Hafizah Husain

Reduced set support vector machines: Application for 2-dimensional datasets

This paper presents the performance of the reduced set (RS) method to approximate the decision boundary for standard support vector machines (SVM) classifier without affecting its generalization performance. The main focus of this work is to demonstrate the capability of the RS method such that even with fewer set of vectors, the generalization performance is not affected. In evaluating the RS method performance, decision boundaries obtained using RS method were benchmarked against the decision boundaries obtained from the standard SVM using sequential minimal optimization (SMO) method. Specifically, the generalization ability of the two methods is not evaluated since the main objective is to analyze the effect of reduced set vector in producing approximation of SVM decision rules. Results obtained demonstrated that the SVM classifier using RS method is comparable with the standard SVM using SMO method. In fact, the RS method is better since it uses fewer set of vectors to produce similar decision boundaries while maintaining the generalization performances.

Implementation of Emotional-Aware Computer Systems Using Typical Input Devices

Emotions play an important role in human interactions. Human Emotions Recognition HER - Affective Computing is an innovative method for detecting users emotions to determine proper responses and recommendations in Human-Computer Interaction HCI. This paper discusses an intelligent approach to recognize human emotions by using the usual input devices such as keyboard, mouse and touch screen displays. This research is compared with the other usual methods like processing the facial expressions, human voice, body gestures and digital signal processing in Electroencephalography EEG machines for an emotional-aware system. The Emotional Intelligence system is trained in a supervised mode by Artificial Neural Network ANN and Support Vector Machine SVM techniques. The result shows 93.20% in accuracy which is around 5% more than the existing methods. It is a significant contribution to show new directions of future research in this topical area of emotion recognition, which is useful in recommender systems.

Implementation of active noise control for hearing test application using PC

Hearing screening test is a method to determine the disorder in human ears. Conventional audiometer and an audiologist are required in order to perform the hearing screening test. However, this procedure is difficult to be implemented especially in a remote site such as factory or school due to the ambient noise which may cause inaccuracy in the test. In this work, the application of active noise control (ANC) is proposed to reduce the ambient noise using a personal computer in hearing screening test. The ANC algorithm was implemented using a computer with data acquisition modules and LabVIEW software. The results showed that the anti-noise was successfully generated in the electrical domain but no reduction was observed in the acoustic domain. The ANC is a deterministic application that requires real-time operating system to respond to the input with precisely timed output. In order to have an effective ANC system, the processing time has to be less than 0.125 ms at 8 KHz sampling rate.

Anterior osteoporosis classification in cervical vertebrae using fuzzy decision tree

Anterior Osteoporosis (AOs) in the cervical vertebrae is an osteoporosis complication and a common condition of vertebral irregularity caused by a decrease in bone density and strength, which can lead to fragile bone and fractures. Consequently, it is crucial to detect the AOs irregularity early so that appropriate pharmacological intervention can be done to reduce further complications. To do so via a computer approach, an efficient method that can provide high classification rate is required. Basing on the fuzzy logic theory, this article affords a new method for AOs (classes and severity) classification of the cervical radiography by designing a fuzzy decision tree (FDT) model. The method involves two main processed, namely i) segmentation process which employs the active shape model (ASM) based on the 9-anatomical points representation (9-APR) to segment the cervical vertebra shape boundary (C-VSB) and ii) fuzzy based feature extraction and classifier development, known as FDT method. The fuzzy set along with its membership functions are derived from the resulting C-VSB segment. It operates by extracting a specific angle descriptor (horizontal, vertical, and corner) as crisp input to the fuzzification inter-face to produce reasonable key indexing to the fuzzy interface system. Then, the defuzzification interface converts it into a crisp output that adequately represents the degree of AOs class and severity as appraisal values. The resulting fuzzy then acts as input to a basic concept of if-then rules called FDT to recognise and distinguish between vertebrae presented with/without AOs. Receiver operating characteristic (ROC) and area under curve (AUC) index evaluation methods are examined to offer quantitative evaluation between the medical ground truth versus FDT classifier predicted results. Results obtained on a set of 400 cervical vertebrae images indicate superb classification rate (R > 90 %) which suggest that the proposed FDT as an appropriate solution to AOs classification process for reliable vertebral fracture diagnosis. In summary, the findings confirmed the effectiveness of FDT as an excellent classifier to recognize and differentiate AOs classes and severity thus, able to provide important basis for pathology.

A Secure and Robust Compressed Domain Video Steganography for Intra- and Inter-Frames Using Embedding-Based Byte Differencing (EBBD) Scheme.

This paper presents a novel secure and robust steganographic technique in the compressed video domain namely embedding-based byte differencing (EBBD). Unlike most of the current video steganographic techniques which take into account only the intra frames for data embedding, the proposed EBBD technique aims to hide information in both intra and inter frames. The information is embedded into a compressed video by simultaneously manipulating the quantized AC coefficients (AC-QTCs) of luminance components of the frames during MPEG-2 encoding process. Later, during the decoding process, the embedded information can be detected and extracted completely. Furthermore, the EBBD basically deals with two security concepts: data encryption and data concealing. Hence, during the embedding process, secret data is encrypted using the simplified data encryption standard (S-DES) algorithm to provide better security to the implemented system. The security of the method lies in selecting candidate AC-QTCs within each non-overlapping 8 × 8 sub-block using a pseudo random key. Basic performance of this steganographic technique verified through experiments on various existing MPEG-2 encoded videos over a wide range of embedded payload rates. Overall, the experimental results verify the excellent performance of the proposed EBBD with a better trade-off in terms of imperceptibility and payload, as compared with previous techniques while at the same time ensuring minimal bitrate increase and negligible degradation of PSNR values.

Comparative analysis of steganographic algorithms within compressed video domain

Steganographic techniques with varying degrees of embedding capacity, perceptual transparency, and security are presented. These techniques have been developed to protect privileged or confidential information restricted to public access and to replace cryptography methods. In this study, three embedding algorithms, namely, least significant bit insertion, bit-plane complexity segmentation, and enhanced version of pixel value difference (EPVD) were implemented and analyzed in terms of the main steganography issues (payload, invisibility, and security) using a different proper performance metrics. This study was motivated by the minimal research focus accorded to hiding data in compression domain for the class of video-based embedding methods. Therefore, analyzing steganographic algorithms is generally based on hiding information in the quantized AC-Coefficients of the frames during the MPEG-2 compression process. Simulation results reveal that EPVD provides better embedded payload and acceptable visual quality, but lower PSNR value.

Development of VLF receiver for remote sensing of low atmospheric activities

VLF (very low frequency) can be used to probe lower atmospheric activities. To detect lower atmospheric activities, two types of VLF receiver were developed in this project. The first type is called SID (sudden ionospheric disturbances) receiver which is used to detect the solar flares activities that affect communication systems. The second one is called natural radio receiver which is used to detect natural radio signals such as sferics, whistlers emitted by the activities of lightning discharges. A wire loop antena was developed to receive VLF signals from outer space for the receivers that has been developed. In the process of developing the VLF receiver, MULTISIM software. It was used to simulate the receivers circuit developed for this purpose. The simulation results was adopted to design a prototype for the circuit for experiments carried out to measure the performance and the ability of the receivers in receiving VLF signals. Lab equipment, such as signal generators and oscilloscopes, were used to test the performance of the receivers. The receivers developed will contribute to the research of D-layer ionospheric activities.

Statistical analysis approach for posture recognition

The aim of this study is to determine the best eigenfeatures of four main human postures based on the rules of thumb of Principal Component Analysis namely the KG-rule, Cumulative Variance and the Scree Test followed by statistical analysis. Accordingly, all three rules of thumb suggest the retention of only 35 main principle components or eigenvalues. Next, these eigenfeatures that we named as dasiaeigenposturespsila are statistically analyzed prior to classification. Thus, the most relevant component of the selected eigenpostures can be ascertained. The statistical significance of the eigenpostures is determined using ANOVA. Further, a multiple comparison procedure (MCP) and homogeneous subsets tests are performed to determine the number of optimized eigenpostures for classification. artificial neural network (ANN) and support vector machine (SVM) were employed for classification. Results attained that the statistical analysis has enabled us to perform effectively the selection of eigenpostures for classification of human postures.

Active Noise Control for Hearing Screening Test: Simulation and Experiment

A hearing screening test is a method to determine human ear disorders and conventional audiometers and audiologists are required to perform the test. However, this procedure is difficult to implement, especially in a remote site such as a factory or a school due to the ambient noise that may cause test inaccuracy. In this work, the application of active noise control ANC is proposed to reduce the ambient noise using a personal computer in a hearing screening test. The ANC algorithm was simulated in MATLAB software and implemented using a computer with data acquisition modules and LabVIEW software. Results show that anti-noise was successfully generated in the electrical domain but no reduction was observed in the acoustic domain. ANC is a deterministic application that requires a real-time operating system to respond to the input with precisely timed output. To have an effective ANC system, the processing time has to be less than 0.125 ms at 8 KHz sampling rate.

Shock Posture for Shape Matching

In this paper, the aim is to develop a framework to efficiently match human posture. We use shock graphs as our shape representation. The algorithms developed are for detection of end points and junction points present in every region of the shock graph for modelling of human posture. We propose a relatively simple posture matching system based on key points correspondences. It is shown that this framework is effective in matching human shape posture. The algorithms were tested with wide varieties of 2D human shapes representing a range of posture. Six key points were determined and preserved. These key points serve as tool for matching of human postures. Initial results of the experiments are encouraging and found that our method can efficiently be applied for posture matching. Research on this promising approach is ongoing, focusing both on new applications and on techniques that further enhance its performance