Search

Data clustering is an important technique in data mining and pattern recognition. In practical tasks the clusters can be of arbitrary shapes. However, many existing algorithms tend to generate only spherical clusters. While density based clustering algorithms are able to deal with arbitrary clusters, they usually involve multiple user-specified parameters. In this paper we propose to solve this problem by making use of the nice properties of dominant set algorithm. Specifically, we use the dominant sets algorithm with histogram equalization transformation to generate initial clusters. These initial clusters are usually subsets of real clusters. Then we expand the initial clusters to final ones with density information captured in the initial clusters. We experiment with clusters of various types and compare with other clustering algorithms to demonstrate the effectiveness of the proposed algorithm

Clustering is an important unsupervised learning approach and widely used in pattern recognition, data mining and image processing, etc. Different from existing clustering algorithms based on partitioning within data, dominant sets clustering extracts clusters in a sequential fashion. Based on graph-theoretic concept of a cluster, dominant sets clustering can be accomplished with a game dynamics efficiently while being able to determine the number of clusters automatically. However, we have observed that the definition of dominant set over weights the importance of high intra-cluster similarity. Consequently, dominant sets clustering is found to be sensitive to similarity parameters and show the tendency to generate over-segmented clustering results. In order to solve these problems, in this paper we present a cluster extension algorithm by making use of the relationship of intra-cluster and inter-cluster similarity. In experiments on eight datasets, our algorithm performs evidently better than the original dominant sets algorithm, and comparably to other state-of-the-art clustering algorithms.

Clustering image pixels is an important technique in image segmentation. While normalized cuts is popularly used in image segmentation, dominant set clustering is another promising method and shown to outperform normalized cuts in some experiments. However, dominant set clustering suffers from the problems of sensitiveness to distance measures and over-segmentation tendency. In this paper we present DSET++ to enhance the original dominant set clustering and solve the two problems. Firstly, we use the histogram equalization in image enhancement to transform the similarity matrix and eliminate the sensitiveness to distance measures. In the second step we extend the dominant set based on density information to overcome the tendency of over-segmentation. Preliminary experiments on data clustering tasks validate the effectiveness of DSET++ clustering.

In this paper a filter algorithm based on vertical-horizontal sub windows and median-mean is proposed. The algorithm detects the center point of the noise image in N × N window. For the noise point, The window will be divided into 2N sub-windows of vertical and horizontal, if the number of information points in the sub-window is odd, calculates their median value, else calculates their mean value, then the center point pixel value will be replaced by the median value of the those median value and mean value in the detection window and the center point pixel value. The center pixel value will be retained for the non noise points. Experimental results indicate the algorithm not only has a strong denoising ability but also keeps the details and texture of the image.

To reduce attributes in incomplete information system, this paper proposed a direct method based on rough set theory, which gives the concept of tolerance relationship similar matrix via using an extension of equivalence relationship of rough set theory. It solves the problem of inconsistency in the incomplete information system through the introduction of restrictions of the generalized decision function. It calculates attribute significance , the core attributes of incomplete information systems via the tolerance relationship similar matrix. Experiment results show that the algorithm is simple and effective.

To deal with attribute reduction in incomplete information systems, this paper proposed a direct method of attribute relative reduction based on rough set theory. This reduction algorithm gives the concept of tolerance relationship similar matrix via extending equivalence relationship of rough set theory, which is called tolerance relationship. It introduces the generalized decision function to solve the problem of inconsistency in the incomplete information system. This algorithm uses the tolerance relationship similar matrix to calculate the core attributes of incomplete information systems. It applies attribute significance, which this paper puts forward based on attribute frequency in the tolerance relationship similar matrix, as the heuristic knowledge. And it makes use of binsearch heuristic algorithm to calculate the candidate attribute expansion so that it can reduce the expansion times to speed up reduction. Experiment results show that the algorithm is simple and effective.

A new class of interacting multiple-model algorithms based on turning rate estimation for maneuvering target tracking is presented. To track turning maneuver, the key is how to estimate the parameter of motion, i.e., target's turning rate. It can be obtained in real time while the target's heading is filtered using multiple motion models, which include current statistical model and constant angular velocity model. Then the result estimated could be integrated with the variation of circular-turn model (VCTM) and the curvilinear model (CM), referred to as modified VCTM (MVCTM) and modified CM (MCM) respectively. The performances of the algorithms presented here are evaluated and compared via simulation of a generic maneuvering target tracking problem. Simulation results show that the MVCTM-based significantly outperform the MCM-based tracker.

In order to recover the missing data in an information system, the paper proposed a new approach based on rough set to reduce the redundant attributes, discretize the continuous attributes and fill in the missing data. According to indiscernible relationship, discernible vector were defined and used the discernible vector addition rule to reduce attributes. And then, depending on the concept of super-club data and entropy of the information table, discretization of the continuous attributes was implemented. Finally, by use of the corresponding relationship of condition attributes and decision attributes, the definition of interval value and interval value addition rule were defined and filled up the incomplete data. The illustration and experimental results indicate that the approach is effective and efficient.

Discretizing continuous attributes in an information table is a major problem in rough set theory and a NP-hard problem also. To solve it, the paper proposed a new method. The paper firstly introduced the concept of breaking points while explaining its role in this algorithm, and described the discretization essential matter. And then defined a formula for scaling the attributes importance and the breaking points importance in the rough sets boundary region respectively, and proved four theorems about them. By use of the above concepts, formulas, theorems and combining the local discretization methods and global ones, the paper put forward an algorithm for discretization problem. This algorithm discretized it on single attribute after discretizing the information table on all attributes, and meanwhile it continually combined the unimportant breaking points and remained the important ones with the iterative constraints of entropy. This algorithm was implemented in VC++ program. Illustration and experimental results indicate the algorithm is concise, effective and efficient.