Probabilistic relaxation labelling in line and edge detection
A Maheswaran, JA Richards · 1st IASTED International Symposium on Signal Processing and its Applications · 1987
Probabilistic relaxation labelling procedures have been applied to edge and line detected imagery to remove edge or line segments that are inconsistent with their local neighbourhoods. These procedures are iterative in nature and have been shown to be successful in these applications and in pixel classification. However, often a deterioration in labelling is observed to take place with number of iterations, after an initial improvement in labelling accuracy. In the case of pixel classification this has been shown in the past to be the result of a degeneration of the process to a weighted averaging in the vicinity of its fixed points. Consequently, careful choice of the parameters available can be made to overcome the deterioration. Here it is shown how a similar choice of the parameters in the relaxation labelling algorithm can be made also for edge and line enhancement to avoid loss of labelling accuracy with iteration. In particular, it is seen how line thickening can be avoided.