Browsing by Author "Liang, SF"

Now showing 1 - 20 of 20

A 3-D surface reconstruction approach based on postnonlinear ICA model
(2005-11-01)
Lin, CT
;
Cheng, WC
;
Liang, SF
;
生物科技學系
;
資訊工程學系
Photometric stereo technique deals with the reconstruction of three-dimensional (3-D) shape of an object by using several images of the same surface taken from the same viewpoint but under illuminations from different directions. In this paper, we propose a new photometric stereo scheme based on a new reflectance model and the postnonlinear (PNL) independent components analysis (ICA) method. The proposed nonlinear reflectance model consists of diffuse components and specular components for modeling the surface reflectance of a stereo object in an image. Unlike the previous approaches, these two components are not separated and processed individually in the proposed model. An unsupervised learning adaptation algorithm is developed to estimate the reflectance model based on image intensities. In this algorithm, the PNL ICA method is used to obtain the surface normal on each point of an image. Then, the 3-D surface model is reconstructed based on the estimated surface normal on each point of image by using the enforcing integrability method. Two experiments are performed to assess the performance of the proposed approach. We test our algorithm on synthetically generated images for the reconstruction of surface of objects and on a number of real images captured from the Yale Face Database B. These testing images contain variability due to illumination and varying albedo in each point of surface of human faces. All the experimental results are compared to those of the existing photometric stereo approaches tested on the same images. The results clearly indicate the superiority of the proposed nonlinear reflectance model over the conventional Lambertian model and the other linear hybrid reflectance model.
A class of physical modeling recurrent networks for analysis/synthesis of plucked string instruments
(2002-09-01)
Su, AWY
;
Liang, SF
;
電控工程研究所
;
Institute of Electrical and Control Engineering
A new approach is proposed that closely synthesizes tones of plucked string instruments by using a class of physical modeling recurrent networks. The strategies employed in this paper consist of a fast training algorithm and a multistage training procedure that are able to obtain the synthesis parameters for a specific instrument automatically. The training vector can be recorded tones of most target plucked instruments with ordinary microphones. The proposed approach delivers encouraging results when it is applied to different types of plucked string instruments such as steel-string guitar, nylon-string guitar, harp, Chin, Yueh-chin, and Pipa. The synthesized tones sound very close to the originals produced by their acoustic counterparts. In addition, this paper presents an embedded technique that can produce special effects such as vibrato and portamento that are vital to the playing of plucked-string instruments. The computation required in the resynthesis processing is also reasonable.
A novel neural-network-based image resolution enhancement
(2003)
Pu, HC
;
Lin, CT
;
Liang, SF
;
Kumar, N
;
電控工程研究所
In this paper, a novel HVS-directed neural-network-based adaptive interpolation scheme for natural image is proposed. A fuzzy decision system built from the characteristics of the human visual system (HVS) is proposed to classify pixels of the input image into human perception nonsensitive class and sensitive class. High-resolution digital images along with supervised learning algorithms are used to automatically train the proposed neural network. Simulation results demonstrate that the proposed new resolution enhancement algorithm can produce higher visual quality of the interpolated image than the conventional interpolation methods.
A novel prosodic-information synthesizer based on recurrent fuzzy neural network for the Chinese TTS system
(2004-02-01)
Lin, CT
;
Wu, RC
;
Chang, JY
;
Liang, SF
;
電控工程研究所
In this paper, a new technique for the Chinese text-to-speech (TTS) system is proposed. Our major effort focuses on the prosodic information generation. New methodologies for constructing fuzzy rules in a prosodic model simulating human's pronouncing rules are developed. The proposed Recurrent Fuzzy Neural Network (RFNN) is a multilayer recurrent neural network (RNN) which integrates a Self-constructing Neural Fuzzy Inference Network (SONFIN) into a recurrent connectionist structure. The RFNN can be functionally divided into two,parts. The first part adopts the SONFIN as a prosodic model to explore the relationship between high-level linguistic features and prosodic information based on fuzzy inference rules. As compared to conventional neural networks, the SONFIN can always construct itself with an economic network size in high learning speed. The second part employs a five-layer network to generate all prosodic parameters by directly using the prosodic fuzzy rules inferred from the first part as well as other important features of syllables. The TTS system combined with the proposed method can behave not only sandhi rules but also the other prosodic phenomena existing in the traditional TTS systems. Moreover, the proposed scheme can even find out some new rules about prosodic phrase structure. The performance of the proposed RFNN-based prosodic model is verified by imbedding it into a Chinese TTS system with a Chinese monosyllable database based on the time-domain pitch synchronous overlap add (TD-PSOLA) method. Our experimental results show that the proposed RFNN can generate proper prosodic parameters including pitch means, pitch shapes, maximum energy levels, syllable duration, and pause duration. Some synthetic sounds are on-line available for demonstration.
A robust digital image stabilization technique based on inverse triangle method and background detection
(2005-05-01)
Hsu, SC
;
Liang, SF
;
Lin, CT
;
電控工程研究所
;
Institute of Electrical and Control Engineering
In this paper, a novel robust digital image stabilization (DIS) technique is proposed to stably remove the unwanted shaking phenomena in the image sequences captured by hand-held camcorders without affecting moving objects in the image sequence and intentional motion of panning condition, etc. It consists of a motion estimation unit and a motion compensation unit. To increase the robustness in adverse image conditions, an inverse triangle method is proposed to extract reliable motion vectors in plain images which are lack of features or contain large low-contrast area, etc., and a background evaluation model is developed to deal with irregular images which contain large moving objects, etc. In the motion compensation unit, a CMV estimation method with an inner feedback-loop integrator is proposed to stably remove the unwanted shaking phenomena without losing the effective area of the image in panning condition. We also propose a smoothness index (SI) to quantitatively evaluate the performances of different image stabilization methods. The experimental results are on-line available to demonstrate the remarkable performance of the proposed DIS technique.(1.2).
Adaptive feature extractions in an EEG-based alertness estimation system
(2005)
Lin, CT
;
Chao, WH
;
Chen, YC
;
Liang, SF
;
腦科學研究中心
During the past years, the public security has become an important issue, especially, the safe manipulation and control of various vehicles. Maintaining high cognition is particularly important for the drivers behind the steering wheel. It requires an optimal estimation system to online continuously detect drivers' cognitive state related to abilities in perception, recognition, and vehicle control. In this paper, we proposed an EEG-based alertness estimation system with automatic feature selection mechanism. The independent component analysis (ICA) is used first to decompose the measured electroencephalogram (EEG). Then, a time-frequency analysis is performed to evaluate the time-frequency characteristic of each ICA component. We also proposed a new adaptive feature extracting mechanism for selections of frequency bands and ICA components. Different ranges of the alpha rhythm of subjects can be evaluated by the adaptive feature extracting mechanism according to the correlation coefficient between the ICA time-frequency response and the driving performance. The extracted features are then trained both by linear regression model and Self-cOnstructing Neuro-Fuzzy Inference Network (SONFIN) for the estimation of driving performance. The training and testing results of SONFIN are 96% and 91%, while the results of linear regression model are 90% and 85%, respectively. It demonstrates that the proposed adaptive feature extracting mechanism can achieve a great performance in alertness estimation with frequency band and component selection.
An on-line ICA-mixture-model-based self-constructing fuzzy neural network
(2005-01-01)
Lin, CT
;
Cheng, WC
;
Liang, SF
;
電控工程研究所
;
Institute of Electrical and Control Engineering
This paper proposes a new fuzzy neural network (FNN) capable of parameter self-adapting and structure self-constructing to acquire a small number of fuzzy rules for interpreting the embedded knowledge of a system from the given training data set. The proposed FNN is inherently a modified Takagi-Sugeno-Kang (TSK)-type fuzzy-rule-based model with neural network's learning ability. There are no rules initiated at the beginning and they are created and adapted through an on-line learning processing that performs simultaneous structure and parameter identification. In the structure identification of the precondition part, the input space is partitioned in a flexible way according to newly proposed on-line independent component analysis (ICA) mixture model. The input space is thus represented by linear combinations of independent, non-Gaussian densities. The first input training pattern is assigned to the first rule initially by the on-line ICA mixture model. Afterwards, some additional significant terms (input variables) selected by the on-line ICA mixture model will be added to the consequent part (forming a liner equation of input variables) incrementally or create a new rule in the learning processing. The combined precondition and consequent structure identification scheme can make the network grow dynamically and efficiently. In the parameter identification, the consequent parameters are tuned by the backpropagation rule and the precondition parameters are turned by the on-line ICA mixture model. Both the structure and parameter identifications are done simultaneously to form a fast learning scheme. The derived on-line ICA mixture model also provide a natural linear transformation for each input variable to enhance the knowledge representation ability of the proposed FNN and reduce the required rules and achieve higher accuracy efficiently. in order to demonstrate the performance of the proposed FNN, several experiments covering the areas of system identification, classification, and image segmentation are carried out. Our experiments show that the proposed FNN can achieve significant improvements in the convergence speed and prediction accuracy with simpler network structure.
Assessment of driver's driving performance and alertness using EEG-based fuzzy neural networks
(2005)
Lin, CT
;
Chen, YC
;
Wu, RC
;
Liang, SF
;
Huang, TY
Traffic fatalities in recent years have become a serious concern to our society. Accidents caused by drivers' drowsiness have a high fatality rate due to the decline of drivers' abilities in perception, recognition, and vehicle control abilities while sleepy. Preventing such an accident requires a technique for detecting, estimating, and predicting the level of alertness of a driver and a mechanism to maintain the driver's maximum performance of driving. This paper proposed a system that combines electroencephalogram (EEG) power spectra estimation, independent component analysis and fuzzy neural network models to estimate drivers' cognitive state in a dynamic virtual-reality-based driving environment. Experimental results show that the quantitative driving performance can be accurately and successfully estimated through analyzing driver's EEG signals by the proposed system.
Classification of driver's cognitive responses from EEG analysis
(2005)
Liang, SF
;
Lin, CT
;
Wu, RC
;
Huang, TY
;
Chao, WH
During the past years, the growing number of traffic fatalities has become an important issue in public security. In this paper, we develop a quantitative analysis for ongoing assessment of cognitive response by investigating the neurobiological brain dynamics in traffic-light experiments. A single-trial event-related-potential (ERP)-based fuzzy neural network (FNN) is applied to recognize different brain potentials stimulated by red/green/yellow traffic-light events. The system consists of a dynamic virtual-reality (VR)-based motion simulation platform, EEG signal detection and analysis units, and FNN-based classifier. The independent component analysis (ICA) algorithms are used to obtain noise-free ERP signals from the multi-channel EEG signals. A novel temporal filter is also proposed to solve time-alignment problems of ERP features and principle component analysis (PCA) is used to reduce dimension of features, which were then fed into a FNN classifier. Experimental results demonstrate the feasibility of detecting and analyzing multiple streams of ERP signals that organize operators' cognitive responses to task events. Comparisons of three kinds of linear and nonlinear classifiers show that our proposed FNN-based classifier can achieve a satisfactory and superior recognition rate (85%). The classification results can be further transformed as the control/biofeedback signals of intelligent driving systems.
EEG-based drowsiness estimation for safety driving using independent component analysis
(2005-12-01)
Lin, CT
;
Wu, RC
;
Liang, SF
;
Chao, WH
;
Chen, YJ
Preventing accidents caused by drowsiness has become a major focus of active safety driving in recent years. It requires an optimal technique to continuously detect drivers' cognitive state related to abilities in perception, recognition, and vehicle control in (near-) real-time. The major challenges in developing such a system include: 1) the lack of significant index for detecting drowsiness and 2) complicated and pervasive noise interferences in a realistic and dynamic driving environment. In this paper, we develop a drowsiness-estimation system based on electroencephalogram (EEG) by combining independent component analysis (ICA), power-spectrum analysis, correlation evaluations, and linear regression model to estimate a driver's cognitive state when he/she drives a car in a virtual reality (VR)-based dynamic simulator. The driving error is defined as deviations between the center of the vehicle and the center of the cruising lane in the lane-keeping driving task. Experimental results demonstrate the feasibility of quantitatively estimating drowsiness level using ICA-based multistream EEG spectra. The proposed ICA-based method applied to power spectrum of ICA components can successfully (1) remove most of EEG artifacts, (2) suggest an optimal montage to place EEG electrodes, and estimate the driver's drowsiness fluctuation indexed by the driving performance measure. Finally, we present a benchmark study in which the accuracy of ICA-component-based alertness estimates compares favorably to scalp-EEG based.
EEG-based fuzzy neural network estimator for driving performance
(2004)
Wu, RC
;
Lin, CT
;
Liang, SF
;
Huang, TY
;
Jung, TP
Accidents caused by drivers' drowsiness have a high fatality rate because of the marked decline in the drivers' vehicle control abilities. Preventing accidents caused by drowsiness is highly desirable but requires techniques for continuously detecting, estimating, and predicting the level. of alertness of drivers. This paper proposes a brain-machine interface that combines electroencephalographic power spectrum estimation, principal. component analysis, and fuzzy neural networks to estimate/predict drivers' drowsiness level in a virtual-reality-based driving simulator. The driving performance is defined as deviation between the center of the vehicle and the center of the cruising lane. Our results demonstrated that the proposed method is feasible to accurately estimate quantitatively driving performance in a realistic driving simulator.
Estimating driving performance based on EEG spectrum analysis
(2005)
Lin, CT
;
Wu, RC
;
Jung, TP
;
Liang, SF
;
Huang, TY
The growing number of traffic accidents in recent years has become a serious concern to society. Accidents caused by driver's drowsiness behind the steering wheel have a high fatality rate because of the marked decline in the driver's abilities of perception, recognition, and vehicle control abilities while sleepy. Preventing such accidents caused by drowsiness is highly desirable but requires techniques for continuously detecting, estimating, and predicting the level of alertness of drivers and delivering effective feedbacks to maintain their maximum performance. This paper proposes an EEG-based drowsiness estimation system that combines electroencephalogram (EEG) log subband power spectrum, correlation analysis, principal component analysis, and linear regression models to indirectly estimate driver's drowsiness level in a virtual-reality-based driving simulator. Our results demonstrated that it is feasible to accurately estimate quantitatively driving performance, expressed as deviation between the center of the vehicle and the center of the cruising lane, in a realistic driving simulator.
Estimating driving performance based on EEG spectrum and fuzzy neural network
(2004)
Wu, RC
;
Lin, CT
;
Liang, SF
;
Huang, TY
;
Chen, YC
The growing number of traffic fatalities in recent years has become a serious concern to society. Accidents caused by drivers' drowsiness behind the steering wheel have a high fatality rate because of the marked decline in the drivers' abilities of perception, recognition! and vehicle control abilities while sleepy. Preventing accidents caused by drowsiness requires a technique for detecting, estimating, and predicting the level of alertness of a driver and a mechanism for maintaining his/her maximum performance. This paper describes a system that combines electroencephalographic (EEG) power spectrum estimation, principal component analysis, and fuzzy neural network model to estimate/predict drivers' drowsiness level in a driving simulator. Our results demonstrated that, for the first time, it is feasible to accurately estimate task performance, accurately estimate quantitatively measured driving performance, expressed as deviation between the center of the vehicle and the center of the cruising lane, in a realistic driving simulation.
Fuzzy neural network design using support vector regression for function approximation with outliers
(2005)
Lin, CT
;
Liang, SF
;
Yeh, CM
;
Fan, KW
;
電控工程研究所
A fuzzy neural network based on support vector learning mechanism for function approximation is proposed in this paper. Support vector regression (SVR) is a novel method for tackling the problems of function approximation and regression estimation based on the statistical learning theory. SVR has been shown to have robust properties against noise. A novel support-vector-regression based fuzzy neural network (SVRFNN) by integrating SVR technology into FNN is developed. The SVRFNN combines the high accuracy and robustness of support vector regression (SVR) and the efficient human-like reasoning of FNN for function approximation. Experimental results show that the proposed SVFNN for function approximation can achieve good approximation performance with drastically reduced number of fuzzy kernel functions.
Machine learning with automatic feature selection for multi-class protein fold classification
(2005-07-01)
Huang, CD
;
Liang, SF
;
Lin, CT
;
Wu, RC
;
電控工程研究所
In machine learning, both the properly used networks and the selected features are important factors which should be considered carefully. These two factors will influence the result, whether for better or worse. In bioinformatics, the amount of features may be very large to make machine learning possible. In this study we introduce the idea of feature selection in the problem of bioinformatics. We use neural networks to complete our task where each input node is associated with a gate. At the beginning of the training, all gates are almost closed, and, at this time, no features are allowed to enter the network. During the training phase, gates are either opened or closed, depending oil the requirements. After the selection training phase has completed, gates corresponding to the helpful features are completely opened while gates Corresponding to the useless features are closed more tightly. Some gates may be partially open, depending oil the importance of the corresponding features. So, the network can not only select features in an online manner during learning, but it also does some feature extraction. We combine feature selection with our novel hierarchical machine learning architecture and apply it to multi-class protein fold classification. At the first level the network classifies the data into four major folds: all alpha, all beta, alpha + beta and alpha/beta. In the next level, we have another set of networks which further classifies the data into twenty-seven folds. This approach helps achieve the following. The gating network is found to reduce the number of features drastically. It is interesting to observe that, for the first level using just 50 features selected by the gating network, we can get a test accuracy comparable to that using 125 features in neural classifiers. The process also helps us get a better insight into the folding process. For example, tracking the evolution of different gates, we call find which characteristics (features) of the data are more important for the folding process. Eventually, it reduces the computation time. The use of the hierarchical architecture helps LIS get a better performance also.
Model-based synthesis of plucked string instruments by using a class of scattering recurrent networks
(2000-01-01)
Liang, SF
;
Su, AWY
;
Lin, CT
;
電控工程研究所
;
Institute of Electrical and Control Engineering
A physical modeling method for electronic music synthesis of plucked-string tones by using recurrent networks is proposed. A scattering recurrent network (SRN) which is used to analyze string dynamics is built based on the physics of acoustic strings, The measured vibration of a plucked string is employed as the training data for the supervised learning of the SRN, After the network is well trained, it can be regarded as the virtual model for the measured string and used to generate tones which can be very close to those generated by its acoustic counterpart. The "virtual string" corresponding to the SRN can respond to different "plucks" just like a real string, which is impossible using traditional synthesis techniques such as frequency modulation and wavetable. The simulation of modeling a cello "A"-string demonstrates some encouraging results of the new music synthesis technique. Some aspects of modeling and synthesis procedures are also discussed.
Modeling and analysis of acoustic musical strings using Kelly-Lochbaum lattice networks
(2004-11-01)
Liang, SF
;
Su, AWY
;
電控工程研究所
;
Institute of Electrical and Control Engineering
Musical strings play important roles in the timbre of a stringed instrument. An instrument installed with different strings can produce very different sounds. To model a plucked-string instrument physically, analyzing its strings may be a good starting point. In this paper, a novel strategy for the modeling and analysis of acoustic musical strings is proposed. In order to obtain the vibration of an acoustic string, a measurement device with multiple sets of electromagnetic pickups is built. A recurrent network based on the Kelly-Lochbaum structure with multiple sets of system parameters is proposed to simulate the time-varying vibrating behavior of a target string. A fast training algorithm combining BPTT and SARPROP is also proposed to train the network automatically. The well-trained recurrent network based model can be regarded as a close simulation model of a real string, and using the model, it becomes easier to understand the characteristics of a string under different conditions when it vibrates.
Neural-network-based adaptive hybrid-reflectance model for 3-D surface reconstruction
(2005-11-01)
Lin, CT
;
Cheng, WC
;
Liang, SF
;
生物科技學系
;
資訊工程學系
This paper proposes a novel neural-network-based adaptive hybrid-reflectance three-dimensional (3-D) surface reconstruction model. The neural network automatically combines the diffuse and specular components into a hybrid model. The proposed model considers the characteristics of each point and the variant albedo to prevent the reconstructed surface from being distorted. The neural network inputs are the pixel values of the two-dimensional images to be reconstructed. The normal vectors of the surface can then be obtained from the output of the neural network after supervised learning, where the illuminant direction does not have to be known in advance. Finally, the obtained normal vectors are applied to enforce integrability when reconstructing 3-D objects. Facial images and images of other general objects were used to test the proposed approach. The experimental results demonstrate that the proposed neural-network-based adaptive hybrid-reflectance model can be successfully applied to objects generally and perform 3-D surface reconstruction better than some existing approaches.
Recurrent neural-network-based physical model for the chin and other plucked-string instruments
(2000-11-01)
Liang, SF
;
Su, AWY
;
電控工程研究所
;
Institute of Electrical and Control Engineering
A new physical model with neural networks is presented. The structure of the network is designed for the analysis of plucked-string instruments, and this network is also used as the corresponding synthesis engine. The proposed approach also provides a general and automatic way of determining suitable synthesis parameters by using a supervised neural network training algorithm with recorded sounds of a specific played instrument as the training vector. This is a general method and can be used for any plucked-string instrument. A traditional Chinese plucked-string instrument, called the Chin, is used as the target instrument to demonstrate this new synthesis method.
Support-vector-based fuzzy neural network for pattern classification
(2006-02-01)
Lin, CT
;
Yeh, CM
;
Liang, SF
;
Chung, JF
;
Kumar, N
Fuzzy neural networks (FNNs) for pattern classification usually use the backpropagation or C-cluster type learning algorithms to learn the parameters of the fuzzy rules and membership functions from the training data. However, such kinds of learning algorithms usually cannot minimize the empirical risk (training error) and expected risk (testing error) simultaneously, and thus cannot reach a good classification performance in the testing phase. To tackle this drawback, a support-vector-based fuzzy neural network (SVFNN) is proposed for pattern classification in this paper. The SVFNN combines the superior classification power of support vector machine (SVM) in high dimensional data spaces and the efficient human-like reasoning of FNN in handling uncertainty information. A learning algorithm consisting of three learning phases is developed to construct the SVFNN and train its parameters. In the first phase, the fuzzy rules and membership functions are automatically determined by the clustering principle. In the second phase, the parameters of FNN are calculated by the SVM with the proposed adaptive fuzzy kernel function. In the third phase, the relevant fuzzy rules are selected by the proposed reducing fuzzy rule method. To investigate the effectiveness of the proposed SVFNN classification, it is applied to the Iris, Vehicle, Dna, Satimage, Ijenn1 datasets from the UCI Repository, Statlog collection and IJCNN challenge 2001, respectively. Experimental results show that the proposed SVFNN for pattern classification can achieve good classification performance with drastically reduced number of fuzzy kernel functions.