Xie Weixin

Parameter optimization in FCM clustering algorithms

Weighting exponent m is an important parameter in fuzzy c-means (FCM) clustering algorithm, which directly affects the performance of the algorithm and the validity of fuzzy cluster analysis. However, so far the optimal choice of m is still an open problem. A method of selecting the optimal m is proposed in this paper, which is based on the fuzzy decision theory. The experimental results obtained demonstrate its effectiveness and arrive a conclusion that the optimal range of m is [1.5, 2.5] in practical applications.

Bispectrum-based radar target classification

With the good performance of higher-order spectra (HOS) techniques for non-Gaussian signal processing and Gaussian noise suppression capability, an AR model parametric bispectrum estimation is presented for conventional radar target return analysis. A reasonable selection of target bispectrum features is made with a formation of target feature vector for target classification. The classification experimental results-on-actual target returns are given, with satisfactory results.

HMM based online hand-drawn graphic symbol recognition

In this paper, an online hand-drawn graphic symbol recognition algorithm based on hidden Markov models is presented. A rearrangement strategy is applied to the hand-drawn symbol points in order to alleviate the influence of the difference in drawing sequence. Based on rearranged drawing points, global distance measure and local angle feature are extracted as the feature vector. After the quantization, a discrete HMM is used as the core recognizer. The experiment shows the recognition rate of our system can be above 85%.

Template based fuzzy c-shells clustering algorithm and its fast implementation

The fuzzy c-shells (FCS) clustering algorithms are widely applied in pattern recognition and computer vision. However, the available FCS algorithms are valid only for detecting the hyper-spherical-shell or hyper-ellipsoidal-shell type clusters, which limits their applications. In this paper, a template based fuzzy c-shells (TBFCS) clustering algorithm is proposed. Its fast implementation is also given. This approach can obtain better clustering performance for any-shell type clusters, so that it is used more widely than the existing FCS algorithms. The effectiveness of the proposed algorithm is illustrated with experimental results.

A novel FCM clustering algorithm for interval-valued data and fuzzy-valued data

It is well known that fuzzy c-means (FCM) algorithm is one of the most popular methods of cluster analysis. However, the traditional FCM algorithm does not work for the interval-valued data and fuzzy-valued data. To this end, a feature mapping method is proposed to preprocess these special type data, and then the traditional FCM algorithm can also be employed to analyze the interval-valued and fuzzy-valued data. Therefore, a novel FCM clustering algorithm is formed for interval-valued data and fuzzy-valued data. The experimental result demonstrates its effectiveness.

Fractional Fourier transform based target number detection

Due to the drawback that a conventional radar can not directly resolve the targets flying in a group in both range and azimuth, a method of target number detection is proposed by using the fractional Fourier transform based on the fact that the targets may be resolved in the Doppler frequency domain. The combination of coarse and fine search steps adopted effectively resolves the contradiction between the computation time and the search accuracy. In addition, being a 1-D linear transform and without the influence of the cross-term as in other time-frequency transforms, the method performs well under either higher SNR or lower SNR. The experimental results of actual target echoes demonstrate the performance of the method.

A new initialization method of cluster centers

The problem of parameters selection for potential function used to initialize cluster centers is discussed, and two formulas are given for determining these parameters. Then a new potential function to initialize cluster centers is also given which is computational effective. Finally, a set of compared experiments is presented to show the effectiveness of the proposed methods.

A robust optical flow computation

This paper presents a new method for robust and accurate optical flow estimation. The significance of this work is twofold. Firstly, the idea of bi-directional scheme is adopted to reduce the model error of optical flow equation, which allows the second order Taylor’s expansion of optical flow equation for accurate solution without much extra computational burden; Secondly, this paper establishs a new optical flow equation based on LSCM (Local Structure Constancy Model) instead of BCM (Brightness Constancy Model), namely the optical flow equation does not act on scalar but on tensor-valued (matrix-valued) field, due to the two reason: (1) structure tensor-value contains local spatial structure information, which provides us more useable cues for computation than scalar; (2) local image structure is less sensitive to illumination variation than intensity, which weakens the disturbance of non-uniform illumination in real sequences. Qualitative and quantitative results for synthetic and real-world scenes show that the new method can produce an accurate and robust results.

Reduction of training datasets via fuzzy entropy for support vector machines

Support vector machines (SVM) are currently the state-of-the-art models for many classification problems but they suffer from the complexity of their training algorithm that is at least quadratic with respect to the number of examples. Hence, it is hard to try to solve real-life problems with more than a few hundreds of thousands examples by SVM. The present paper proposes a new heuristic method based on the fuzzy entropy. Under the circumstances that there are little support vectors in the original training set, this new method can effectively preselect the boundary subset which contain overwhelming majority support vectors. By substituting the boundary subset for original training set, our method greatly reduces the training time, while the ability of support vector machine to classification is unaffected. Comparing to other analogous methods, the merit of our method is that there are no parameters for determining the border of subset. The preliminary experimental results indicate that our approach is efficient and practical.

An initialization method for multi-type prototype fuzzy clustering

Fuzzy clustering is an important branch of unsupervised classification, and has been widely used in pattern recognition and image processing. However, most existing fuzzy clustering algorithms are sensitive to initialization, and strongly depend on the number of clusters, which limits their applications. Moreover, it these algorithms also need to know the type and number of prototypes in advance in multi-type prototype fuzzy clustering. To overcome these limitations, a method for acquiring a priori knowledge about the clustering prototype is proposed in this paper, which obtains better performance in initializing multi-type prototype fuzzy clustering.