Dzati Athiar Ramli

Automatic syllables segmentation for frog identification system

Automatic recognition of frog sound according to particular species is considered a worthy tool for biological research and environmental monitoring. As a result, automatic recognition of frog sound offers many advantages rather than manual method that depending on physical observation procedure. This study evaluates the accuracy of frog sound identification from 12 species that recorded from Malaysia forest. By applying short time energy and short time average zero crossing rate, the frog sound samples are automatically segmented into syllables. A syllable feature extraction method i.e, Mel-Frequency Cepstrum Coefficients is employed to extract the segmented signal. Finally, nonparametric k-nearest neighbor classifier with Euclidean distance has been employed to recognize the frog species. A comparison between automatic segmentation and manual segmentation is applied and results show that automatic segmentation outperforms to identify the frog species with an accuracy of 97% compared to 82.33% for manual segmentation.

Comparative Study on Feature, Score and Decision Level Fusion Schemes for Robust Multibiometric Systems

Multibiometric system employs two or more behavioral or physical information from a person’s traits for the verification and identification processes. Many researches have proved that multibiometric system can improve the performances of single biometric system. In this study, three types of fusion levels i.e feature level fusion, score level fusion and decision level fusion have been tested. Feature level fusion involves feature concatenation of the features from two modalities before the pattern matching process while score level fusion is executed by calculating the mean score from both biometrics scores produced after the pattern matching. Finally, for the decision level fusion, the logic AND and OR are performed on the final decision of the two modalities. In this study, speech signal is used as a biometric trait to the biometric verification system while lipreading image is used as a second modality to assist the performance of the single modal system. For speech signal, Mel Frequency Ceptral Coefficient (MFCC) is used as speech features while region of interest (ROI) of lipreading is used as visual features. Consequently, support vector machine (SVM) is executed as classifier. Performances of the systems for each fusion level is compared based on accuracy percentage and Receiver Operation Characteristic (ROC) curve by plotting Genuine Acceptance Rate (GAR) versus False Acceptance Rate (FAR. Experimental results show that score level fusion performance is the most outstanding method followed by feature level fusion and finally the decision level fusion. The accuracy percentages using 20 training data are observed as 99.9488%, 99.7534% and 99.6639% for the score level fusion, feature level fusion and decision level fusion, respectively.

Frog Sound Identification System for Frog Species Recognition

Physiological research reported that certain frog species contain antimicrobial substances which is potentially and beneficial in overcoming certain health problem. As a result, there is an imperative need for an automated frog species identification to assist people in physiological research in detecting and localizing certain frog species. This project aims to develop a frog sound identification system which is expected to recognize frog species according to the recorded bio acoustic signals. The Mel Frequency Cepstrum Coefficient (MFCC) and Linear Predictive Coding (LPC) are used as the feature extraction techniques for the system while the classifier employed is k-Nearest Neighbor (K-NN). Database from AmphibiaWeb has been used to evaluate the system performances. Experimental results showed that system performances of 98.1% and 93.1% have been achieved for MFCC and LPC techniques, respectively.

A multi-sample single-source model using spectrographic features for biometric authentication

In this paper we propose a novel approach by using spectrographic features and correlation filters as classifiers to perform speaker authentication. Visual displays (spectrograms) from speech signals produced from different persons are used as features for the verification task In order to achieve better verification results, the exclusion of low energies and the inclusion morphological image processing steps are applied to the spectrograms. It is discovered that, by applying these two techniques, the verification performance improves significantly. For the classification modeling, unconstrained minimum average correlation energy (UMACE) filter is implemented. We propose a multi-sample approach by fusing multiple samples from different utterances at the score level. By using the average operator, both the theoretical and empirical results show that by integrating as many samples as possible can improve the overall reliability of the system. This model is called as multi-sample single-source (MSSS) model. A digit database has been used for performance evaluation, yielding an overall performance of 99.6%.

Diagnosis of Cervical Cancer Using Hybrid Multilayered Perceptron (HMLP) Network

Cancer of the cervix is the second most common cancer among females in Malaysia after breast cancer. In most cases, cervical cancer takes many years to develop from normal to advanced stage. Therefore, the mortality related to cervical cancer can be reduced through early detection and treatment. Pap test is one of the early diagnosis that should be done to reduce the mortality rate related to cervical cancer. Neverthenles, low accuracy, sensitivity and specificity become a problem in diagnosing cervical cancer by using the Pap test . Recently, artificial intelligent based on neural network such as radial basis function, multi-layered perceptron and modular knowledge-based network have been implemented widely as cervical cancer diagnosis system. The networks is used to classify the cervical cells into normal and abnormal cells. In this paper, a hybrid multi-layered perceptron using recursive least square algorithm is introduced to diagnose the cervical cancer. The network has high ability to classify the cervical cells into normal, low-grade squamous intraepithelial lesions and high-grade squamous intraepithelial lesions. Furthermore, it has been prove to achieve better accuracy, sensitivity and specificity with smaller false negative and false positive compared to the conventional techniques. The results also proved that by using the network which has superior ability to be implemented as cervical cancer diagnosis system, the Pap test performance can be improved.

MFCC based frog identification system in noisy environment

Identification of frog sound is useful tool and competent in biological research and environmental monitoring. In contrast with traditional methods that not practical due to the time consuming, expensive or detrimental to the animals welfare, this study proposes an automatic frog call identification system. 750 data species that recorded from Malaysia forest is used as data signals and have been corrupted by 10dB and 20dB noise to determine the performance of accuracy in noisy environment. MFCC parameter is employed as feature extraction. An analysis of signals for different number of MFCCs (8, 12, 15, 20 and 25) is presented and the results are provided using MFCC, Delta Coefficients (ΔMFCC) and Delta Delta Coefficients (ΔΔMFCC). Subsequently, kNN classifier is applied to evaluate the performance in the frog identification system. The results show the accuracy range from 84.67% to 85.78%, 61.33% to 68.89% and 59.33% to 67.33% in clean environment, 10dB and 20dB, respectively.

Score Information Decision Fusion Using Support Vector Machine for a Correlation Filter Based Speaker Authentication System

In this paper, we propose a novel decision fusion by fusing score information from multiple correlation filter outputs of a speaker authentication system. Correlation filter classifier is designed to yield a sharp peak in the correlation output for an authentic person while no peak is perceived for the imposter. By appending the scores from multiple correlation filter outputs as a feature vector, Support Vector Machine (SVM) is then executed for the decision process. In this study, cepstrumgraphic and spectrographic images are implemented as features to the system and Unconstrained Minimum Average Correlation Energy (UMACE) filters are used as classifiers. The first objective of this study is to develop a multiple score decision fusion system using SVM for speaker authentication. Secondly, the performance of the proposed system using both features are then evaluated and compared. The Digit Database is used for performance evaluation and an improvement is observed after implementing multiple score decision fusion which demonstrates the advantages of the scheme.

Person identification using lip motion sequence

In this paper, we propose a new method based on lip motion sequence and the Unconstrained Minimum Average Correlation Energy (UMACE) filter as a classifier for person identification. UMACE filter has advantages in terms of their characteristics such as distortion-tolerant, shift-invariant and high discrimination ability. Therefore, executing UMACE filter provides an ideal application for person identification because we deal with the variations of the lip appearance in each frame sequence. The results obtained using a Digit Database show that the use of lip motion sequence and UMACE filter offer a good potential and can be an alternative technique for a person identification system.

Multilayered perceptron (MLP) network trained by recursive least squares algorithm

In my research, the performance of multilayered perceptron (MLP) network which trained by recursive least square (RLS) algorithm is investigated. The network has been implemented to classify the cervical cells into normal, low-grade squamos intraepithelial lesion(LSIL) and high-grade squamos intraepithelial lesion(HSIL). Based on Bathesda System, it has achieved to classify the cervical cells with high accuracy, sensitivity and specificity as well as lower false negative and false positive but more work should be done to enhance the system accuracy.

A robust and fast computation touchless palm print recognition system using LHEAT and the IFkNCN classifier

Mobile implementation is a current trend in biometric design. This paper proposes a new approach to palm print recognition, in which smart phones are used to capture palm print images at a distance. A touchless system was developed because of public demand for privacy and sanitation. Robust hand tracking, image enhancement, and fast computation processing algorithms are required for effective touchless and mobile-based recognition. In this project, hand tracking and the region of interest (ROI) extraction method were discussed. A sliding neighborhood operation with local histogram equalization, followed by a local adaptive thresholding or LHEAT approach, was proposed in the image enhancement stage to manage low-quality palm print images. To accelerate the recognition process, a new classifier, improved fuzzy-based k nearest centroid neighbor (IFkNCN), was implemented. By removing outliers and reducing the amount of training data, this classifier exhibited faster computation. Our experimental results demonstrate that a touchless palm print system using LHEAT and IFkNCN achieves a promising recognition rate of 98.64%.