#include <osl_edge_detector.h>

Inheritance diagram for osl_edge_detector:

Public Member Functions
	osl_edge_detector (osl_edge_detector_params const &)
	~osl_edge_detector ()
void	detect_edges (vil1_image const &image, vcl_list< osl_edge * > *edges, bool maintain_topology=true)
Public Attributes
float	sigma_
float	low_
float	gauss_tail_
bool	fill_gaps_
bool	verbose_
Private Member Functions
void	Sub_pixel_interpolation ()
	A procedure that performs sub-pixel interpolation for all edges greater than the threshold by parabolic fitting.
void	Thicken_threshold (int, int)
	Thickens the threshold image around each good pixel to take account for the smoothing kernel (almost a dilation with a square structuring element).
void	Set_thresholds ()
	Takes the thresh_ image that contains threshold values near to where non-maximal suppression succeeded, and zero elsewhere, and extend the values to all areas of the image.
void	Fill_holes ()
	Finds all pixels that are surrounded by four edgels, but which are themselves not edgels.
void	Thin_edges ()
	Method to thin the image using the variation of Tsai-Fu thinning used by Van-Duc Nguyen in Geo-Calc.
void	Follow_curves (vcl_list< osl_edge * > *)
	Follow all edgel chains that have pixel values above their corresponding threshold values (thin_[x][y] > thresh_[x][y]).
void	Follow (int, int, vcl_list< int > , vcl_list< int > , vcl_list< float > *, int)
	Following routine looking for connectiveness of edgel chains, and accounts for single pixel gaps in the chains.
void	Find_junctions ()
	Searches for the junctions in the image.
void	Find_junction_clusters ()
	Locate junction clusters using the following method of hysteresis.
void	Follow_junctions (int, int, vcl_list< int > , vcl_list< int > )
	Following routine looking for searching out junction clusters.
void	Cluster_centre (vcl_list< int > &, vcl_list< int > &, int &, int &)
	Finds which member of the lists lies closest to the centre of the list.
Private Attributes
unsigned int	width_
unsigned int	k_size_
float *	kernel_
unsigned int	xstart_
unsigned int	ystart_
unsigned int	xsize_
unsigned int	ysize_
float **	dx_
float **	dy_
float **	grad_
float **	smooth_
float **	thin_
float **	theta_
float **	thresh_
int **	dist_
int **	junction_
int **	jx_
int **	jy_
vcl_list< int > *	xjunc_
vcl_list< int > *	yjunc_
vcl_list< osl_Vertex * > *	vlist_
float	jval_
int	chain_no_
int	vertidcount_
bool	gradient_histogram_
float	max_gradient_
int	histogram_resolution_

Detailed Description

Definition at line 63 of file osl_edge_detector.h.

Constructor & Destructor Documentation

osl_edge_detector::osl_edge_detector ( osl_edge_detector_params const & params )

Definition at line 33 of file osl_edge_detector.cxx.

osl_edge_detector::~osl_edge_detector ( )

Definition at line 55 of file osl_edge_detector.cxx.

Member Function Documentation

void osl_edge_detector::Cluster_centre	(	vcl_list< int > &	xc,
		vcl_list< int > &	yc,
		int &	x0,
		int &	y0
	)		`[private]`

Finds which member of the lists lies closest to the centre of the list.

Definition at line 977 of file osl_edge_detector.cxx.

void osl_edge_detector::detect_edges	(	vil1_image const &	image,
		vcl_list< osl_edge * > *	edges,
		bool	maintain_topology = `true`
	)

Definition at line 80 of file osl_edge_detector.cxx.

void osl_edge_detector::Fill_holes ( ) [private]

Finds all pixels that are surrounded by four edgels, but which are themselves not edgels.

These `holes' cause the construction of complex topological descriptions. To simplify matters, we raise the thin_ value of the central pixel and so force it to be an edgel.

Definition at line 590 of file osl_edge_detector.cxx.

void osl_edge_detector::Find_junction_clusters ( ) [private]

Locate junction clusters using the following method of hysteresis.

Definition at line 882 of file osl_edge_detector.cxx.

void osl_edge_detector::Find_junctions ( ) [private]

Searches for the junctions in the image.

Definition at line 845 of file osl_edge_detector.cxx.

void osl_edge_detector::Follow	(	int	x,
		int	y,
		vcl_list< int > *	xc,
		vcl_list< int > *	yc,
		vcl_list< float > *	grad,
		int	reverse
	)		`[private]`

Following routine looking for connectiveness of edgel chains, and accounts for single pixel gaps in the chains.

Definition at line 773 of file osl_edge_detector.cxx.

void osl_edge_detector::Follow_curves ( vcl_list< osl_edge * > * edges ) [private]

Follow all edgel chains that have pixel values above their corresponding threshold values (thin_[x][y] > thresh_[x][y]).

Definition at line 620 of file osl_edge_detector.cxx.

void osl_edge_detector::Follow_junctions	(	int	x,
		int	y,
		vcl_list< int > *	xc,
		vcl_list< int > *	yc
	)		`[private]`

Following routine looking for searching out junction clusters.

Definition at line 946 of file osl_edge_detector.cxx.

void osl_edge_detector::Set_thresholds ( ) [private]

Takes the thresh_ image that contains threshold values near to where non-maximal suppression succeeded, and zero elsewhere, and extend the values to all areas of the image.

This is done using chamfer masks so that the final threshold assigned at any one point (ie. a point that was initially zero) is functionally dependent on the strengths of the nearest good edges. At present we linearly interpolate between the two (approximately) closest edges.

Try to do the same process using Delauney triangulation (CAR, March 1995), in an attempt to image the efficiency from a memory management point of view. However, the triangulation becomes so complex that the computation time becomes incredibly long. Therefore putting up with the Chamfer method for the moment.

The histogram calculation was added to support edgel change detection-JLM May 1995

Definition at line 383 of file osl_edge_detector.cxx.

void osl_edge_detector::Sub_pixel_interpolation ( ) [private]

A procedure that performs sub-pixel interpolation for all edges greater than the threshold by parabolic fitting.

Writes edges into the thresh_ image if they are maxima and above low_. This gives a good indication of the local edge strengths. Stores sub-pixel positions in dx_ and dy_, and set the orientations in theta_.

Definition at line 184 of file osl_edge_detector.cxx.

void osl_edge_detector::Thicken_threshold	(	int	x,
		int	y
	)		`[private]`

Thickens the threshold image around each good pixel to take account for the smoothing kernel (almost a dilation with a square structuring element).

Definition at line 346 of file osl_edge_detector.cxx.

void osl_edge_detector::Thin_edges ( ) [private]

Method to thin the image using the variation of Tsai-Fu thinning used by Van-Duc Nguyen in Geo-Calc.

This relies on computing the genus of an edge location, and removing it if it is not a dangling chain as has genus zero. We also order the edges by strength and try to remove the weaker ones first. This accounts for non-maximal suppression, and does it in a topology preserving way. Note that we are creating a vcl_list with a large number of elements, and then sorting it - this is likely to be quite slow. An alternative implementation would be better.

Definition at line 506 of file osl_edge_detector.cxx.