mln::registration Namespace Reference

Namespace of "point-wise" expression tools. More...

Classes
class	closest_point_basic
	Closest point functor based on map distance. More...
class	closest_point_with_map
	Closest point functor based on map distance. More...
Functions
template<typename P, typename F>
algebra::quat	get_rot (const p_array< P > &P_, const vec3d_f &mu_P, const vec3d_f &mu_Yk, const F &closest_point, const algebra::quat &qR, const vec3d_f &qT)
	FIXME: work only for 3d images.
template<typename P, typename F>
composed< translation< P::dim, float >, rotation< P::dim, float > >	icp (const p_array< P > &P_, const p_array< P > &X, const F &closest_point)
template<typename P, typename F>
std::pair< algebra::quat, mln_vec(P)>	icp (const p_array< P > &P_, const p_array< P > &X, const F &closest_point, const algebra::quat &initial_rot, const mln_vec(P)&initial_translation)
	Base version of the ICP algorithm. It is called in other variants.
template<typename P>
composed< translation< P::dim, float >, rotation< P::dim, float > >	registration1 (const box< P > &domain, const p_array< P > &P_, const p_array< P > &X)
	Call ICP once and return the resulting transformation.
template<typename P>
composed< translation< P::dim, float >, rotation< P::dim, float > >	registration2 (const box< P > &domain, const p_array< P > &P_, const p_array< P > &X)
	Call ICP 10 times.
template<typename P>
composed< translation< P::dim, float >, rotation< P::dim, float > >	registration3 (const box< P > &domain, const p_array< P > &P_, const p_array< P > &X)
	Call ICP 10 times.

---

Detailed Description

Namespace of "point-wise" expression tools.

---

Function Documentation

template<typename P, typename F>

algebra::quat mln::registration::get_rot	(	const p_array< P > &	P_,
		const vec3d_f &	mu_P,
		const vec3d_f &	mu_Yk,
		const F &	closest_point,
		const algebra::quat &	qR,
		const vec3d_f &	qT
	)			[inline]

FIXME: work only for 3d images.

References mln::p_array< P >::nsites().

template<typename P, typename F>

composed< translation<P::dim,float>,rotation<P::dim,float> > mln::registration::icp	(	const p_array< P > &	P_,
		const p_array< P > &	X,
		const F &	closest_point
	)			[inline]

Register point in c using a function of closest points closest_point.

Parameters:

[in]	P_	The cloud of points.
[in]	X	the reference surface.
[in]	closest_point	The function of closest points.

Returns:

the rigid transformation which may be use later to create a registered image.

template<typename P, typename F>

std::pair< algebra::quat, mln_vec(P)> mln::registration::icp	(	const p_array< P > &	P_,
		const p_array< P > &	X,
		const F &	closest_point,
		const algebra::quat &	initial_rot,
		const mln_vec(P)&	initial_translation
	)			[inline]

Base version of the ICP algorithm. It is called in other variants.

Register point in c using a function of closest points closest_point. This overload allows to specify initial transformations.

Parameters:

[in]	P_	The cloud of points.
[in]	X	the reference surface.
[in]	closest_point	The function of closest points.
[in]	initial_rot	An initial rotation.
[in]	initial_translation	An initial translation.

Returns:

the rigid transformation which may be use later to create a registered image.

WARNING: the function closest_point *MUST* take float/double vector as arguments. Otherwise the resulting transformation may be wrong due to the truncation of the vector coordinate values.

Precondition:

P_ and X must not be empty.

Reference article: "A Method for Registration of 3-D Shapes", Paul J. Besl and Neil D. McKay, IEEE, 2, February 1992.

References mln::geom::bbox(), mln::literal::black, mln::set::compute(), mln::duplicate(), mln::box< P >::enlarge(), mln::data::fill(), mln::literal::green, mln::io::ppm::save(), and mln::literal::white.

template<typename P>

composed< translation< P::dim, float >, rotation< P::dim, float > > mln::registration::registration1	(	const box< P > &	domain,
		const p_array< P > &	P_,
		const p_array< P > &	X
	)			[inline]

Call ICP once and return the resulting transformation.

template<typename P>

composed< translation< P::dim, float >, rotation< P::dim, float > > mln::registration::registration2	(	const box< P > &	domain,
		const p_array< P > &	P_,
		const p_array< P > &	X
	)			[inline]

Call ICP 10 times.

Do the first call to ICP with all sites then work on a subset of which size is decreasing. For each call, a distance criterion is computed on a subset. Sites part of the subset which are too far or too close are removed. Removed sites are *NOT* reused later in the subset.

template<typename P>

composed< translation< P::dim, float >, rotation< P::dim, float > > mln::registration::registration3	(	const box< P > &	domain,
		const p_array< P > &	P_,
		const p_array< P > &	X
	)			[inline]

Call ICP 10 times.

Do the first call to ICP with all sites then work on a subset. For each call, a distance criterion is computed on a subset. A new subset is computed from the whole set of points according to this distance. It will be used in the next call. Removed Sites *MAY* be reintegrated.