vtb::core::Convolution< T, N > Class Template Reference

Multidimensional convolution based on Fourier transform. More...

#include <convolution.hh>

Public Types
typedef Fourier_transform< T, N >	transform_type
typedef transform_type::shape_type	shape_type
typedef transform_type::array_type	array_type
typedef blitz::RectDomain< N >	domain_type

Public Member Functions
	Convolution (const shape_type &signal_shape, const shape_type &kernel_shape)
void	shape (const shape_type &signal_shape, const shape_type &kernel_shape)
void	padded_block_shape (const shape_type &block_size, const shape_type &padding)
shape_type	padded_block_shape () const noexcept
const shape_type &	block_shape () const noexcept
const shape_type &	padding () const noexcept
array_type	operator() (array_type signal, array_type kernel)
array_type	convolve (array_type signal, array_type kernel)

Detailed Description

template<class T, int N>
class vtb::core::Convolution< T, N >

Multidimensional convolution based on Fourier transform.

Date

2018-11-03

Author

Ivan Gankevich

• Slicing is done in specified dimension with specified padding.

Definition at line 21 of file convolution.hh.

Member Function Documentation

convolve()

template<class T , int N>

auto vtb::core::Convolution< T, N >::convolve	(	array_type	signal,
		array_type	kernel
	)

Zero-pad kernel to be of length block_size + padding.

Take forward FFT of padded kernel.

Decompose input signal into blocks of length block_size.

Zero-pad each part to be of length block_size + padding.

Take forward FFT of each padded part.

Multiply two FFTs.

Take backward FFT of the result.

Copy padded part back overlapping it with adjacent parts.

Definition at line 36 of file convolution.cc.