Component labeling in pyramid architecture

摘要

A new pyramid algorithm for labeling connected components of an image is presented. The algorithm uses explicit node to node connections in a pyramid architecture for data transfer operations and does not assume any complex pyramid machine with random access read (RAR) or random access write (RAW) capabilities. The algorithm is more practical from a programming point of view as it is easily implementable on some of the existing pyramid machines (Schaeffer, Proc. 18th Annual Hawaii Conf. on Systems Science, Vol. 1, pp. 178–184 (1985); Tanomoto et al., Parallel Computer Vision, pp. 43–83. Academic Press, Orlando, Florida (1987)). The time complexity of the algorithm is O(s log n) for n × n image size and s components in the image. Moreover, the algorithm is applicable on both binary and multilevel images and hence is useful for segmentation of multilevel images. Experimental results on binary and range images are cited in this paper to highlight distinctive features of the algorithm.