Image threshold selection with Isoperimetric partition

Graph theory has attracted great attention for the problem of image segmentation in recent years. In this paper we deduce the unified form of the normalized cut (Ncut) algorithm and the Isoperimetric algorithm, and then a new Isoperimetric-based thresholding algorithm is proposed. Unlike Tao and Jin's Ncut-based thresholding algorithm, the proposed algorithm need not to search all possible thresholds, nevertheless yields similar results. A large number of examples are presented to show the effectiveness of the proposed algorithm.