Muhammad Imran Ahmad

Non-stationary feature fusion of face and palmprint multimodal biometrics

Multimodal biometrics provides high performance in biometric recognition systems with respect to unibiometric systems as they offer a more universal approach, added security and better recognition accuracy. Moreover, data acquisition at the feature level brings out rich information from the traits, thus fusion of modalities at this level is desirable. In this paper we propose a novel fusion technique called non-stationary feature fusion where a new structure of interleaved matrix is constructed using local features extracted from two modalities i.e. face and palmprint images. A block based Discrete Cosine Transform (DCT) algorithm is used to construct a fused feature vector by extracting independent feature vectors from each spatial image. This fused feature vector contains nonlinear information that is used to train a Gaussian Mixture Models (GMM) based statistical model. The model provides accurate estimation of the class conditional probability density function of the fused feature vector. The proposed method produces recognition rates as high as 99.7% and 97% when tested on benchmark databases-ORL-PolyU and FERET-PolyU respectively. These rates are achieved using 23% low frequency DCT coefficients. The new technique is shown to outperform existing feature level fusion methods including methods based on matching and decision level fusion.

Classification of EEG signals for brain-computer interface applications: Performance comparison

This paper presents a comparison of Electroencephalogram (EEG) signals classification for Brain Computer-Interfaces (BCI). At present, it is a challenging task to extract the meaningful EEG signal patterns from a large volume of poor quality data and simultaneously with the presence of artifacts noises. Selection of the effective classification technique of the EEG signals at classification stage is very important to get the robust BCI system. Support Vector Machine (SVM), k-Nearest Neighbour (k-NN), Multi-Layer Perceptron Artificial Neural Network (MLP-ANN) and Logistic Regression (LR) were evaluated in this paper. A BCI competition IV — Dataset 1 is used for testing the classifiers. It is shown that LR and SVM are the most efficient classifier with the highest accuracy of 73.03% and 68.97%.

Investigation of information fusion in face and palmprint multimodal biometrics

This paper reviews several information fusion techniques and strategies in the application of multimodal biometrics system using face and palmprint images. Multimodal biometric is able to overcome several limitations in single modal biometric such as intra-class variations, less discriminative power, noise data and redundant features. By consolidating two kinds of modality a better performance can be achieved. Information fusion in multimodal biometrics can be carried out at three possible levels, i.e. feature, matching score and decision levels. Fusions at these three levels have their own attributes, thus this paper is aimed to compare their effectiveness. A specific fusion rule is necessary to combine the information at each level. Several numbers of analyses on verification and identification shows matching score fusion is able to achieve the best performance which is 98% recognition rates and 98.5% GAR at 0.1% FAR when tested using AR face and PolyU palmprint datasets.

An efficient processing element architecture for pairwise sequence alignment

One of the most challenging tasks in sequence alignment is its repetitive and time-consuming alignment matrix computations. Alignment matrix scores are crucial for identifying regions of homology between biological sequences. In this paper, a parametrizable and area efficient processing element (PE) architecture for performing biological sequence alignment task especially for pairwise biological sequence alignment is designed. Its corresponding PE architecture realization was prototyped on Xilinx FPGA platform. FPGA has been chosen as it able to realize an array of systolic array-based PEs. Execution of the proposed parameterizable PE architecture have been conducted and comparison results have shown that the systolic arrays with parameterizable PE has gained at least 15x speed-up as compared to the well-known SSEARCH 35 solution.

Fusion of Low Frequency Coefficients of DCT Transform Image for Face and Palmprint Multimodal Biometrics

In this paper, we propose multimodal biometric feature fusion using alternating concatenation of DCT coefficients exist in face and plamprint images. Discrete cosine transform (DCT) is used to extract low frequency features which has high discrimination feature at the top left corner of the DCT transform image. The fuse feature vector is projected to the most principal component of eigenvector to produces low dimensional fused feature vector which contains important information about the face and palmprint images. Distance classifier is then implemented as a classifier to compute the nearest distance of test feature data point with a template to evaluate the recognition process. PolyU and FERET dataset is used to validate the propose method and the result shows fusion by using alternating concatenation of face and palmprint is able to produce a better recognition rates compare to concatenation method. The best recognition rate is 95%.

Improving speaker verification using MFCC order

This paper presents a text-dependent speaker verification using Mel-Frequency Cepstral Coefficients (MFCC) and Support Vector Machine (SVM). Mel-Frequency Cepstral Coefficients technique has been used to extract the characteristic from the recorded voice spoken by the user and SVM is used to classify the all models of the speakers and impostors. A Malay spoken digit database is utilized for the training and testing. The aim of this paper is to improve the performance of SVM by selecting the best order of Mel-Frequency Cepstral Coefficients. Five types of Mel-Frequency Cepstral Coefficients order (5, 10, 15, 20, 25) have been tested and classified using SVM. It is shown that 20th and 25th order of MFCC achieved the best total success rate (TSR) and Equal Error Rate (EER).

Combination of gait multiple features at matching score level

This paper focus to analyze several fusion rule at matching score level to combine important features extracted from gait sequence images for human identification system. Gait sequence image is a non-stationary data and can be modelled using a statistical learning technique. The propose technique consists of three different stages. The pre-processing stage computes the average silhouette images to capture the important information and get a better representation for gait silhouette data. Then a principle component analysis (PCA) technique is applied on the average silhouette to extract the important gait features and reduce a dimension of gait data. Three different features are fused at matching score level by using sum, product and max rule. The proposed algorithm has been tested using a benchmark CASIA datasets. The experimental results show that the best recognition rate is 90% when the fusion is performed using sum rule.

Score level normalization and fusion of iris recognition

Biometric traits such as an iris texture is one of the dependable physiological biometric traits because of its uniqueness. In this paper, we explore a different approach of matching score fusion and the effect of normalization method to the fusion process. Despite a plenty of work of iris recognition methods have been proposed in recent years, many are paying attention to the feature extraction process and classification method. Less number of method focuses on the information fusion of iris images. Fusion is believed to produce a better discrimination power due to the rich information can be utilized from both of iris images. We conduct an analysis to investigate which fusion rule is able to produce the best result for iris recognition system. Experimental analysis using CASIA dataset shows sum rule fusion produces 99% recognition accuracy. The verification analysis shows the best result is GAR = 95% at the FRR = 0.1% when using min-max normalization method to preprocess the matching score before the fusion process.

Improving hybrid speaker verification in noisy environments using least mean-square adaptive filters

In this paper, we present a hybrid speaker verification system based on the Hidden Markov Models (HMMs) and Vector Quantization(VQ) and Least Mean-Square (LMS) adaptive filtering. The aim of using hybrid speaker verification is to improve the HMMs performance, while LMS adaptive filtering is to improve the hybrid speaker verification performance in noisy environments. A Malay spoken digit database is used for the training and testing. It is shown that, in a clean environment a Total Success Rate (TSR) of 99.97% is achieved using hybrid VQ and HMMs. For speaker verification, the true speaker rejection rate is 0.06% while the impostor acceptance rate is 0.03% and the equal error rate (EER) is 11.72%. In noisy environments without LMS adaptive filtering TSRs of between 62.57%-76.80% are achieved for Signal to Noise Ratio (SNR) of 0-30 dBs. Meanwhile, after LMS filtering, TSRs of between 77.31%-76.87% are achieved for SNRs of 0-30 dB.

An embedded delta modulator system for coding speech signals

This paper discusses the comparison between two modulation systems, pulse code modulation and delta modulation. The objective is to identify a suitable modulation for speech coding by comparing delta modulation (DM) with pulse code modulation (PCM). The reconstruction performance of the delta modulation is compared with the pulse code modulation by using MATLAB, SIMULINK and finally implemented DSP Processor for real time realization. The simulation result using SIMULINK shows that the delta modulation performs better than pulse code modulation. Hence, delta modulator is implemented for real time test on DSP board, TMS320C6416. Delta modulation is successfully implemented in real time realization and can be further improved to reduce noise for future works.