Operators

points_sojka (Operator)

Name

points_sojka — Find corners using the Sojka operator.

Description

points_sojka defines a corner as the point of intersection of two straight, non-collinear gray value edges. To decide whether a point of the input image Image is a corner or not, a neighborhood of MaskSize x MaskSize points is inspected. Only those image regions that are relevant for the decision are considered. Pixels with a magnitude of the gradient of less than MinGrad are ignored from the outset.

Furthermore, only those of the remaining points are used that belong to one of the two gray value edges that form the corner. For this, the so called Apparentness is calculated, which is an indicator of the probability that the examined point actually is a corner point. Essentially, it is determined by the number of relevant points and their gradients. A point can only be accepted as a corner when its Apparentness is at least MinApparentness. Typical values of MinApparentness should range in the region of a few multiples of MinGrad.

To calculate the Apparentness, each mask point is weighted according to two criteria: First, the influence of a mask point is weighted with a Gaussian of size SigmaW according to its distance from the possible corner point. SigmaW should be roughly a quarter to the half of MaskSize to obtain a reasonable proportion of the size of the weighting function to the mask size. Secondly, the distance of the point from the (assumed) ideal gray value edge is estimated and the point is weighted with a Gaussian of size SigmaD according to that distance. I.e., pixels that (due to the discretization of the input image) lie farther from the ideal gray value edge have less influence on the result than pixels with a smaller distance. Typically, it is not necessary to modify the default value 0.75 of SigmaD .

As a further criterion, the angle is calculated, by which the gray value edges change their direction in the corner point. A point can only be accepted as a corner when this angle is greater than MinAngle.

The position of the detected corner points is returned in (Row, Column). Row and Column are calculated with subpixel accuracy if Subpix is 'true'. They are calculated only with pixel accuracy if Subpix is 'false'.

Parallelization

• Multithreading type: reentrant (runs in parallel with non-exclusive operators).
• Processed without parallelization.

Parameters

Image (input_object)  singlechannelimage object (byte / int1 / int2 / uint2 / int4 / real)

Input image.

Required filter size.

Default value: 9

List of values: 5, 7, 9, 11, 13

SigmaW (input_control)  number (real / integer)

Sigma of the weight function according to the distance to the corner candidate.

Default value: 2.5

Suggested values: 2.0, 2.2, 2.4, 2.5, 2.6, 2.8, 3.0

SigmaD (input_control)  number (real / integer)

Sigma of the weight function for the distance to the ideal gray value edge.

Default value: 0.75

Suggested values: 0.6, 0.7, 0.75, 0.8, 0.9, 1.0

MinGrad (input_control)  number (real / integer)

Threshold for the magnitude of the gradient.

Default value: 30.0

Suggested values: 20.0, 15.0, 30.0, 35.0, 40.0

MinApparentness (input_control)  number (real / integer)

Threshold for Apparentness.

Default value: 90.0

Suggested values: 30.0, 60.0, 90.0, 150.0, 300.0, 600.0, 1500.0

Threshold for the direction change in a corner point (radians).

Default value: 0.5

Restriction: 0.0 <= MinAngle && MinAngle <= pi

Subpix (input_control)  string (string)

Subpixel precise calculation of the corner points.

Default value: 'false'

List of values: 'false', 'true'

Row (output_control)  point.y-array (real)

Row coordinates of the detected corner points.

Column (output_control)  point.x-array (real)

Column coordinates of the detected corner points.

Result

points_sojka returns 2 (H_MSG_TRUE) if all parameters are correct and no error occurs during the execution. If the input is empty the behavior can be set via set_system('no_object_result',<Result>). If necessary, an exception is raised.

References

Eduard Sojka: “A New and Efficient Algorithm for Detecting the Corners in Digital Images”. Pattern Recognition, Luc Van Gool (Editor), LNCS 2449, pp. 125-132, Springer Verlag, 2002.

Module

Foundation

 Operators