This algorithm is based on an electro-dynamic analogon and therefore it is not limited to window based operations. A found contour is rapidly traced with a fixed step width from its beginning to its end while collecting additional properties that are added to the contour. The traced contour spots are segmented using least square regression lines and an efficient corner detection algorithm.
Here is a summary of the contour tracer´s properties:
  • extracting whole contours as connection edge lines
  • two step melting of short edge lines, recomputing the regressing line with the resulting contour spots
  • correction of camera internal distortions (Lenz & Tsai [1]) between the two steps of melting
  • local adaptive threshold for edge detecting ensures finding week contours
  • adaptive corner detection algorithm 
  • efficient starting point search 
  • possibility of prioritizing or restricting parts of the image
  • real time capability due to image part priorities, examining important parts of the image first. Thus reaching at least a suboptimal result when the deadline is reached.
  • edge line properties:
    • starting and ending points
    • information about connected lines
    • length of edge line
    • averaged first derivative in x- and y- direction
    • averaged second derivative 
    • type of starting and ending point
  • only an average of 5% percent of all pixels have to be touched to extract the important contours
  • accuracy better than 1 pixel
  • short processing time (<30ms on state-of-the-art computers for 512x512 8Bit grey scale images
  • written under UNIX in C