Classes
class	gpufilter::dvector< T >
	Device Vector class. More...
struct	gpufilter::_clamp
	Access pattern: clamp-to-border. More...
struct	gpufilter::_repeat
	Access pattern: repeat. More...
struct	gpufilter::_mirror
	Access pattern: mirror. More...
class	gpufilter::base_timer
	Base timer class. More...
class	gpufilter::gpu_timer
	GPU specialization of the timer class. More...
class	gpufilter::cpu_timer
	CPU specialization of the timer class. More...
class	gpufilter::scoped_timer_stop
	Scope-limited timer class. More...
class	gpufilter::timer_pool
	Pool of timers. More...
class	gpufilter::Vector< T, N >
	Vector class. More...
class	gpufilter::Matrix< T, M, N >
	Matrix class. More...
Enumerations
enum	gpufilter::initcond { gpufilter::zero, gpufilter::clamp, gpufilter::repeat, gpufilter::mirror }
	Enumerates possible initial conditions for 2D-image access. More...
Functions
template<class T >
T *	gpufilter::cuda_new (size_t elements)
	Allocates a new memory space in the GPU.
template<class T >
void	gpufilter::cuda_delete (T *d_ptr)
	Deallocates a memory space in the GPU.
void	gpufilter::calc_borders (int &left, int &top, int &right, int &bottom, const int &w, const int &h, const int &extb)
	Calculate image borders.
bool	gpufilter::extend (const int &w, const int &h, const int &extb)
	Verify if an image needs to be extended.
void	gpufilter::cuda_error (const std::string &msg)
	Check error in device.
template<class T >
T	gpufilter::lookat (const T *in, const int &i, const int &n, const initcond &ic, const int &p=1)
	Look in an input at a given index range.
template<class T >
T	gpufilter::lookat (const T *img, const int &i, const int &j, const int &h, const int &w, const initcond &ic)
	Look in an image at a given position.
template<class T >
T	gpufilter::qs (const T &s)
	Compute recursive filtering scaling factor.
template<class T >
std::complex< T >	gpufilter::ds (const std::complex< T > &d, const T &s)
	Rescale poles of the recursive filtering z-transform.
template<class T >
T	gpufilter::ds (const T &d, const T &s)
	Rescale poles in the real-axis of the recursive filtering z-transform.
template<class T1 , class T2 >
void	gpufilter::weights1 (const T1 &s, T2 &b0, T2 &a1)
	Compute first-order weights.
template<class T1 , class T2 >
void	gpufilter::weights2 (const T1 &s, T2 &b0, T2 &a1, T2 &a2)
	Compute first- and second-order weights.
template<class T >
void	gpufilter::copy_to_symbol (const std::string &name, const T &value)
	Copy value(s) to symbol.
void	gpufilter::copy_to_symbol (const std::string &name, unsigned long value)
void	gpufilter::copy_to_symbol (const std::string &name, long value)
template<class T >
void	gpufilter::copy_to_symbol (const std::string &name, const std::string &size_name, const std::vector< T > &items)
template<class T >
void	gpufilter::copy_to_symbol (const std::string &name, const std::vector< T > &items)
Variables
timer_pool	gpufilter::timers
	Global pool of timers.

Enumeration Type Documentation

enum gpufilter::initcond

Enumerates possible initial conditions for 2D-image access.

Enumerator:

zero	Zero-border (outside image everything is zero)
clamp	Clamp-to-border (image border continues forever)
repeat	Repeat (image repeats itself forever)
mirror	Mirror (image repeats itself mirrored forever)

Function Documentation

void gpufilter::calc_borders	(	int &	left,
		int &	top,
		int &	right,
		int &	bottom,
		const int &	w,
		const int &	h,
		const int &	extb
	)

Calculate image borders.

Given the image size and an extension to consider calculates the four-directional border pixels (left, top, right and bottom) needed by algorithm 4 and 5.

Parameters:

[out]	left	Extension (in pixels) in the left border of the image
[out]	top	Extension (in pixels) in the top border of the image
[out]	right	Extension (in pixels) in the right border of the image
[out]	bottom	Extension (in pixels) in the bottom border of the image
[in]	w	Width of the image
[in]	h	Height of the image
[in]	extb	Extension (in blocks) to consider outside image (default 0)

template<class T >

void gpufilter::copy_to_symbol	(	const std::string &	name,
		const T &	value
	)

Copy value(s) to symbol.

Parameters:

[in]	name	Name of the symbol in device (GPU)
[in]	value	Value to copy to symbol

Template Parameters:

T	Value type used in the GPU

void gpufilter::copy_to_symbol	(	const std::string &	name,
		unsigned long	value
	)		`[inline]`

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

Parameters:

[in]	name	Name of the symbol in device (GPU)
[in]	value	Value to copy to symbol

void gpufilter::copy_to_symbol	(	const std::string &	name,
		long	value
	)		`[inline]`

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

Parameters:

[in]	name	Name of the symbol in device (GPU)
[in]	value	Value to copy to symbol

template<class T >

void gpufilter::copy_to_symbol	(	const std::string &	name,
		const std::string &	size_name,
		const std::vector< T > &	items
	)

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

Parameters:

[in]	name	Name of the symbol in device (GPU)
[in]	size_name	Name of the symbol storing the size of the items to be copied
[in]	items	Host (STL) Vector values

Template Parameters:

T	Vector values type used in the CPU and the GPU

template<class T >

void gpufilter::copy_to_symbol	(	const std::string &	name,
		const std::vector< T > &	items
	)

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

Parameters:

[in]	name	Name of the symbol in device (GPU)
[in]	items	Host (STL) Vector values

Template Parameters:

T	Vector values type used in the CPU and the GPU

template<class T >

void gpufilter::cuda_delete ( T * d_ptr )

Deallocates a memory space in the GPU.

This function deletes device (GPU) memory space.

Parameters:

[in] d_ptr Device pointer (in the GPU memory)

Template Parameters:

T	Memory values type

void gpufilter::cuda_error ( const std::string & msg ) [inline]

Check error in device.

This function checks if there is a device error.

Parameters:

[in] msg Message to appear in case of a device error

template<class T >

T* gpufilter::cuda_new ( size_t elements )

Allocates a new memory space in the GPU.

This function allocates device (GPU) memory space.

Parameters:

[in] elements Number of elements to allocate

Returns:: Pointer to the device memory allocated

Template Parameters:

T	Memory values type

template<class T >

std::complex<T> gpufilter::ds	(	const std::complex< T > &	d,
		const T &	s
	)

Rescale poles of the recursive filtering z-transform.

Given a complex-valued pole on |z|=1 ROC of the recursive filter z-transform compute a rescaled pole representing a true Gaussian filter convolution with arbitrary support sigma.

See also:: [vanVliet:1998] cited in weights2()

Parameters:

[in]	d	Complex-valued pole of a stable recursive filter
[in]	s	Sigma support of the true Gaussian filter

Returns:: Rescaled complex-valued pole of the recursive filter approximation

Template Parameters:

T	Sigma value type

template<class T >

T gpufilter::ds	(	const T &	d,
		const T &	s
	)

Rescale poles in the real-axis of the recursive filtering z-transform.

Given a real pole on |z|=1 ROC of the recursive filter z-transform compute a rescaled pole representing a true Gaussian filter convolution with arbitrary support sigma.

See also:: [vanVliet:1998] cited in weights2()

Parameters:

[in]	d	Real pole of a stable recursive filter
[in]	s	Sigma support of the true Gaussian filter

Returns:: Rescaled real pole of the recursive filter approximation

Template Parameters:

T	Sigma value type

bool gpufilter::extend	(	const int &	w,
		const int &	h,
		const int &	extb
	)

Verify if an image needs to be extended.

Given the image size and an extension to consider verifies if the image should be extended to met the requirement of each dimension be a multiple of 32.

Parameters:

[in]	w	Width of the image
[in]	h	Height of the image
[in]	extb	Extension (in blocks) to consider outside image (default 0)

Returns:: True if corresponding image should be extended

template<class T >

T gpufilter::lookat	(	const T *	in,
		const int &	i,
		const int &	n,
		const initcond &	ic,
		const int &	p = `1`
	)

Look in an input at a given index range.

This function looks in an input array at an arbitrary index (possibly outside the input array) using a given initial condition and a pitch value for arrays with 2D geometry.

Parameters:

[in]	in	Input array
[in]	i	Index to look at
[in]	n	Input range [0,n)
[in]	ic	Initial condition
[in]	p	Pitch (e.g. image width represented by the input array)

Returns:: Value at given position

Template Parameters:

T	Input array value type

template<class T >

T gpufilter::lookat	(	const T *	img,
		const int &	i,
		const int &	j,
		const int &	h,
		const int &	w,
		const initcond &	ic
	)

Look in an image at a given position.

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts. This function looks in an input image at an arbitrary position (possibly outside the image) using a given initial condition and the 2D image size.

Parameters:

[in]	img	Input image
[in]	i	Row index
[in]	j	Column index
[in]	h	Image height
[in]	w	Image width
[in]	ic	Initial condition

Returns:: Value at given position

Template Parameters:

T	Input image value type

template<class T >

T gpufilter::qs ( const T & s )

Compute recursive filtering scaling factor.

Compute the scaling factor of the recursive filter representing a true Gaussian filter convolution with arbitrary support sigma.

See also:: [vanVliet:1998] cited in weights2()

Parameters:

[in] s Sigma support of the true Gaussian filter

Returns:: Scaling factor q of the recursive filter approximation

Template Parameters:

T	Sigma value type

template<class T1 , class T2 >

void gpufilter::weights1	(	const T1 &	s,
		T2 &	b0,
		T2 &	a1
	)

Compute first-order weights.

Given a Gaussian sigma value compute the feedforward and feedback first-order coefficients.

See also:: [vanVliet:1998] cited in weights2()

Parameters:

[in]	s	Gaussian sigma
[out]	b0	Feedforward coefficient
[out]	a1	Feedback first-order coefficient

Template Parameters:

T1	Sigma value type
T2	Coefficients value type

template<class T1 , class T2 >

void gpufilter::weights2	(	const T1 &	s,
		T2 &	b0,
		T2 &	a1,
		T2 &	a2
	)

Compute first- and second-order weights.

Given a Gaussian sigma value compute the feedforward and feedback first- and second-order coefficients. Refer to the following paper for more information:

@inproceedings{vanVliet:1998,
  author = {L. J. {van Vliet} and I. T. Young and P. W. Verbeek},
  title = {Recursive {Gaussian} derivative filters},
  booktitle = {Proceedings of the 14th International Conference on Pattern Recognition},
  year = {1998},
  pages = {509--514 (v. 1)}
}

Parameters:

[in]	s	Gaussian sigma
[out]	b0	Feedforward coefficient
[out]	a1	Feedback first-order coefficient
[out]	a2	Feedback second-order coefficient

Template Parameters:

T1	Sigma value type
T2	Coefficients value type

Classes

Enumerations

Functions

Variables

Enumeration Type Documentation

Function Documentation