A skeleton pruning method based on saliency sorting

Abstract

The pruning of spurious skeletons is an important issue when apply skeletons for object analysis and recognition. Most existing pruning methods use threshold-like parameter to control the effects of pruning. Such approaches usually encounter difficulties in parameter tuning and the quality of pruning. A skeleton pruning method is proposed in this work. This method decomposes a skeleton into a number of skeletal components (SCs), and the terminal SCs are removed one by one according to a saliency measure. SCs may be merged into a new one after each removal. The removal process continues until the desirable number of terminals are achieved. The method can essentially be regarded as the sorting of the terminals, and the truncation of the terminals at the given number. This method is suitable for applications where the desired number of terminals can be determined a priori. Experimental results show that the method can effectively prune the skeletons from the test image bases, and it has the advantage that the terminal number parameter is very intuitive and easy to set for the applications under concern.