Yue Zhao

Ear and face based multimodal recognition based on KFDA

Ear recognition is proved to be a promising authentication technique. Because of earpsilas special physiological structure and location, the fusion of ear and face biometrics could fully utilize their connection relationship of physiological location and the supplement between these two biometrics, and possess the advantage of recognizing people without their cooperation. In this paper a novel feature fusion algorithm based on KFDA is proposed and applied to multimodal recognition based on fusion of ear and profile face. With the algorithm, the fusion discriminant vectors of ear and profile face are established and nonlinear feature fusion projection could be implemented. The experimental results show that the method is efficient for feature-level fusion and the ear and face based multimodal recognition performs better than ear or profile face unimodal biometric recognition.

A telecom clients’ credit risk rating model based on active learning

The available cases with actual classes are not enough for building telecom clientspsila credit classification model in practice, especially for the newly established system in which old customerspsila data do not exist. For evaluating telecom clientspsila credit, a classifier based on active learning is proposed in this paper. Active learning aims at reducing the number of training examples to be labeled by automatically processing the unlabeled examples, then selecting the most informative ones with respect to a given cost function for a human to label. Experimental results show the model built by the active learning algorithm with less labeled training data can reach the same accuracy as passive learning. This can reduce annotation cost for credit evaluation experts.

The study of multimodal recognition based on ear and face

Because of earpsilas special physiological structure and location, it is reasonable to combine ear with profile face for recognition in such scenarios as frontal face images are not available. First, only the face profile-view images are captured for recognition. Then according to the algorithm of canonical correlation analysis, a kind of associated feature based on ear and face was proposed for classification and personal identity recognition. The experimental results show that multimodal recognition fusing ear and profile face results in improvement over either ear or profile face unimodal biometric recognition.

A kind of algorithm of faint object recognition based on higher order cumulant spectrum

A new algorithm of faint object recognition was presented, through the principles of restraining to noise of HOC and SVD being analyzed, and then based on the idea of combining HOC with SVD being given, in this paper. The integrated and valid recognition frame, included Gaussian, or non-gaussian noise controlled, higher order cumlant spectrum analyzed, recognition template designed and trained, and the classifying function raised, is formed in the algorithm. Finally, the simulation was carried out founded on the experimental data., and the results are illustrated the algorithm is effective and valid.

MBBNTree Classifier Algorithm Based on Active Learning from Unlabeled Samples

MBBNTree algorithm, which integrates the advantage of Markov blanket Bayesian networks (MBBN) and decision tree, would behave better performance than other Bayesian networks for classification. But the available training samples with actual classes are not enough for building MBBNTree classifier in practice. Active learning aims at reducing the number of training examples to be labeled by automatically processing the unlabeled examples, then selecting the most informative ones with respect to a given cost function for a human to label. In this paper, the MBBNTree classifier algorithm based on active learning would be presented to solve the problem of learning MBBNTree classifier from unlabeled samples. Experimental results show that the proposed algorithm can reach the same accuracy as passive learning with few labeled training examples.

Research on the algorithm of adaptive variable structure control and synchronization for a class of continuous-time chaotic system

Research on the problem of the control and synchronization for a class of continuous-time chaos system with unknown parameter and time-delayed property (CTCS-UP-TP) was investigated. An effective synchronization controlling algorithm combined variable structure control with adaptive parameter estimation was given under the certain hypothetical conditions. At the same time, the realizing principle of the presented nonlinear variable structure controller was analyzed and the asymptotic stability of the control and synchronization chaos system was proved. Furthermore, the simulation on the presented algorithm would be carried out, and the results illustrated that the designed controller was effective and valid.

A kind of trading off algorithm of object classification based on intelligent data fusion

A kind of trading off algorithm of target classification combined with double mode intelligent fusion is presented, Applying neuron-fuzzy technique to the synthesis of the complementary information between radar and infrared, through combining neuron-fuzzy technique with D-S evidence fusion theory, the ability of target classification could be improved. At the same time, according to the effective detection range of sensors, reasonably select the target features that can ensure the accurate target classification so that the computation load of the algorithm can be largely decreased compared to the infrared single model fusion. Simulation results illustrate that the trading off algorithm is effective.

Research on the Algorithm of Adaptive Coupling Control and Synchronization for a Kind of Continuous-time Chaos System

The research on the problem of control and synchronization of the chaotic system would be carried out in this paper. The simple and reliable controller based on linear coupling rules and the theory of parameters estimation was designed, for the continuous-time chaos system with unknown parameter and delayed-time property(CTCS-UP-DTP). At the same time, the asymptotic stability of the control and synchronization continuous-time chaos system and the realizable character of unknown parameters estimation would be proved or analyzed in theory. And then the simulation was carried out and the results illustrated that the given algorithm is valid and effective.

A new MBBCTree classification algorithm based on active learning

MBBCTree algorithm, which integrates the advantage of Markov blanket Bayesian networks (MBBC) and Decision Tree, performances better than other Bayesian Networks for classification. But MBBCTree classifier was built by the traditional passive learning. The available training samples with actual classes are not enough for passive learning method for modelling MBBCTree classifier in practice. Active learning aims at reducing the number of training examples to be labeled by automatically processing the unlabeled examples, then selecting the most informative ones with respect to a given cost function for a human to label. In this paper, a new MBBCTree classifier algorithm based on active learning is present to solve the problem of building MBBCTree classifier from unlabelled samples. Experimental results show that the proposed algorithm can reach the same accuracy as passive learning with few labeled training examples.

An active MBBNTree classifier learning from unlabeled samples

Obtaining labeled training examples for some classification tasks is often expensive, such as text classification, mail filtering, while gathering large quantities of unlabeled examples is usually very cheap. Active learning aims at reducing the number of training examples to be labeled by automatically processing the unlabeled examples, then selecting the most informative ones with respect to a given cost function for a human to label. MBBNTree algorithm, which integrates the advantage of Markov Blanket Bayesian Networks (MBBN) and Decision Tree, would behave better performance than other Bayesian Networks for classification. But the available training samples with actual classes are not enough for building MBBNTree classifier in practice. In this paper, the MBBNTree classifier algorithm based on the Query-by-Committee of active learning would be presented to solve the problem of learning MBBNTree classifier from unlabeled samples. Experimental results show that the proposed algorithm can reach the same accuracy as passive learning with few labeled training examples.