SepImageConvolver.h

Classes

SepImageConvolver -- This class does separable convolution of an image (full description)

template <class T> class SepImageConvolver

Interface

Public Members

SepImageConvolver (ImageInterface<T>& image, T &os, Bool showProgress)

SepImageConvolver(const SepImageConvolver<T> &other)

~SepImageConvolver()

SepImageConvolver &operator=(const SepImageConvolver<T> &other)

void setKernel(uInt axis, const Vector<T>& kernel)

void setKernel(uInt axis, VectorKernel::KernelTypes kernelType, const Quantum<Double>& width, Bool autoScale, Bool useImageShapeExactly=True, Double scale=1.0)

void setKernel(uInt axis, VectorKernel::KernelTypes kernelType, Double width, Bool autoScale, Bool useImageShapeExactly=True, Double scale=1.0)

Vector<T> getKernel(uInt axis)

uInt getKernelShape(uInt axis)

void convolve(ImageInterface<T>& imageOut)

void convolve()

Private Members

void checkAxis(uInt axis)

Bool isTempImage (const ImageInterface<Float>* pIm) const

void zero()

void smoothProfiles (ImageInterface<T>& in, const Int& axis, const T<T>& psf)

Description

Review Status

Date Reviewed:: yyyy/mm/dd

Prerequisite

Etymology

This class handles convolution of images by separable kernels.

Synopsis

Convolution kernels can be separable or not. For example, convolution of an image by a 2-D gaussian when the position angle of the Gaussian is along one of the axes is separable. If the position angle is otherwise, it is not separable. When the kernel is separable, an N-dimensional specification of the convolution kernel is straightforward.

Although this class is templated, it will only work for Float and Double types.

Example

Motivation

Separable and non-separable convolution are standard requirements.

To Do

Member Description

SepImageConvolver (ImageInterface<T>& image, T &os, Bool showProgress)

Constructor

SepImageConvolver(const SepImageConvolver<T> &other)

Copy constructor. Uses reference semantics.

~SepImageConvolver()

Destructor

SepImageConvolver &operator=(const SepImageConvolver<T> &other)

Assignment operator. Uses reference semantics.

void setKernel(uInt axis, const Vector<T>& kernel)

Set convolution kernel vector. The specified axis is convolved by the given kernel.

void setKernel(uInt axis, VectorKernel::KernelTypes kernelType, const Quantum<Double>& width, Bool autoScale, Bool useImageShapeExactly=True, Double scale=1.0)
void setKernel(uInt axis, VectorKernel::KernelTypes kernelType, Double width, Bool autoScale, Bool useImageShapeExactly=True, Double scale=1.0)

Set convolution kernel. The specified axis is convolved by the given kernel. If autoScale is True then kernel volume is unity, else kernel peak is 1 * scale. If useImageShapeExactly is True, the kernel will be the shape of the axis, else it will be big enough to accomodate the kernel width (e.g. +/- 5sigma for Gaussian)

Vector<T> getKernel(uInt axis)

Get the convolution kernel for the specified axis

uInt getKernelShape(uInt axis)

Get the convolution kernel shape for the specified axis

void convolve(ImageInterface<T>& imageOut)
void convolve()

Perform the convolution either outputting to a new image or in-situ. The error checking for the convolution parameters is done when you call this function. If outputting a new image, and it needs a mask and doesn't have one, the it will be given one if possible and the input mask will be transferred to the output. Masked pixels are zeroed before convolving