lanczos interpolations are about three times faster than the
|
|
* corresponding function evaluation interpolations. Interpolation from the
|
|
* implemented LUTs provides a maximum error of +/- 1/2^16, so this class and
|
|
* its derived classes are fully accurate for 8-bit and 16-bit integer images.
|
|
*
|
|
* \sa BidimensionalInterpolation, LanczosInterpolation,
|
|
* Lanczos3LUTInterpolation, Lanczos4LUTInterpolation, Lanczos5LUTInterpolation
|
|
*/
|
|
template <typename T, int m_n>
|
|
class PCL_CLASS LanczosLUTInterpolationBase : public BidimensionalInterpolation<T>
|
|
{
|
|
public:
|
|
|
|
/*!
|
|
* Constructs a %LanczosLUTInterpolationBase instance.
|
|
*
|
|
* \param clamp Clamping threshold. Clamping is applied to fix undershoot
|
|
* (aka ringing) artifacts. A value of this parameter within
|
|
* the [0,1] range enables clamping: The lower the clamping
|
|
* threshold, the more aggressive deringing effect is
|
|
* achieved. A negative threshold value disables the clamping
|
|
* feature. The default value is 0.3. For more information,
|
|
* refer to the documentation for the
|
|
* SetClampingThreshold( float ) member function.
|
|
*/
|
|
LanczosLUTInterpolationBase( float clamp )
|
|
: m_lut( PCL_InitializeLanczosIntLUT( m_n ) )
|
|
, m_Lx( 2*m_n )
|
|
, m_Ly( 2*m_n )
|
|
, m_clampTh( Range( clamp, 0.0F, 1.0F ) )
|
|
, m_clampThInv( 1 - m_clampTh )
|
|
, m_clamp( clamp >= 0 )
|
|
{
|
|
PCL_PRECONDITION( m_n >= 1 )
|
|
PCL_PRECONDITION( clamp < 0 || 0 <= clamp && clamp <= 1 )
|
|
PCL_CHECK( m_lut != nullptr )
|
|
}
|
|
|
|
/*!
|
|
* Copy constructor.
|
|
*/
|
|
LanczosLUTInterpolationBase( const LanczosLUTInterpolationBase& ) = default;
|
|
|
|
/*!
|
|
* Virtual destructor.
|
|
*/
|
|
virtual ~LanczosLUTInterpolationBase()
|
|
{
|
|
}
|
|
|
|
/*!
|
|
* Interpolated value at \a {x,y} location.
|
|
*
|
|
* \param x,y %Coordinates of the interpolation point (horizontal,
|
|
* vertical).
|
|
*/
|
|
double operator()( double x, double y ) const override
|
|
{
|
|
PCL_PRECONDITION( m_data != nullptr )
|
|
PCL_PRECONDITION( m_width > 0 && m_height > 0 )
|
|
PCL_PRECONDITION( x >= 0 && x < m_width )
|
|
PCL_PRECONDITION( y >= 0 && y < m_height )
|
|
|
|
// Central grid coordinates
|
|
int x0 = Range( TruncInt( x ), 0, m_width-1 );
|
|
int y0 = Range( TruncInt( y ), 0, m_height-1 );
|
|
|
|
// Precalculate function values
|
|
int dx = RoundInt( (x - x0)*__PCL_LANCZOS_LUT_INT_RESOLUTION );
|
|
int dy = RoundInt( (y - y0)*__PCL_LANCZOS_LUT_INT_RESOLUTION );
|
|
for ( int j = -m_n + 1, k = 0; j <= m_n; ++j, ++k )
|
|
{
|
|
int d0 = j*__PCL_LANCZOS_LUT_INT_RESOLUTION;
|
|
m_Lx[k] = m_lut[Abs( d0 - dx )];
|
|
m_Ly[k] = m_lut[Abs( d0 - dy )];
|
|
}
|
|
|
|
double sp = 0; // positive filter values
|
|
double sn = 0; // negative filter values
|
|
double wp = 0; // positive filter weight
|
|
double wn = 0; // negative filter weight
|
|
int i, k; // row and coefficient indices
|
|
|
|
// Clipped rows at top
|
|
for ( i = -m_n + 1, k = 0; i <= m_n; ++i, ++k )
|
|
{
|
|
int y = y0 + i;
|
|
if ( y >= 0 )
|
|
break;
|
|
if ( m_fillBorder )
|
|
FillRow( sp, sn, wp, wn, m_Ly[k] );
|
|
else
|
|
InterpolateRow( sp, sn, wp, wn, m_data - 2*int64( y )*m_width, x0, m_Ly[k] );
|
|
}
|
|
|
|
// Unclipped rows
|
|
for ( ; i <= m_n; ++i, ++k )
|
|
{
|
|
int y = y0 + i;
|
|
if ( y == m_height )
|
|
break;
|
|
InterpolateRow( sp, sn, wp, wn, m_data + int64( y )*m_width, x0, m_Ly[k] );
|
|
}
|
|
|
|
// Clipped rows at bottom
|
|
for ( ; i <= m_n; ++i, ++k )
|
|
{
|
|
if ( m_fillBorder )
|
|
FillRow( sp, sn, wp, wn, m_Ly[k] );
|
|
else
|
|
InterpolateRow( sp, sn, wp, wn, m_data + int64( 2*m_height - 2 - y0 - i )*m_width, x0, m_Ly[k] );
|
|
}
|
|
|
|
// Clamping
|
|
if ( m_clamp )
|
|
{
|
|
// Empty data?
|
|
if ( sp == 0 )
|
|
return 0;
|
|
|
|
// Clamping ratio: s-/s+
|
|
double r = sn/sp;
|
|
|
|
// Clamp for s- >= s+
|
|
if ( r >= 1 )
|
|
return sp/wp;
|
|
|
|
// Clamp for c < s- < s+
|
|
if ( r > m_clampTh )
|
|
{
|
|
r = (r - m_clampTh)/m_clampThInv;
|
|
double c = 1 - r*r;
|
|
sn *= c, wn *= c;
|
|
}
|
|
}
|
|
|
|
// Weighted convolution
|
|
return (sp - sn)/(wp - wn);
|
|
}
|
|
|
|
/*!
|
|
* Returns true iff the interpolation clamping feature has been enabled for
|
|
* this object.
|
|
*
|
|
* \sa EnableClamping(), ClampingThreshold()
|
|
*/
|
|
bool IsClampingEnabled() const noexcept
|
|
{
|
|
return m_clamp;
|
|
}
|
|
|
|
/*!
|
|
* Enables (or disables) the interpolation clamping feature.
|
|
*
|
|
* \sa IsClampingEnabled(), DisableClamping(), SetClampingThreshold()
|
|
*/
|
|
void EnableClamping( bool enable = true ) noexcept
|
|
{
|
|
m_clamp = enable;
|
|
}
|
|
|
|
/*!
|
|
* Disables (or enables) the interpolation clamping feature.
|
|
*
|
|
* \sa IsClampingEnabled(), EnableClamping(), SetClampingThreshold()
|
|
*/
|
|
void DisableClamping( bool disable = true ) noexcept
|
|
{
|
|
EnableClamping( !disable );
|
|
}
|
|
|
|
/*!
|
|
* Returns the current <em>clamping threshold</em> for this object.
|
|
*
|
|
* See the documentation for SetClampingThreshold( float ) for a detailed
|
|
* description of the clamping mechanism.
|
|
*
|
|
* \sa SetClampingThreshold(), IsClampingEnabled(), EnableClamping()
|
|
*/
|
|
float ClampingThreshold() const noexcept
|
|
{
|
|
return m_clampTh;
|
|
}
|
|
|
|
/*!
|
|
* Defines a threshold to trigger interpolation \e clamping.
|
|
*
|
|
* Lanczos interpolation generates strong undershoot (aka ringing) artifacts
|
|
* when the negative lobes of the interpolation function fall over bright
|
|
* isolated pixels or edges. The clamping mechanism acts by limiting the
|
|
* high-pass component of the interpolation filter selectively to fix these
|
|
* problems.
|
|
*
|
|
* The specified clamping threshold \e clamp must be in the [0,1] range.
|
|
* Lower values cause a more aggressive deringing effect. Too strong of a
|
|
* clamping threshold can degrade performance of the Lanczos filter to some
|
|
* degree, since it tends to block its high-pass behavior.
|
|
*
|
|
* \note The interpolation clamping feature must be enabled for this
|
|
* threshold to have any effect. See the constructor for this class and the
|
|
* documentation for IsClampingEnabled().
|
|
*
|
|
* \sa ClampingThreshold(), IsClampingEnabled(), EnableClamping()
|
|
*/
|
|
void SetClampingThreshold( float clamp ) noexcept
|
|
{
|
|
PCL_PRECONDITION( 0 <= clamp && clamp <= 1 )
|
|
m_clampTh = Range( clamp, 0.0F, 1.0F );
|
|
}
|
|
|
|
private:
|
|
|
|
const float* m_lut; // filter LUT
|
|
mutable FVector m_Lx, m_Ly; // precalculated function values
|
|
double m_clampTh; // clamping threshold in [0,1]
|
|
double m_clampThInv; // 1 - m_clampTh
|
|
bool m_clamp; // clamping enabled ?
|
|
|
|
/*
|
|
* Interpolate a row of pixels.
|
|
* Can be either an unclipped row or a mirrored border row.
|
|
*/
|
|
void InterpolateRow( double& sp, double& sn, double& wp, double& wn, const T* f, int x0, float Ly ) const noexcept
|
|
{
|
|
int j, k;
|
|
|
|
// Clipped pixels at the left border
|
|
for ( j = -m_n + 1, k = 0; j <= m_n; ++j, ++k )
|
|
{
|
|
int x = x0 + j;
|
|
if ( x >= 0 )
|
|
break;
|
|
double L = m_Lx[k] * Ly;
|
|
double s = (m_fillBorder ? m_fillValue : double( f[-x] )) * L;
|
|
PCL_LANCZOS_ACC()
|
|
}
|
|
|
|
// Unclipped pixels
|
|
for ( ; j <= m_n; ++j, ++k )
|
|
{
|
|
int x = x0 + j;
|
|
if ( x == m_width )
|
|
break;
|
|
double L = m_Lx[k] * Ly;
|
|
double s = f[x] * L;
|
|
PCL_LANCZOS_ACC()
|
|
}
|
|
|
|
// Clipped pixels at the right border
|
|
for ( ; j <= m_n; ++j, ++k )
|
|
{
|
|
int x = x0 + j;
|
|
double L = m_Lx[k] * Ly;
|
|
double s = (m_fillBorder ? m_fillValue : double( f[2*m_width - 2 - x] )) * L;
|
|
PCL_LANCZOS_ACC()
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Interpolate a clipped pixel row with border filling.
|
|
*/
|
|
void FillRow( double& sp, double& sn, double& wp, double& wn, float Ly ) const noexcept
|
|
{
|
|
for ( int j = -m_n + 1, k = 0; j <= m_n; ++j, ++k )
|
|
{
|
|
double L = m_Lx[k] * Ly;
|
|
double s = m_fillValue * L;
|
|
PCL_LANCZOS_ACC()
|
|
}
|
|
}
|
|
};
|
|
|
|
// ----------------------------------------------------------------------------
|
|
|
|
/*!
|
|
* \class Lanczos3LUTInterpolation
|
|
* \brief Two dimensional LUT-based 3rd-order Lanczos interpolation algorithm.
|
|
*
|
|
* This class implements 3rd-order Lanczos interpolation through precalculated
|
|
* look-up tables. For a description of the Lanczos algorithm and information
|
|
* on its performance and features, refer to the documentation for the
|
|
* LanczosInterpolation class.
|
|
*
|
|
* LUT-based Lanczos interpolations are about three times faster than the
|
|
* corresponding function evaluation interpolations. Interpolation from the
|
|
* implemented LUTs provides a maximum error of +/- 1/2^16, so this class is
|
|
* fully accurate for 8-bit and 16-bit integer images.
|
|
*
|
|
* \sa LanczosInterpolation, LanczosLUTInterpolationBase,
|
|
* Lanczos4LUTInterpolation, Lanczos5LUTInterpolation
|
|
*/
|
|
template <typename T>
|
|
class PCL_CLASS Lanczos3LUTInterpolation : public LanczosLUTInterpolationBase<T,3>
|
|
{
|
|
public:
|
|
|
|
/*!
|
|
* Constructs a %Lanczos3LUTInterpolation instance.
|
|
*
|
|
* \param clamp Clamping threshold. Clamping is applied to fix undershoot
|
|
* (aka ringing) artifacts. A value of this parameter within
|
|
* the [0,1] range enables clamping: The lower the clamping
|
|
* threshold, the more aggressive deringing effect is
|
|
* achieved. A negative threshold value disables the clamping
|
|
* feature. The default value is 0.3. For more information,
|
|
* refer to the documentation for the
|
|
* SetClampingThreshold( float ) member function.
|
|
*/
|
|
Lanczos3LUTInterpolation( float clamp = __PCL_LANCZOS_CLAMPING_THRESHOLD )
|
|
: LanczosLUTInterpolationBase<T,3>( clamp )
|
|
{
|
|
PCL_PRECONDITION( 0 <= clamp && clamp <= 1 )
|
|
}
|
|
|
|
/*!
|
|
* Copy constructor.
|
|
*/
|
|
Lanczos3LUTInterpolation( const Lanczos3LUTInterpolation& ) = default;
|
|
|
|
/*!
|
|
* Virtual destructor.
|
|
*/
|
|
virtual ~Lanczos3LUTInterpolation()
|
|
{
|
|
}
|
|
};
|
|
|
|
// ----------------------------------------------------------------------------
|
|
|
|
/*!
|
|
* \class Lanczos4LUTInterpolation
|
|
* \brief Two dimensional LUT-based 4th-order Lanczos interpolation algorithm.
|
|
*
|
|
* This class implements 4th-order Lanczos interpolation through precalculated
|
|
* look-up tables. For a description of the Lanczos algorithm and information
|
|
* on its performance and features, refer to the documentation for the
|
|
* LanczosInterpolation class.
|
|
*
|
|
* LUT-based Lanczos interpolations are about three times faster than the
|
|
* corresponding function evaluation interpolations. Interpolation from the
|
|
* implemented LUTs provides a maximum error of +/- 1/2^16, so this class is
|
|
* fully accurate for 8-bit and 16-bit integer images.
|
|
*
|
|
* \sa LanczosInterpolation, LanczosLUTInterpolationBase,
|
|
* Lanczos3LUTInterpolation, Lanczos5LUTInterpolation
|
|
*/
|
|
template <typename T>
|
|
class PCL_CLASS Lanczos4LUTInterpolation : public LanczosLUTInterpolationBase<T,4>
|
|
{
|
|
public:
|
|
|
|
/*!
|
|
* Constructs a %Lanczos4LUTInterpolation instance.
|
|
*
|
|
* \param clamp Clamping threshold. Clamping is applied to fix undershoot
|
|
* (aka ringing) artifacts. A value of this parameter within
|
|
* the [0,1] range enables clamping: The lower the clamping
|
|
* threshold, the more aggressive deringing effect is
|
|
* achieved. A negative threshold value disables the clamping
|
|
* feature. The default value is 0.3. For more information,
|
|
* refer to the documentation for the
|
|
* SetClampingThreshold( float ) member function.
|
|
*/
|
|
Lanczos4LUTInterpolation( float clamp = __PCL_LANCZOS_CLAMPING_THRESHOLD )
|
|
: LanczosLUTInterpolationBase<T,4>( clamp )
|
|
{
|
|
PCL_PRECONDITION( 0 <= clamp && clamp <= 1 )
|
|
}
|
|
|
|
/*!
|
|
* Copy constructor.
|
|
*/
|
|
Lanczos4LUTInterpolation( const Lanczos4LUTInterpolation& ) = default;
|
|
|
|
/*!
|
|
* Virtual destructor.
|
|
*/
|
|
virtual ~Lanczos4LUTInterpolation()
|
|
{
|
|
}
|
|
};
|
|
|
|
// ----------------------------------------------------------------------------
|
|
|
|
/*!
|
|
* \class Lanczos5LUTInterpolation
|
|
* \brief Two dimensional LUT-based 5th-order Lanczos interpolation algorithm.
|
|
*
|
|
* This class implements 5th-order Lanczos interpolation through precalculated
|
|
* look-up tables. For a description of the Lanczos algorithm and information
|
|
* on its performance and features, refer to the documentation for the
|
|
* LanczosInterpolation class.
|
|
*
|
|
* LUT-based Lanczos interpolations are about three times faster than the
|
|
* corresponding function evaluation interpolations. Interpolation from the
|
|
* implemented LUTs provides a maximum error of +/- 1/2^16, so this class is
|
|
* fully accurate for 8-bit and 16-bit integer images.
|
|
*
|
|
* \sa LanczosInterpolation, LanczosLUTInterpolationBase,
|
|
* Lanczos3LUTInterpolation, Lanczos4LUTInterpolation
|
|
*/
|
|
template <typename T>
|
|
class PCL_CLASS Lanczos5LUTInterpolation : public LanczosLUTInterpolationBase<T,5>
|
|
{
|
|
public:
|
|
|
|
/*!
|
|
* Constructs a %Lanczos5LUTInterpolation instance.
|
|
*
|
|
* \param clamp Clamping threshold. Clamping is applied to fix undershoot
|
|
* (aka ringing) artifacts. A value of this parameter within
|
|
* the [0,1] range enables clamping: The lower the clamping
|
|
* threshold, the more aggressive deringing effect is
|
|
* achieved. A negative threshold value disables the clamping
|
|
* feature. The default value is 0.3. For more information,
|
|
* refer to the documentation for the
|
|
* SetClampingThreshold( float ) member function.
|
|
*/
|
|
Lanczos5LUTInterpolation( float clamp = __PCL_LANCZOS_CLAMPING_THRESHOLD )
|
|
: LanczosLUTInterpolationBase<T,5>( clamp )
|
|
{
|
|
PCL_PRECONDITION( 0 <= clamp && clamp <= 1 )
|
|
}
|
|
|
|
/*!
|
|
* Copy constructor.
|
|
*/
|
|
Lanczos5LUTInterpolation( const Lanczos5LUTInterpolation& ) = default;
|
|
|
|
/*!
|
|
* Virtual destructor.
|
|
*/
|
|
virtual ~Lanczos5LUTInterpolation()
|
|
{
|
|
}
|
|
};
|
|
|
|
// ----------------------------------------------------------------------------
|
|
|
|
#undef PCL_LANCZOS_ACC
|
|
|
|
#undef m_width
|
|
#undef m_height
|
|
#undef m_fillBorder
|
|
#undef m_fillValue
|
|
#undef m_data
|
|
|
|
// ----------------------------------------------------------------------------
|
|
|
|
} // pcl
|
|
|
|
#endif // __PCL_LanczosInterpolation_h
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// EOF pcl/LanczosInterpolation.h - Released 2022-03-12T18:59:29Z
|