-
[show abstract]
[hide abstract]
ABSTRACT: This paper introduces a framework that employs the Fisher linear discriminant model (FLDM) and classifier (FLDC) on integrated facial appearance and facial expression features. The principal component analysis (PCA) is firstly applied for dimensionality reduction. The normalized fusion method is then applied to the reduced lower dimensional subspaces of these two features. Finally, the FLDM is used for generalizing the most expressive and discriminant feature space for enhancing better generalization performance. Experimental results show that 1) the integrated features of the facial appearance and facial expressions carry the most expressive and discriminant information and 2) the intra-personal variation, indeed, can assist the extra-personal separation. In particular, the proposed method achieves 100% for our database recognition accuracy using only 9 features.
Pattern Recognition, 2008. ICPR 2008. 19th International Conference on; 01/2009
-
[show abstract]
[hide abstract]
ABSTRACT: Physiological and/or behavioural characteristics of humans such as face, gait and/or voice have been used in biometric recognition technology. Apart from these characteristics (which have been reported in the literature), the hypothesis of this research was to investigate if facial behaviour could be used for human identification. We analysed and proposed a multiple experts system, called Adaptive Multiple Experts System (AMES), for validating our hypothesis and analysis. We used the Japanese Female Facial Expression (JAFFE) database as it provides the facial behaviour traits for data collection. The experimental results indicate that facial behaviours may provide information about individual difference and, thus may be used as another behavioural biometric.
Multimedia Signal Processing, 2008 IEEE 10th Workshop on; 11/2008
-
[show abstract]
[hide abstract]
ABSTRACT: Physiological and/or behavioral characteristics of humans such as face, gait and/or voice have been used in biometric recognition technology. Apart from those characteristics reported in the literature, the hypothesis of this research was to initially investigate if human facial behaviors could also be used as another behavioral traits for human identification. We used kernel subspace analysis method to analyze the data so as to support our hypothesis. We used the Japanese Female Facial Expression (JAFFE) database as it provides the facial behavior traits for data collection. The experimental results indicate that facial behaviors may provide information about individual differences, thus may be used as another behavioral biometric.
Systems, Man and Cybernetics, 2007. ISIC. IEEE International Conference on; 11/2007
-
[show abstract]
[hide abstract]
ABSTRACT: In face recognition, if the extracted input data contains misleading information (uncertainty), the classifiers may produce degraded classification performance. In this paper, we employed kernel-based discriminant analysis method for the non-separable problems in face recognition under facial expression changes. The effect of the transformations on a subsequent classification was tested in combination with learning algorithms. We found that the transformation of kernel-based discriminant analysis has a beneficial effect on the classification performance. The experimental results indicated that the nonlinear discriminant analysis method dealt with the uncertainty problem very well. Facial expressions can be used as another behavior biometric for human identification. It appears that face recognition may be robust to facial expression changes, and thus applicable.
Neural Networks, 2007. IJCNN 2007. International Joint Conference on; 09/2007
-
[show abstract]
[hide abstract]
ABSTRACT: First, a hierarchical modelling method, VQSVM, is introduced, and some remarks are discussed. Secondly the proposed VQSVM is applied to a nonstandard learning environment, imbalanced data sets. In cases of extremely imbalanced dataset with high dimensions, standard machine learning techniques tend to be overwhelmed by the large classes. The hierarchical VQSVM contains a set of local models i.e. codevectors produced by the vector quantization and a global model, i.e. support vector machine, to rebalance datasets without significant information loss. Some issues, e.g. distortion and support vectors, have been discussed to address the trade-off between the information loss and undersampling rate. Experiments compare VQSVM with random resampling techniques on some imbalanced datasets with varied imbalance ratios, and results show that the performance of VQSVM is superior or equivalent to random resampling techniques, especially in case of extremely imbalanced large datasets.
Neural Networks, 2007. IJCNN 2007. International Joint Conference on; 09/2007
-
[show abstract]
[hide abstract]
ABSTRACT: The prediction of stock market has been an important issue in the field of finance, mathematics and engineering due to its great potential financial gain. In addition, uncertainty in the prediction of the financial time series has attracted interest from many researchers. In this study, we present recent developments in stock market prediction models, and discuss their strengths and limitations. In addition, we investigate diverse macroeconomic factors and their issues in the prediction of stock market. From this study, we found that incorporating event information into the prediction models plays important roles for more accurate prediction. Hence, an accurate event weighting method and a stable automated event extraction system are required for more accurate and reliable stock market prediction
9th International Multitopic Conference, IEEE INMIC 2005; 01/2006
-
[show abstract]
[hide abstract]
ABSTRACT: This paper surveys machine learning techniques for stock market prediction. The prediction of stock markets is regarded as a challenging task of financial time series prediction. In this paper, we present recent developments in stock market prediction models, and discuss their advantages and disadvantages. In addition, we investigate various global events and their issues on predicting stock markets. From this survey, we found that incorporating event information with prediction model plays very important roles for more accurate prediction. Hence, an accurate event weighting method and a stable automated event extraction system are required to provide better performance in financial time series prediction
Computational Intelligence for Modelling, Control and Automation, 2005 and International Conference on Intelligent Agents, Web Technologies and Internet Commerce, International Conference on; 12/2005
-
[show abstract]
[hide abstract]
ABSTRACT: Face recognition has been recognized as most simple and non-intrusive technology that can be applied in many places. However, there are still many unsolved face recognition problems such as facial deformations, pose or illumination variations. Nonetheless, little research has been done on facial deformation problems. The hypothesis of this research was to determine if a face recognition system could provide robustness to facial deformation problem and its potential applicability. We used the Japanese female facial expression (JAFFE) database as it provides the deformed facial traits for data collection. We used subspace model analysis to analyze the data so as to support our hypothesis. The experimental results indicate that face recognition may be robust to facial changes and applicable.
Systems, Man and Cybernetics, 2005 IEEE International Conference on; 11/2005
-
Computational Intelligence for Measurement Systems and Applications, 2005. CIMSA. 2005 IEEE International Conference on; 08/2005
-
[show abstract]
[hide abstract]
ABSTRACT: This paper is devoted to theoretic algorithms development and experimental research of automatic target detection of acoustic signals, especially for boats generated signals. In this paper, an observation space is created by sampling and dividing input analog acoustic signal into multiple frames and each frame is transformed into frequency domain. In the created observation space, a median constant false alarm rate (MCFAR) and post detection integration algorithms have been proposed for an effective automatic target detection of boat generated acoustic signals, in which a low constant false alarm rate is kept with relative high detection rate. The proposed algorithms have been tested on real boat generated acoustic signals. The statistical analysis and experimental results showed that the proposed algorithm has kept a very low false alarm rate and relatively high detection rate.
Signal Processing and Information Technology, 2004. Proceedings of the Fourth IEEE International Symposium on; 01/2005
-
[show abstract]
[hide abstract]
ABSTRACT: Background modelling is widely used in computer vision for the detection of foreground objects in a frame sequence. The more accurate the background model, the more correct is the detection of the foreground objects. In this paper, we present an approach to background modelling based on a mean-shift procedure. The mean shift vector convergence properties enable the system to achieve reliable background modelling. In addition, histogram-based computation and the new concept of local basins of attraction allow us to meet the stringent real-time requirements of video processing.
Image Processing, 2004. ICIP '04. 2004 International Conference on; 11/2004
-
[show abstract]
[hide abstract]
ABSTRACT: Recent medical studies show that there exist aesthetic ideal features for facial beauty based on facial proportions. Automated tools that can provide information about the prediction of how the surgery will improve the patients' perceived beauty or 'peer-esteem' will find applications in various areas. In our previous work, we introduced an automated procedure based on image analysis and supervised learning that confirmed the existence of general rules in peer-esteem measurement. In this paper, we further experimented our automated system by extending the analysis of classification tools and human data by comparing a number of classifiers, namely decision trees, multi-layer perceptron and kernel density estimators. Results are good since the standardized distance is generally less than one class, and prove that these classifiers can be used to reliably predict the consensus of a large and varied population of human referees, hence providing peer-esteem information for patients.
Systems, Man and Cybernetics, 2004 IEEE International Conference on; 11/2004
-
[show abstract]
[hide abstract]
ABSTRACT: In multimedia applications such as MPEG-4, an efficient model is required to encode and classify video objects such as human, car and building. Recently, support vector machine (SVM) has been shown to be a good classifier; however, its large computational requirement prohibited its use in real time video processing applications. In this paper, a model is proposed that enables use of SVM in video applications. This paper aims to merge multi-scale based selective encoding/classification technique and locality-enhanced support vector machine (SVM). The proposed model allows selected image scales (of interest) to be encoded and classified more accurately by complex classifier such as SVM, whilst other image scales of less significance to be encoded and classified by simpler encoder/classifier. Image scales of interest are readily selected from multi-scale image processing paradigm. SVM is used to encode visual object information of significant image scale only; hence its use is efficient. Experiment with MPEG-4 video object encoding and classification shows that the performance of the proposed model is comparable with other models, however with significantly reduced computational requirements.
Systems, Man and Cybernetics, 2004 IEEE International Conference on; 11/2004
-
T. Jan
[show abstract]
[hide abstract]
ABSTRACT: In automated visual surveillance systems (AVSS), reliable detection of suspicious human behavior is of great practical importance. Many conventional classifiers have shown to perform inadequately because of unpredictable nature of human behavior. Flexible models such as artificial neural network (ANN) models can perform better; however, computational requirement of ANN models can be prohibitively large for realtime video processing. It is interesting to construct a small-sized ANN classifier that can perform well for threat assessment in video-based surveillance system. In this paper, modified probabilistic neural network (MPNN) is introduced that can achieve reliable classification, with significantly reduced computation. Experiment on visual surveillance application shows that MPNN achieves good classification but with much reduced computation compared to other ANN models. In this application, trajectory profile and motion history image information from the observed human subject are used for threat assessment.
Neural Networks, 2004. Proceedings. 2004 IEEE International Joint Conference on; 08/2004
-
[show abstract]
[hide abstract]
ABSTRACT: In this paper, a model is proposed which combines multiple local linear models with a novel modified probabilistic neural network (MPNN). The proposed model is shown to provide improved regularization with reduced computation utilizing semiparametric model approach and efficient vector quantization of data space. In this paper, the proposed model is shown to generalize better with reduced variance and model complexity in short-term financial prediction application.
Computational Intelligence for Measurement Systems and Applications, 2004. CIMSA. 2004 IEEE International Conference on; 08/2004
-
T. Jan
[show abstract]
[hide abstract]
ABSTRACT: In this paper, a hybrid classifier is introduced which combines a linear discriminant classifier and a nonlinear non-parametric neural network based classifier such as the radial basis function neural networks. This hybrid model provides a linear parametric coding of the coarse-level information about the underlying image, and then uses the neural networks to encode the finer-level information of the same image. This model allows the selected image regions of interest be analyzed and encoded in the finer scales by a non-parametric neural network models whilst the image regions of no-interest are analyzed and encoded in coarse scales by a simple parametric model. The experiment on video image compression shows that the proposed model achieves significantly reduced computations for similar compression performance compared to other conventional methods.
Signal Processing and Information Technology, 2003. ISSPIT 2003. Proceedings of the 3rd IEEE International Symposium on; 01/2004
-
T. Jan
[show abstract]
[hide abstract]
ABSTRACT: In this paper, an ensemble of models is introduced which combines a linear parametric model and a nonlinear non-parametric model such as artificial neural network (ANN). This model aims to embody the desirable characteristics of linear parametric model such as stable generalization capability while retaining the data-based learning and prediction capacity of ANNs. The proposed model is applied for short term time series prediction and the results show that the proposed model achieves good generalization (prediction) performance utilizing the nonparametric ANN model component while achieving much improved stability utilizing the linear model component. The experiment compares the proposed model to other ANN models and linear models for generalization.
Signal Processing and Information Technology, 2003. ISSPIT 2003. Proceedings of the 3rd IEEE International Symposium on; 01/2004
-
[show abstract]
[hide abstract]
ABSTRACT: In automated visual surveillance applications, detection of suspicious human behaviors is of great practical importance. However due to random nature of human movements, reliable classification of suspicious human movements can be very difficult. Artificial neural network (ANN) classifiers can perform well however their computational requirements can be very large for real time implementation. In this paper, a data-based modeling neural network such as modified probabilistic neural network (MPNN) is introduced which partitions the decision space nonlinearly in order to achieve reliable classification, however still with acceptable computations. The experiment shows that the compact MPNN attains good classification performance compared to that of other larger conventional neural network based classifiers such as multilayer perceptron (MLP) and self organising map (SOM).
Neural Networks for Signal Processing, 2003. NNSP'03. 2003 IEEE 13th Workshop on; 10/2003
-
T. Jan
[show abstract]
[hide abstract]
ABSTRACT: In many business applications, accurate short term prediction is vital for survival. Many different techniques have been applied to model business data in order to produce accurate prediction. Artificial neural network (ANN) have shown excellent potential however it requires better extrapolation capacity in order to provide reliable prediction. In this paper, a combination of piecewise linear regression model in parallel with general regression neural network is introduced for short term financial prediction. The experiment shows that the proposed hybrid model achieves superior prediction performance compared to the conventional prediction techniques such as the multilayer perceptron (MLP) or Volterra series based prediction.
Neural Networks, 2003. Proceedings of the International Joint Conference on; 08/2003
-
[show abstract]
[hide abstract]
ABSTRACT: For separation of signals with overlapping spectra. Classical linear filters fail to perform effectively. Nonlinear filters such as Volterra filters or artificial neural networks (ANNs) can perform better but their implementations are often impractical due to their computational complexity. In this paper an ANN based hyperspace signal modeling is used to separate signals with overlapping spectra. The computational complexity of the ANN is reduced significantly by a simple feature extraction utilizing the unique temporal characteristics of the signals. The results show that difficult signal separation and filtering can be achieved efficiently by employing an ANN and an effective feature extraction
Adaptive Systems for Signal Processing, Communications, and Control Symposium 2000. AS-SPCC. The IEEE 2000; 02/2000
