@article{240663148f764424bdce3f9d5e81c2b8,
title = "A Model-Fitting Approach to Cluster Validation with Application to Stochastic Model-Based Image Segmentation",
abstract = "A clustering scheme is used for model parameter estimation. Most of the existing clustering procedures require prior knowledge of the number of classes, which is often, as in unsupervised image segmentation, unavailable and must be estimated. This problem is known as the cluster validation problem. For unsupervised image segmentation the solution of this problem directly affects the quality of the segmentation. A model-fitting approach to the cluster validation problem based on Akaike's information criterion (AIC) is proposed, and its efficacy and robustness are demonstrated through experimental results for synthetic mixture data and image data.",
keywords = "Cluster validation, computer vision, image interpretation, scene segmentation, texture discrimination",
author = "J. Zhang and Modestino, {J. W.}",
note = "Funding Information: Clustering is a process in which observed data or entities are grouped together to form a number of clusters in such a way that the entities within a cluster are more similar to each other than to those in other clusters. Various clustering procedures have been developed for such diverse fields as biology, statistical data analysis, and pattern recognition [l]. In some previous work on stochastic model-based image segmentation, clustering algorithms have been used to estimate the model parameters for different image classes from the observed image [2]-[4]. The clustering algorithm used in this work was the K-means algorithm [5], [6], a very effective and popular algorithm developed within the pattern recognition community. However, an important problem existing with the K-means algorithm, as is with most clustering algorithms, is that the number of clusters in the data must be specified prior to using the algorithm. In many practical applications, such as our previous work on image segmentation, it is desired to estimate this number directly from the observed data since a priori knowledge is generally not available, and the data are often vectors of dimension higher than two which do not have visually apparent clusters. This problem is of great practical importance for many clustering algorithms, and is known as the cluster validation problem [7]. For stochastic model-based image segmentation, such as the schemes described in [2]-[4], the solution of this problem directly affects the quality of the resulting segmentation. If the estimated number of clusters, or image classes, is too small, different homogeneous regions in Manuscript received December 9, 1987; revised March 26, 1990. Recommended for acceptance by A. K. Jain. This work was supported in part by the U.S. Air Force Rome Air Development Center under Contract F30602-85-C-0008, and in part by the National Science Foundation under Contract ECS8 1-13432. Copyright: Copyright 2015 Elsevier B.V., All rights reserved.",
year = "1990",
month = oct,
doi = "10.1109/34.58873",
language = "English (US)",
volume = "12",
pages = "1009--1017",
journal = "IEEE Transactions on Pattern Analysis and Machine Intelligence",
issn = "0162-8828",
publisher = "IEEE Computer Society",
number = "10",
}