Miao Liu

Fuzzy kernel clustering based on particle swarm optimization

A novel fuzzy kernel clustering algorithm is presented based on Particle Swarm Optimization algorithm (PSO). The idea of the algorithm is firstly map the data in the original space to a high-dimensional feature space by using Mercer kernel functions where the data are expected to be more separable then perform Fuzzy C-means (FCM) in the high dimensional space. The iteration process is replaced by the PSO based on gradient descent of FCM in feature space, which makes the algorithm have a strong global searching capacity and avoids the local minimum problems of FCM. At the same time, FCM is no longer a large degree dependent on the initialization values. The results of simulation experiments show the feasibility and effectiveness of the kernel clustering algorithm based on PSO.

The methods of improving variable illumination for face recognition

The illumination fluctuation is one of the important factors that influence the precision of face recognition. When environmental illumination is changed, the implementation method which is often used in the laboratory may not be effective any longer. Several traditional methods for compensating and improving variable illumination are discussed in this paper. Then a novel algorithm based on wavelet analysis is presented to find the invariance of illumination. Several empirical tests are given to demonstrate the effectiveness of our method. The method can be applied to a real system. And it can also improve the system robustness and adaptability.

Face recognition using adaptive resonance theory

Human face detection and recognition are challenged questions in pattern recognition field. After the facial features such as eyes, nose and mouth are detected in an image which contains a face, the rectangle area surrounding facial features is obtained. The pixels number of the rectangle area is large and the intensity values of these pixels are often treated as a feature vector. It is very important to drop the dimension of the vector for an effective recognition. Three means for dimensional reduction in the feature extraction field are often used, including average values of weighted intensity, wavelet transform and principle component analysis. The compact face feature vector is the eigenvector to be recognized. A face recognition method using ART2 is proposed in the paper. Experiment results show that it is preferable in recognition as well as it could increase or decrease samples rapidly.

Colony Evolution in Social Networks Based on Multi-agent System

According to the sociological principium, this paper designed a model aimed at social networks and implemented it using the multi-agent system. Based on this model, we established a simulation system to research the evolution of colony in the social networks, analyzed the effects on the evolution by the characteristics of individuals and achieved meaningful conclusions.

Rules Extraction from ANN Based on Clustering

We propose a novel algorithm based on clustering to extract rules from artificial neural networks. After networks Beijing trained and pruned successfully, inner-rules are generated by discrete activation values of hidden units. Then, weights between input and hidden units are clustered to decrease the complexity of rules extraction. In clustering phase, the clustered number of weights can be adjusted dynamically according to activation values of their corresponding hidden units. The experimental results demonstrate that this algorithm is effective.

Differential evolution fuzzy clustering algorithm based on kernel methods

A new fuzzy clustering algorithm is proposed. By using kernel methods, this paper maps the data in the original space into a high-dimensional feature space in which a fuzzy dissimilarity matrix is constructed. It not only accurately reflects the difference of attributes among classes, but also maps the difference among samples in the high-dimensional feature space into the two-dimensional plane. Using the particularity of strong global search ability and quickly converging speed of Differential Evolution (DE) algorithms, it optimizes the coordinates of the samples distributed randomly on a plane. The clustering for random distributing shapes of samples is realized. It not only overcomes the dependence of clustering validity on the space distribution of samples, but also improves the flexibility of the clustering and the visualization of high-dimensional samples. Numerical experiments show the effectiveness of the proposed algorithm

Global Structure Constraint: A Fast Pre-location for Active Appearance Model

In this paper, a global structure constraint model, GSC, is described. This model represents the target pattern as a set of landmark points reserving both geometric relationships (shape model) and color information (color model). Each patch with small color variations in the target pattern is denoted by one or several points, differing with Active Shape Model, ASM, and Active Appearance Model, AAM, which model the local boundary information of target patterns. With the information of global distributions, GSC model can be exploited to estimate the initial status of objects in target images in a more generalized way, for its consequent models, such as ASM and AAM. Low dimension representation and the flexible selection of transformation parameters lead to the rapid match to the complex models. The experiment results show the models are effective.

Multi-Objective Evolutionary Algorithm Based on Max-Min Distance Density

This paper proposed a multi-objective differential evolution algorithm based on max-min distance density. The algorithm proposed the definiteness of max-min distance density and a Pareto candidate solution set maintenance method, and ensured the diversity of the Pareto solution set. Using Pareto dominance relationship among individuals and max-min distance density ensured the convergence of the algorithm, realized solving multi-objective optimization problems. The proposed algorithm is applied to five ZDT test functions and compared with others multi-objective evolutionary algorithms. Experimental result and analysis show that the algorithm is feasible and efficient

Researches on inhibition mechanism in P2P networks

A peer-to-peer (P2P) system will starve from resource if every user acts as a free rider who only takes resources from the system but never contributes. While developing a scheme, it is also important to inhibit free riders besides incentivizing users to contribute resources. In this paper, we adopt game theory to explore behaviors of nodes in P2P networks and present a Tit-for-Tat based mechanism. The reputation of one node is scored dispersedly and stored locally by its counterparts, according to its past reactions to their requests. Experimental results have prove that free riders are inhibited effectively since it takes more time for them to download resources than altruistic nodes.

Object Localization Based on Global Structure Constraint Model and Optimal Algorithm

We present a novel object localization approach based on the global structure constraint model (GSC) and optimal algorithm. In GSC, Objects are described as constellations of points satisfied with their specific global structure constraints. The spatial relations among all the patches having stable color information and their representative color information around patches are encoded. Then, the searching algorithm, i.e. genetic algorithm, is used to locate target objects in images by finding out the exactly matched position. In the experiment, we tested the approach on a collection of human face images and the results demonstrated the approach is simple, effective and efficient.