A NEW 2D PULSE DOMAIN TRANSFORM (PDT) FEATURE EXTRACTION FOR FINGERPRINT BIOMETRICS

Abstract

Feature extraction is a technique to extract derived value which is informative and non-redundant. There are many techniques which have been reported in the literature on its usage and application on signal and image. Some of the common techniques of feature extraction are Fast Fourier Transform, Wavelet Transform, Histogram of Oriented Gradients (HOG), Speeded Up Robust Features (SURF), Local Binary Patterns (LBP), Haar Wavelets, and Colour Histograms. Recently, a new feature extraction technique based on Pulse Width Modulation (PWM) has been proposed. Although PWM is widely used for power control but it can be applied to feature extraction as well. Pulse Domain Transform (PDT) is the feature extraction technique that derived from PWM. Initially, PDT applies on electrocardiogram (ECG) signal and draws out a non-invertible output. This output is presented as a pulse waveform which is generated by comparing amplitude of ECG signal with a periodic triangular waveform to get the intersection points. The time location of the intersection points is then used as a transition state for output pulse to raise or fall. Since the PDT concept is only applied to 1D signal, it is interesting if the technique can be extended to a 2D signal. Therefore, this research work comes up and presents as 2D PDT feature extraction technique. Also, the concept of 2D PDT is based on the principle of PWM to process the incoming image (2D signal) with the aid of a 2D triangular waveform to form an output of 2D pulse waveform. In respect of 2D triangular waveform, it is set up based on the desired threshold for its frequency and amplitude. These thresholds are modulated in accordance with the size of the image. As for the output, the superposition process is proceeding between the image and modulated 2D triangular waveform to generate the intersection points to form the output of 2D pulse waveform. The technique is then tested on fingerprint biometric to show its ability to generate unique feature from the fingerprint images. This uniqueness takes shape which is caused by the inconstancy of the frequency and amplitude of pulse waveform. More than that, it is able to provide the matching percentage of 87% on average and for Area Under the Receiver Operating Characteristic curve (AUROC or AUC) reading of 0.86 on average. From the result, 2D PDT shows good feature extraction capability which is comparable with other conventional techniques.