Catalogue of Artificial Intelligence Techniques

   

Jump to: Top | Entry | References | Comments

View Maths as: Images | MathML

Boundary Detection

Aliases: Curve Detection, Line Finding

Keywords: boundary tracking, transform-space

Categories: Pattern Recognition and Image Processing


Author(s): Bob Beattie

The conventional approach to Image Segmentation in early visual processing is Edge Detection followed by boundary detection. Edge detection produces primitive edge elements, possibly with properties of magnitude and/or direction, at places in the image where the edge detection operator has `fired'. The task of the boundary detection process is to produce a set of boundaries by appropriately connecting up primitive edge elements. There are two main methods:

Since a boundary is, by definition, a connected set of edge elements, boundary tracking chooses any edge element and looks for its neighbours, which are then connected to the initial element to form a boundary. This search process is then continually repeated for the elements at the ends of the boundary until some termination criterion is met (such as not being able to find any nearby neighbours). In the Hough technique, each primitive edge element is transformed into a curve in a transform space representing all the possible boundaries the edge element could be part of. The transform-space curves produced by edge elements on the same image boundary should all intersect at the same point in transform space, which can be subjected to the inverse transform to characterise the boundary in the image. Originally applicable to straight boundaries only, this method has been generalised to deal with any curve describable in an analytic or tabular form.


References:


Comments:

Add Comment

No comments.