Clustering is inherently a difficult problem, both with respect to the definition of adequate models as well as to the optimization of the models. We present a model for the cluster problem that does not need knowledge about the number of clusters a priori. This property is among others useful in the image segmentation domain, which we especially address. Further, we propose a cellular coevolutionary algorithm for the optimization of the model. Within this scheme multiple agents are placed in a regular two-dimensional (2-D) grid representing the image, which imposes neighboring relations on them. The agents cooperatively consider pixel migration from one agent to the other in order to improve the homogeneity of the ensemble of the image regions they represent. If the union of the regions of neighboring agents is homogeneous then the agents form alliances. On the other hand, if an agent discovers a deviant subject, it isolates the subject. In the experiments we show the effectiveness of the proposed method and compare it to other segmentation algorithms. The efficiency can easily be improved by exploiting the intrinsic parallelism of the proposed method.

We present a partitional cluster algorithm that minimizes the sum-of-squared-error criterion while imposing a hard constraint on the cluster variance. Conceptually, hypothesized clusters act in parallel and cooperate with their neighboring clusters in order to minimize the criterion and to satisfy the variance constraint. In order to enable the demarcation of the cluster neighborhood without crucial parameters, we introduce the notion of foreign cluster samples. Finally, we demonstrate a new method for cluster tendency assessment based on varying the variance constraint parameter.

Studies the motion correspondence problem for which a diversity of qualitative and statistical solutions exist. We concentrate on qualitative modeling, especially in situations where assignment conflicts arise either because multiple features compete for one detected point or because multiple detected points fit a single feature point. We leave out the possibility of point track initiation and termination because that principally conflicts with allowing for temporary point occlusion. We introduce individual, combined, and global motion models and fit existing qualitative solutions in this framework. Additionally, we present a tracking algorithm that satisfies these-possibly constrained-models in a greedy matching sense, including an effective way to handle detection errors and occlusion. The performance evaluation shows that the proposed algorithm outperforms existing greedy matching algorithms. Finally, we describe an extension to the tracker that enables automatic initialization of the point tracks. Several experiments show that the extended algorithm is efficient, hardly sensitive to its few parameters, and qualitatively better than other algorithms, including the presumed optimal statistical multiple hypothesis tracker.

In this paper, a generalization of parameter hypotheses tests used in statistical inference is presented. This leads to fuzzy hypotheses defined as imprecise predictions about the unknown parameter of a family of distribution functions in question, rather than exactly and sharply given assertions about its location in the parameter space. The notion of fuzzy partitions is discussed in the first part of this paper and details regarding their structure in relation to possibility theoretical interpretations are investigated subsequently. In the second part the authors introduce the notions of fuzzy tests and disclose the impacts of an ill defined counter hypothesis (i.e., hypotheses, partly supporting the same subset of the parameter space) on the reliability of such a fuzzy test. An example demonstrates the consistency of the notion of fuzzy tests with the classical crisp case in statistical inference.>

Describes a new training method, the DRBP-algorithm, for sigmoid-function based multilayer networks. The key step in DRBP is the dynamical selection and autonomous control of the learning rate. Various experiments have shown that the DRBP-algorithm has achieved its goal of fast speed, secure stability and easy parameter selection in practice.>

The usefulness of the fractal Brownian model in combination with pyramidal data structures is discussed. The model enables one to optimize the initialization of the texture pyramid. The ideas which are applied during the initialization phase are also applied during the segmentation. The segmentation algorithm is based on the complete linked pyramid and does not require prior knowledge about the number of segments. The segmentation result is obtained by walking along the links, guided by an error measure. The method does not require any predefined thresholds. All experimental PC program has been developed for the 1-D case.>

In this paper, we present the general methodology for extending different 2-D thresholding techniques to meet 3-D requirements. Several typical thresholding techniques in their 3-D forms have been implemented. The extended techniques have been tested using 3-D synthetic images and the results will be evaluated. Problems related to the extension and improvements made in the implementation will also be discussed.

This correspondence concentrates on the detection of stable clustering results when using the nonparmetric clustering technique of Jarvis and Patrick. This technique incorporates the concept of similarity based on sharing of near neighbors. Essentially, in this clustering scheme, two parameters are involved: neighborhood depth and similarity threshold. Combining a penalty for chaining with the detection of so-called hierarchical stable solutions in the field of all possible solutions, due to the setting of the two parameters, is experimentally shown to provide a powerful method for discriminating between reliable and unreliable results.

Clustering is primarily used to uncover the true underlying structure of a given data set and, for this purpose, it is desirable to subject the same data to several different clustering algorithms. This paper attempts to put an order on the various partitions of a data set obtained from different clustering algorithms. The goodness of each partition is expressed by means of a performance measure based on a fuzzy set decomposition of the data set under consideration. Several experiments reported in here show that the proposed performance measure puts an order on different partitions of the same data which is consistent with the error rate of a classifier designed on the basis of the obtained cluster labelings.

This paper describes a procedure for segmenting muscle cell pictures. The segmentation procedure is broken into two logical parts. The first part segments the picture into regions composed of cells or clumps of cells using a number of low-level operations. The second part of the procedure involves the segmentation of the cell clumps into individual cells. This is done by using a hierarchical clustering algorithm to group together those boundary points of a cell clump that belong to the same globally convex sections of the boundary. The dissimilarity measure used by the clustering algorithm is based only on information about the shape of the boundary, where this information is derived from line segments interior to the boundary. This procedure has given us satisfactory results on a number of test pictures.