ImFusion::DistanceTransform Class Reference

#include <ImFusion/Base/DistanceTransform.h>

Distance transformation algorithm. More...

Detailed Description

Distance transformation algorithm.

Static Public Member Functions
static std::unique_ptr< TypedImage< float > >	computeChamfer (const TypedImage< unsigned char > &mask, bool inverse=false, bool considerSpacing=false, int thresholdMask=0)
	Computes a fast Chamfer distance transformation from mask of non-zero values.
static void	extendChamfer (TypedImage< float > &result, bool considerSpacing, float threshold=0.0)
	Extends a given distance-volume by performing a forward-backward Chamfer pass.
static std::unique_ptr< TypedImage< float > >	computeChamferSigned (const TypedImage< unsigned char > &mask, bool considerSpacing=false, int thresholdMask=0, double insideWeight=1.0)
	Version of the method above which also computes distances inside the object (whose negative values are scaled with insideWeight).
static std::unique_ptr< TypedImage< float > >	computeSequentialEuclidean (const TypedImage< unsigned char > &mask, int iterations=-1)
	Computes an absolute Sequential Euclidean Distance Transform.
static std::unique_ptr< TypedImage< float > >	computePreciseSigned (const Mesh &input, double resolution, int fillZ=0, double insideWeight=1.0, double offset=3.0)
	Compute signed distance volume with sub-pixel precise distances in narrow band about surface.
static std::unique_ptr< TypedImage< float > >	computePreciseSigned (const Mesh &input, const vec3 &resolution, int fillZ=0, double insideWeight=1.0, double offset=3.0, vec3i *targetDim=nullptr, TypedImage< unsigned char > *inputMask=nullptr, bool distInPixels=true, int marginPx=10)
static std::unique_ptr< TypedImage< float > >	computePreciseSignedEigen (const Mesh &input, const vec3 &resolution, int fillZ=0, double insideWeight=1.0, double offset=3.0, vec3i *targetDim=nullptr, TypedImage< unsigned char > *inputMask=nullptr, bool distInPixels=true, int marginPx=10)
	Compute precise signed distance transform using portable Eigen-based vectorization.
static std::unique_ptr< TypedImage< float > >	computePreciseAbsolute (const Mesh &input, const vec3 &resolution, double offset=3.0)
	Compute absolute distance volume with sub-pixel precise distances in narrow band about surface.
static std::unique_ptr< TypedImage< float > >	computePreciseAbsolute (const Mesh &input, double resolution, double offset=3.0)
static std::unique_ptr< TypedImage< float > >	computeFastSweeping (const TypedImage< unsigned char > &label, bool considerSpacing)
	Compute a distance volume with the fast sweeping method (https://en.wikipedia.org/wiki/Fast_sweeping_method).

Member Function Documentation

computeChamfer()

std::unique_ptr< TypedImage< float > > ImFusion::DistanceTransform::computeChamfer ( const TypedImage< unsigned char > & mask, bool inverse = false, bool considerSpacing = false, int thresholdMask = 0 )

static

Computes a fast Chamfer distance transformation from mask of non-zero values.

Parameters

mask	Input binary image, all values > threshold will be interpreted as foreground.
inverse	If false, will compute the distances of the background to the foreground. If true, will compute the distances of the foreground to the background.
considerSpacing	Flag whether to approximate image distances (incorporating pixel spacing) instead of pixel distances.
thresholdMask	Optional threshold value for the mask, larger than it is considered foreground

Note

The resulting distance map contains Chamfer distances that try to approximate Euclidean distances but are not exact. As a rule of thumb you can assume the resulting distances to underestimate the exact Euclidean distance by up to 10% for pixel distances and 15% for image distances (incorporating pixel spacing).

extendChamfer()

void ImFusion::DistanceTransform::extendChamfer ( TypedImage< float > & result, bool considerSpacing, float threshold = 0.0 )

static

Extends a given distance-volume by performing a forward-backward Chamfer pass.

Parameters

threshold

- pixels with value less than this parameter are not updated. Ignored if zero.

Warning

The threshold is applied during both the forward and the backward pass. If the value difference of neighboring pixels exceeds their chamfer distance, a pixel value originally above threshold can become smaller than threshold after the forward pass and as a result be incorrectly ignored during the backward pass

computeSequentialEuclidean()

std::unique_ptr< TypedImage< float > > ImFusion::DistanceTransform::computeSequentialEuclidean ( const TypedImage< unsigned char > & mask, int iterations = -1 )

static

Computes an absolute Sequential Euclidean Distance Transform.

This function is more accurate than computeChamfer() but also more expensive to compute.

Unless a specific number of iterations is requested, this will run until it has converged to a final solution (typically 3-4 iterations). Note that the result is not always exact, as the scheme will fail in case of disconnected Voronoi regions. See 7.3 in http://www.agencia.fapesp.br/arquivos/survey-final-fabbri-ACMCSurvFeb2008.pdf For random 2D input this occurs roughly every 10^7 pixels and leads to an error of typically 0.05 pixels.

computePreciseSigned() [1/2]

std::unique_ptr< TypedImage< float > > ImFusion::DistanceTransform::computePreciseSigned ( const Mesh & input, double resolution, int fillZ = 0, double insideWeight = 1.0, double offset = 3.0 )

static

Compute signed distance volume with sub-pixel precise distances in narrow band about surface.

The region inside the object is weighted by the factor insideWeight The input mesh has to be closed for this method to work If it is not the case, the parameter fillZ can be set to 1 (resp. -1) in order to close the mesh in the upper (resp. lower) Z-part

computePreciseSigned() [2/2]

std::unique_ptr< TypedImage< float > > ImFusion::DistanceTransform::computePreciseSigned ( const Mesh & input, const vec3 & resolution, int fillZ = 0, double insideWeight = 1.0, double offset = 3.0, vec3i * targetDim = nullptr, TypedImage< unsigned char > * inputMask = nullptr, bool distInPixels = true, int marginPx = 10 )

static

Parameters

inputMask	Optional mask of the input mesh, will be recomputed if not provided.
distInPixels	If true, the distances will be computed in pixel units. Otherwise it will be in metric units.
offset	Distances smaller than this (in pixels, if distInPixels is set, otherwise metric) will be computed precisely from the Mesh, the further distances are obtained from this via the Chamfer distance. If spacing is not isotropic, metric distance is recommended.

computePreciseSignedEigen()

std::unique_ptr< TypedImage< float > > ImFusion::DistanceTransform::computePreciseSignedEigen ( const Mesh & input, const vec3 & resolution, int fillZ = 0, double insideWeight = 1.0, double offset = 3.0, vec3i * targetDim = nullptr, TypedImage< unsigned char > * inputMask = nullptr, bool distInPixels = true, int marginPx = 10 )

static

Compute precise signed distance transform using portable Eigen-based vectorization.

This implementation uses Eigen's vectorization capabilities which auto-vectorize on all platforms (NEON on ARM, SSE/AVX on x86) without platform-specific intrinsics.

See also

computePreciseSignedThreadpool for detailed parameter documentation

computePreciseAbsolute()

std::unique_ptr< TypedImage< float > > ImFusion::DistanceTransform::computePreciseAbsolute ( const Mesh & input, const vec3 & resolution, double offset = 3.0 )

static

Compute absolute distance volume with sub-pixel precise distances in narrow band about surface.

The input mesh can be either closed or open

computeFastSweeping()

std::unique_ptr< TypedImage< float > > ImFusion::DistanceTransform::computeFastSweeping ( const TypedImage< unsigned char > & label, bool considerSpacing )

static

Compute a distance volume with the fast sweeping method (https://en.wikipedia.org/wiki/Fast_sweeping_method).

Computes the (unsigned) distance transform to pixels that are nonzero in label.

Parameters

`label`	the input label map, where pixels that are non-zero yield zero values in the distance map.
`considerSpacing`	if true, the distance is computed in metric units, otherwise in pixel units. For nonisotropic spacing, distance values may be inexact relative to the precision obtained for isotropic spacing.

---

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

• ImFusion/Base/DistanceTransform.h