Shih Yin Ooi

Offline Signature Verification through Biometric Strengthening

The offline signature verification rests on the hypothesis that each writer has similarity among signature samples, with small distortion and scale variability. In this paper we propose a novel method to increase the accuracy in biometric matching which we term biometric strengthening. We reported 1.1% equal error rate (EER) over the independent database on random forgery, while casual forgery on EER 1.2% and lastly skilled forgery on EER 2.1% along the paper. Our experiments show that biometric strengthening reduces the false acceptance rate (FAR) and false rejection rate (FRR) by increasing the disparity between the features of the two persons, which tends to tolerate more intrapersonal variance which can reduce the FRR without increasing the probability of false accepts.

Video and voice transmission over LTE networks

Video and voice transmission over wireless broadband has become popular and attracted more attention ever since. More and more hand phone owners use their phones to play video and voice over the Internet. Transmitting video and voice in a good quality over the wireless networks is a challenge to service providers. LTE, also known as one of the beyond 3G wireless network technology is designed to have a greater delivery service for multimedia, voice and video applications to end users. However, the QoS provisioning of the voice/video/multimedia in the intricacy wireless network is highly depended on the QoS framework of a network. The QoS framework for LTE network is studied and comparisons among the scheduling algorithms are made in this study. Extensive simulation results showed that the performance of the scheduling algorithms could enhance the video and voice delivery quality.

Compatibility of biometric strengthening with probabilistic neural network

There are growing concerns about the privacy invasion of the biometric technology. This is due to the fact that biometric characteristics are immutable and hence their compromise is permanent. Thus, reissuable biometrics was devised to denote biometric templates that can be reissued and replaced. Biometric Strengthening is a form of reissuable biometrics which strengthens the biometric templates by altering their original values thru the Gaussian distribution, thus generating a new set of values. However, the main drawback of Biometric Strengthening is its great degradation in performance when the legitimate token is stolen and used by the imposter to claim as the legitimate user. In this paper, we employ the probabilistic neural network (PNN) as the classifier to alleviate this problem. The compatibility of Biometric Strengthening with PNN is discussed, along with the experiments that are tested on our own independent offline signature data set.

Experimental Study of Elman Network in Temporal Classification

Elman network is an extension of multilayer perceptron (MLP), where it introduces single hidden layer architecture, as well as an additional state table to store the time units for hidden neurons. This additional state table allows it to do the sequential prediction which is not possible in MLP. To examine its general performance as a temporal classifier, a Weka version of Elman network is exploited on 11 public temporal datasets released by UCI Machine Repository.

Enhanced independent spectral histogram representations in face recognition

A spectral histogram descriptor computes a set of marginal distributions based on the filter bank’s responses, and further encodes them into the images. The encoding process for local image structure takes place during the filtering stage, whereas the encoding process of global image feature is conducted during the histogram stage. One drawback of spectral histogram descriptors is their performances will be greatly deteriorated when the filter bank’s responses are not stochastically independent. To tackle this problem, a computational technique named Enhanced Independent Spectral Histogram Feature (EISHF) is proposed. EISHF is composed of four working modules: (1) unsupervised independent filter bank responses computation, (2) binary hashing, (3) XOR bitwise operation and feature encoding, and lastly, (4) block-wise histogramming. To ensure the performance of ordinary spectral histogram descriptors, an XOR operation has been delicately adopted to increase the independency of the filter responses. Tested on three public face databases, the experimental results have substantiated the performance of EISHF in handling different kinds of facial expressions, illuminations, time spans as well as facial makeup effects.

Temporal Sleuth Machine with decision tree for temporal classification

Temporal data classification is an extension field of data classification, where the observed datasets are temporally related across sequential domain and time domain. In this work, an inductive learning temporal data classification, namely Temporal Sleuth Machine (TSM), is proposed. Building on the latest release of C4.5 decision tree (C4.8), we consider its limitations in handling a large number of attributes and inherited information gain ratio problem. Fuzzy cognitive maps is incorporated in the TSM initial learning mechanism to adaptively harness the temporal relations of TSM rules. These extracted temporal values are used to revisit the information gain ratio and revise the number of TSM rules during the second learning mechanism, hence, yielding a stronger learner. Tested on 11 UCI Repository sequential datasets from diverse domains, TSM demonstrates its robustness by achieving an average classification accuracy of more than 95% in all datasets.

Classifying Human Activities with Temporal Extension of Random Forest

Sensor-Based Human Activity Recognition (HAR) is a study of recognizing the human’s activities by using the data captured from wearable sensors. Avail the temporal information from the sensors, a modified version of random forest is proposed to preserve the temporal information, and harness them in classifying the human activities. The proposed algorithm is tested on 7 public HAR datasets. Promising results are reported, with an average classification accuracy of ~98 %.

Anomaly Based Intrusion Detection through Temporal Classification

Many machine learning techniques have been used to classify anomaly-based network intrusion data, encompassing from single classifier to hybrid or ensemble classifiers. A nonlinear temporal data classification is proposed in this work, namely Temporal-J48, where the historical connection records are used to classify the attack or predict the unseen attack. With its tree-based architecture, the implementation is relatively simple. The classification information is readable through the generated temporal rules. The proposed classifier is tested on 1999 KDD Cup Intrusion Detection dataset from UCI Machine Learning Repository. Promising results are reported for denial-of-service (DOS) and probing attack types.