Image Segmentation Based on Distribution of a Measure of Centrality
Weon Geun Oh, Saburo Tsuji · Systems and Computers in Japan · 1989
Abstract Edge detection can be executed by computer at a high speed and thus lends itself to region extraction. However, the edges detected from most images are imperfect due to noise and broken edges. This paper presented an edge‐based region segmentation algorithm using a global measurement called degree of centrality. Methods of computing the degree of centrality from the input image, analyzing its distribution and detecting gaps between edges will be described. The spatial relationship of disjoint boundaries and pseudoregions will be analyzed to determine region splitting and/or merging. The performance of the proposed algorithm is demonstrated via experiment using images of an indoor scene. Advantages of this algorithm include its simple work‐flow, and its ability to perform feature extraction while performing region segmentation.