No global_affine anymore zxhTransformLocalAffines.h: interface for the zxhTransformLocalAffines class. T = first Region_Affines,then Global_Affine = G ( R(x) ) R= sum wi(Gi)+w_facets, these {wi} are computed from the distance_transform images, hence it is better to have the facets fixed on the test image boundary, because local_affine is identity outside the test image facet or fixed facet w_facets is regarded as 6*ZXH_LocalAffineFacetWeighting identity local affine Length_finite_difference = Length_advance < resample/2 Compute similarity measures, e.g. histogram table, i regions based on Ri, and background regions based on interpolated R(x) For each region i Update the histogram table with finite difference step to compute the gradient against affine parameters; Update Ri R(x) is overlap corrected distance weighting interpolation between Ri; If need overlap correction, then update the regions and recomputed the distance transformation for them Local Affine interpolation: (1) based on affine (2) based on closest points' displacment GRADIENT all 1 based on influence_object. More...

#include <zxhTransformLocalAffines.h>

Inheritance diagram for zxhTransformLocalAffines:

List of all members.

Public Member Functions
	zxhTransformLocalAffines ()
	constructor
virtual	~zxhTransformLocalAffines ()
	deconstructor
virtual bool	SetTransformFromStream (std::ifstream &ifs)
	from stream
virtual std::string	GetPrintString ()
	set transformation as a identity transform, affine doesn't change centre
virtual std::string	GetTransformType ()
virtual zxhTransformBase *	Clone (zxhTransformBase *&pRet)
virtual bool	PreComputeDistance2WeightAndSetModel ()
virtual void	TransformWorldToWorld (const float fVectorFromWorld[ImageDimensionMax], float fVectorToWorld[ImageDimensionMax])
virtual void	LocalTransformPointToWorld (float fVectorFrom[ImageDimensionMax], float fVectorToWorld[ImageDimensionMax])
virtual void	LocalTransformWorldToWorld (const float fVectorFromWorld[ImageDimensionMax], float fVectorToWorld[ImageDimensionMax])
virtual void	LocalTransformWorldToWorldByDisplace (const float fVectorFromWorld[ImageDimensionMax], float fVectorToWorld[ImageDimensionMax])
virtual void	LocalTransformWorldToWorldByAffine (const float fVectorFromWorld[ImageDimensionMax], float fVectorToWorld[ImageDimensionMax])
virtual bool	GetJacobianMatrixWorld (float fWorld[ImageDimensionMax], float pJacobian[ImageDimensionMax *ImageDimensionMax])
virtual bool	GetLocalAffineJacobianMatrixByWorld (float fWorld[ImageDimensionMax], float pJacobian[ImageDimensionMax *ImageDimensionMax])
	only give the jacobian of the local affines interpolation, exclude global
virtual bool	GetLocalAffineJacobianMatrixByWorldForByDisplace (float fWorld[ImageDimensionMax], float pJacobian[ImageDimensionMax *ImageDimensionMax])
	finite difference
virtual float	GetMinLocalAffineJacobian (float wFrom[], float wTo[], float wIntval[], zxhImageData pMaskTest=0, zxhImageData pMaskRef=0)
virtual float	GetMinLocalAffineJacobianOnTestImage (zxhImageData pImageTest, float tFrom[], float tTo[], float tIntval[], zxhImageData pMaskTest=0, zxhImageData *pMaskRef=0)
virtual void	SetLocalRegions (zxhImageData *p)
virtual zxhImageData *	GetLocalRegions ()
virtual bool	SetFileNameOfLocalRegions (std::string *s)
virtual const std::string *	GetFileNameOfLocalRegions ()
virtual bool	HasSetFileNameOfLocalRegions ()
virtual bool	SetLocalRegionsFromPreSetFileNames ()
virtual void	SetWorldRoi (float wfrom, float wto, int dimension)
	it is important for interpolating transformation, if unset, then is infinate wfrom and wto just two corners, may be wfrom>wto in right-hand coordinate
virtual bool	GetWorldRoi (float wfrom[], float wto[])
	return whether world roi set
virtual bool	WithinInterpolateRegionWorld (const float World[])
	can be interpolate using inverse distance
virtual void	SetInterpolateByDisplacement (bool b)
virtual bool	GetInterpolateByDisplacement ()
virtual void	SetIRBReg (bool b)
virtual bool	GetIRBReg ()
virtual bool	ComputeGradientBasedIRBReg (zxhMetricMILocalAffines pMetric, zxhTransformLocalAffines pInfluence=0)
	Region based registration.
virtual bool	ComputeGradientBasedAccReg (zxhMetricMILocalAffines pMetric, zxhTransformLocalAffines pInfluence=0)
	Accessory region using image and local_region as the target and source images.
virtual bool	ComputeMetricGradient (zxhMetricBase pMetric, zxhTransformBase pInfluence=0)
Protected Member Functions
float	ComputeWeightFromDistance (float dis)
float	ComputeDerivativeByDistance (float dis)
float	ComputeWeightFromDistanceBoundary (float dis)
void	LocalTransformWorldToWorldOutlier (const float fVectorFromWorld[ImageDimensionMax], float fVectorToWorld[ImageDimensionMax])
float	ComputeRoiBoundaryWeightByWorld (const float fWorld[])
	6 facets
void	ComputeDerivativeWeightOfRoiBoundary (float NablaRoiW[], float co[], float fWorld[])
void	PreComputeDistanceImageForLocalRegions ()
void	SetModelWeight ()
	after setting, all resample to same dimension as first image
Protected Attributes
zxhImageData *	m_aLocalRegions
	can have diff spacing, size compared to test image, BUTall local region image should be the same
std::string	m_astrFileNameOfLocalRegions [ZXH_LocalAffineMaxNumber]
bool	m_bSetFileNameOfLocalRegions
zxhImageData	m_aimgLoadLocalImagesFromPreSetFileNames [ZXH_LocalAffineMaxNumber]
zxhImageDataT< float >	m_aLocalDistance2Weight [ZXH_LocalAffineMaxNumber]
	distance transformation for each regions
zxhImageDataT< float >	m_roiDistance2Weight
zxhImageModelingLinearT< float, float >	m_aModelDistance2Weight [ZXH_LocalAffineMaxNumber+1]
	point to the distance images
float	m_afWorldRoi [2][ImageDimensionMax]
bool	m_bSetWorldRoi
bool	m_bInterpolateByDisplacement
bool	m_bIRBReg

Detailed Description

No global_affine anymore zxhTransformLocalAffines.h: interface for the zxhTransformLocalAffines class. T = first Region_Affines,then Global_Affine = G ( R(x) ) R= sum wi(Gi)+w_facets, these {wi} are computed from the distance_transform images, hence it is better to have the facets fixed on the test image boundary, because local_affine is identity outside the test image facet or fixed facet w_facets is regarded as 6*ZXH_LocalAffineFacetWeighting identity local affine Length_finite_difference = Length_advance < resample/2 Compute similarity measures, e.g. histogram table, i regions based on Ri, and background regions based on interpolated R(x) For each region i Update the histogram table with finite difference step to compute the gradient against affine parameters; Update Ri R(x) is overlap corrected distance weighting interpolation between Ri; If need overlap correction, then update the regions and recomputed the distance transformation for them Local Affine interpolation: (1) based on affine (2) based on closest points' displacment GRADIENT all 1 based on influence_object.

Constructor & Destructor Documentation

zxhTransformLocalAffines::zxhTransformLocalAffines ( )

constructor

roi region

Member Function Documentation

zxhTransformBase * zxhTransformLocalAffines::Clone ( zxhTransformBase *& pRet ) [virtual]

Returns:

Reimplemented from zxhTransformMultipleAffines.

Reimplemented in zxhTransformLocalAffines2.

bool zxhTransformLocalAffines::ComputeMetricGradient	(	zxhMetricBase *	pMetric,
		zxhTransformBase *	pInfluence = `0`
	)		`[virtual]`

computer gradient, influence control status for optimisation if infl==0, then use m_fDifferentialStep

Reimplemented from zxhTransformMultipleAffines.

bool zxhTransformLocalAffines::GetJacobianMatrixWorld	(	float	fWorld[ImageDimensionMax],
		float	pJacobian[ImageDimensionMax *ImageDimensionMax]
	)		`[virtual]`

Get partial differential of transformation [Y1 Y2 Y3]=T(x), with respect to the transformation parameter Ti with index return whether success, results (e.g. 3D, [d(Y1)/d(Ti) d(Y2)/d(Ti) d(Y3)/d(Ti)]) returned in g[4]

Jac(global) * Jac(locallyaffine), Y=Glb(Loc(X))

Reimplemented from zxhTransformAffine.

bool zxhTransformLocalAffines::GetLocalAffineJacobianMatrixByWorld	(	float	fWorld[ImageDimensionMax],
		float	pJacobian[ImageDimensionMax *ImageDimensionMax]
	)		`[virtual]`

only give the jacobian of the local affines interpolation, exclude global

Jacobian[0][0..3] dX/dx.dy.dz.dt, zxhtodo, still error

not in regions or boundary, 3D only Joc= (wi/SumWj*NablaTi)+ ([Tix Tiy Tiz]^T * (NablaWi[]/SumWj-wi*SumNablaWj[]/SumWj^2) ) NablaWi[]=-Exponent*di^(-e-1)*Ddi/Dx SumNablaWj[]=-e*( dj^(-e-1)*Ddj/Dx)

Joc= (wi/SumWj*NablaTi)+ ([Tix Tiy Tiz]^T * (NablaWi[]/SumWj-wi*SumNablaWj[]/SumWj^2) )

virtual std::string zxhTransformLocalAffines::GetTransformType ( ) [inline, virtual]

Returns:

Reimplemented from zxhTransformMultipleAffines.

void zxhTransformLocalAffines::LocalTransformPointToWorld	(	float	fVectorFrom[ImageDimensionMax],
		float	fVectorToWorld[ImageDimensionMax]
	)		`[virtual]`

1)ForwardMaskDilatedOverlap 2)ReComputeRegionDistanceTransform

Returns:: the Worldical coordinate after transform, and return whether the Worldical coordinate is outlier of test image Worldical size;

void zxhTransformLocalAffines::LocalTransformWorldToWorld	(	const float	fVectorFromWorld[ImageDimensionMax],
		float	fVectorToWorld[ImageDimensionMax]
	)		`[virtual]`

1) outside interpolation field, use nearest displacement todo 2) Get the distance2weight for each local region and transform using local affine 3) addup the distance2weight from facet and identity transform 4) compute the toWorld

bool zxhTransformLocalAffines::PreComputeDistance2WeightAndSetModel ( ) [virtual]

Finish setting NoLA, LocalRegion, ROIWorld compute distance images, all resample to same dimension as first image

1 set NoLA, LocalRegion, LocalDistance, ROIWorld

2 --- not use nowComputeRoiDistance2Weight 3 --- not use nowCompute the distance images into weights 4 --- not use nowDonotNormalise the Weights, set m_aModelDistance2Weight[NoLA] to RoiWeight, distance 10 error 0.62%

bool zxhTransformLocalAffines::SetTransformFromStream ( std::ifstream & ifs ) [virtual]

from stream

set pretransformation by aligning the image size centre "centre" (2) or gray level centre "gcentre" (3) test and ref images should be set

Reimplemented from zxhTransformMultipleAffines.

Reimplemented in zxhTransformLocalAffines2.

virtual void zxhTransformLocalAffines::SetWorldRoi	(	float *	wfrom,
		float *	wto,
		int	dimension
	)		`[inline, virtual]`

it is important for interpolating transformation, if unset, then is infinate wfrom and wto just two corners, may be wfrom>wto in right-hand coordinate

return no. of pixels masked out by dilation 1. recompute the distance transform 2. m_aCheckOverlapLocalRegions 3. recompute StructRegistration.m_aComputeLocalGradientRegions (zxh::ReComputeLocalGradientRegions NOT in here)

virtual void zxhTransformLocalAffines::TransformWorldToWorld	(	const float	fVectorFromWorld[ImageDimensionMax],
		float	fVectorToWorld[ImageDimensionMax]
	)		`[inline, virtual]`

T = first Region_Affines,then Global_Affine = G ( Tj(R(x)) ) need to ForwardMaskDilatedOverlap with true first before use tranform

Reimplemented from zxhTransformAffine.

Member Data Documentation

float zxhTransformLocalAffines::m_afWorldRoi[2][ImageDimensionMax] [protected]

roi region, fixed boundaries for interpolation, default:-ZXH_InfiniteLargeFloat~ZXH_InfiniteLargeFloat

The documentation for this class was generated from the following files:

zxhTransformLocalAffines.h
zxhTransformAffine.cpp
zxhTransformLocalAffines.cpp

zxhTransformLocalAffines Class Reference

Public Member Functions

Protected Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation