On combining graph-partitioning with non-parametric clustering for image segmentation

A Segmentation Algorithm for Noisy Images

“…One of the most frequently used techniques to partition a graph is by means of the cut cost function. Several alternatives to the cut criterion have been proposed [1][2][3]. Of particular note is the normalized cut criterion (Ncut) of Shi and Malik [1], which attempts to rectify the tendency of the cut algorithm to favor isolated nodes of the graph.…”

“…Several algorithms have been introduced to tackle this problem. Among them are approaches based on graph partitioning [1][2][3]. Their common point is the building of a weighted graph.…”

Efficient Closed Contour Extraction from Range Image’s Edge Points

“…Graph theory has long been used in the 2D image segmentation problem (e.g., [5], [7], [8], [9], [10], [11]). Our proposed approach differs from them not only due to the fact that it is focused on range image processing but also in the following aspects.…”

“…Our proposed approach differs from them not only due to the fact that it is focused on range image processing but also in the following aspects. Firstly, some of those techniques were meant for partitioning a gray-level image into connected homogeneous regions-region-based approaches ( [7], [8], [9])-; for example, [7] introduces a 2D image segmentation algorithm using minimum spanning trees. By minimizing the sum of gray levels variations a minimum spanning tree is partitioned into subtrees, representing different homogeneous regions.…”

“…By minimizing the sum of gray levels variations a minimum spanning tree is partitioned into subtrees, representing different homogeneous regions. Similarly, [8] proposes a graph-partitioning with non-parametric clustering approach for 2D image segmentation. Secondly, those approaches, proposed to extract closed contours, were intended to compute the external object boundary ( [5], [10]), but not closed boundaries of the object's regions.…”