Yee Wan Wong

Radial Basis Function Neural Network With Incremental Learning for Face Recognition

Conventional face recognition suffers from problems such as extending the classifier for newly added people and learning updated information about the existing people. The way to address these problems is to retrain the system which will require expensive computational complexity. In this paper, a radial basis function (RBF) neural network with a new incremental learning method based on the regularized orthogonal least square (ROLS) algorithm is proposed for face recognition. It is designed to accommodate new information without retraining the initial network. In our proposed method, the selection of the regressors for the new data is done locally, hence avoiding the expensive reselecting process. In addition, it accumulates previous experience and learns updated new knowledge of the existing groups to increase the robustness of the system. The experimental results show that the proposed method gives higher average recognition accuracy compared to the conventional ROLS-algorithm-based RBF neural network with much lower computational complexity. Furthermore, the proposed method achieves higher recognition accuracy as compared to other incremental learning algorithms such as incremental principal component analysis and incremental linear discriminant analysis in face recognition.Conventional face recognition suffers from problems such as extending the classifier for newly added people and learning updated information about the existing people. The way to address these problems is to retrain the system which will require expensive computational complexity. In this paper, a radial basis function (RBF) neural network with a new incremental learning method based on the regularized orthogonal least square (ROLS) algorithm is proposed for face recognition. It is designed to accommodate new information without retraining the initial network. In our proposed method, the selection of the regressors for the new data is done locally, hence avoiding the expensive reselecting process. In addition, it accumulates previous experience and learns updated new knowledge of the existing groups to increase the robustness of the system. The experimental results show that the proposed method gives higher average recognition accuracy compared to the conventional ROLS-algorithm-based RBF neural network with much lower computational complexity. Furthermore, the proposed method achieves higher recognition accuracy as compared to other incremental learning algorithms such as incremental principal component analysis and incremental linear discriminant analysis in face recognition.

Dual optimal multiband features for face recognition

Illumination and expression variations degrade the performance of a face recognition system. In this paper, a novel dual optimal multiband features method for face recognition is presented. This method aims to increase the robustness of face recognition system to both illumination and expression variations. The wavelet packet transform decomposes image into frequency subbands and the multiband feature fusion technique is incorporated to select optimal multiband feature sets that are invariant to illumination and expression variation separately. Parallel radial basis function neural networks are employed to classify the two sets of feature. The scores generated are then combined and processed by an adaptive fusion mechanism. In this mechanism, the level of illumination variations of the input image is estimated and the weights are assigned to the scores accordingly. Experiments based on Yale, YaleB, AR and ORL databases show that the proposed method outperformed other algorithms.

New Parallel Models for Face Recognition

Subspace methods such as principal component analysis (PCA) and linear discriminant analysis (LDA) extract the features based on space domain. Transformation such as discrete cosine transform (DCT) extracts features based on frequency domain. In this paper, we present two parallel models which intend to utilize the features extracted from frequency and space domain of facial images. Both features are combined under a fusion based scheme. FERET database is chosen to evaluate the performance of the proposed method. Simulation results indicate that the proposed method outperforms other traditional methods and enhance the representation of facial image under low-dimensional features.

Audio-Visual Recognition System with Intra-Modal Fusion

In this paper, a new multimodal biometric recognition system based on feature fusion is proposed to increase the robustness and circumvention of conventional multimodal recognition system. The feature sets originating from the output of the visual and audio feature extraction systems are fused and being classified by RBF neural network. Other than that, 2DPCA is proposed to work in conjunction with LDA to further increase the recognition performance of the visual recognition system. The experimental result shows that the proposed system achieves a higher recognition rate as compared to the conventional multimodal recognition system. Besides, we also show that the 2DPCA+LDA achieves a higher recognition rate as compared with PCA, PCA+LDA and 2DPCA.

A new multi-purpose audio-visual UNMC-VIER database with multiple variabilities

Audio-visual recognition system is becoming popular because it overcomes certain problems of traditional audio-only recognition system. However, difficulties due to visual variations in video sequence can significantly degrade the recognition performance of the system. This problem can be further complicated when more than one visual variation happen at the same time. Although several databases have been created in this area, none of them includes realistic visual variations in video sequence. With the aim to facilitate the development of robust audio-visual recognition systems, the new audio-visual UNMC-VIER database is created. This database contains various visual variations including illumination, facial expression, head pose, and image resolution variations. The most unique aspect of this database is that it includes more than one visual variation in the same video recording. For the audio part, the utterances are spoken in slow and normal speech pace to improve the learning process of audio-visual speech recognition system. Hence, this database is useful for the development of robust audio-visual person, speech recognition and face recognition systems.

M-band wavelet transform in face recognition system

The low-frequency subband of DWT decomposition has been used as the important subband in face recognition system. But we believe that some subbands other than low-frequency subband contain discriminative information. Consequently, a series of experiments are carried out to test the recognition accuracy on various combinations of the low-frequency with some mid- and high-frequency subbands and subbands that perform the best are selected. We use the M-band decomposition to decompose a face image into smaller frequency subbands. The resulting best combination of low-frequency subband and some subbands named as MBCom achieves a higher recognition rate as compared to the low-frequency subband alone. The MBCom is proposed to work in conjunction with PCA and it outperforms other feature extraction methods in term of recognition rate.

A novel tree trunk detection method for oil-palm plantation navigation

A novel tree trunk detection method for oil palm plantations is proposed.A combination of colour images and depth information is used for detection.The proposed method produced a 97.8% tree trunk detection rate in field tests. This paper presents a novel tree trunk detection algorithm that uses the Viola and Jones detector along with a proposed pre-processing method, combined with tree trunk detection via depth information. The proposed method tackles the issue of the high false positive rate when the Viola and Jones detector is used on its own, due to the low contrast between tree trunks and the surrounding environment. The pre-processing method uses colour space combination and segmentation to eliminate the ground not covered by trees from the images and feeding the resulting image into a cascade detector for identifying the location of the trunks in the image. Depth information is obtained via the use of the Microsoft KINECT sensor to further increase the accuracy of the detector. Our proposed method had better performance when compared to both Neural Network based and Support Vector Machine based detectors with a detection rate of 91.7% and had the lowest false acceptance rate out of other detectors, including the original Viola and Jones detector. The performance of the proposed method was also tested on live video feeds with the use of a robot prototype in an oil-palm plantation, which proved the high accuracy of the method, with a 97.8% detection rate. The inclusion of depth information resulted in more accurate detections during low levels of light and at night, where reliance on pure depth information resulted in a 100% detection rate of tree trunks within the range of the sensor.

Incremental Unsupervised Learning Algorithm for Power Fluctuation Event Detection in PV Grid-Tied Systems

Obstacles for solar photovoltaic (PV) system to be a reliable energy source is its intermittent and stochastic output power. The randomness output power could trigger power fluctuation event. Subsequently, more power quality issues such as frequency fluctuation, voltage variation and harmonic distortion could happen. Thus, this paper introduces a Self-Organizing Incremental Neural Network (SOINN) to predict the output power and subsequently detect the power fluctuation events in order to enhance the reliability of a PV grid-tied system. The SOINN is developed from the growing neural gas and competitive hebbian learning. It could be trained without predefined the structure of the network. To train the SOINN, input data to the PV system such as irradiance and temperature are used. The trained SOINN will be compared with the Self-Organizing Map (SOM) network. Results show that the SOINN prediction engine achieves an accuracy of 100 % in identifying power fluctuation event through predicted output power.

Robust video authentication system over internet protocol

This paper presents a video authentication system over internet protocol that is insusceptible to illumination and expression variations. The illumination and expression invariant features are extracted using multi-band feature extraction. These features are classified by a radial basis function neural network. A new adaptive decision fusion method is proposed to combine the scores from different modalities and the different frames during the authentication process. Three levels of decision fusion are carried out in the proposed adaptive decision fusion. Depending on the level of decision fusion, the level of illumination influence is taken into account during the decision making.

Audio-Visual Recognition System in Compression Domain

This paper presents a highly efficient audio-visual recognition system in compression domain. For face recognition systems, the multiband feature fusion method selects the wavelet subbands that are invariant to illumination and facial expression variations. These subbands will be extracted directly from the inverse quantization in the compression system. By taking the inverse quantized wavelet coefficient of the video as the input, the inverse wavelet transform which corresponds to image reconstruction is omitted. As a result, the computational complexity of the conventional video-based face recognition system is reduced. We also present a set of new face localization methods to localize the facial wavelet coefficients from the wavelet subband image. The dual optimal multiband feature fusion method is then used to fuse the two set of wavelet coefficients and generate the visual scores. Experimental results show that with low computational complexity, the proposed system achieves high recognition accuracy in UNMC-VIER, CUAVE, and XM2VTS audio-visual database.