Browsing by Author "Mathekga, Mmamolatelo E"

Now showing 1 - 11 of 11
  • Thumbnail Image
    Item
    The application of Hough transform-based fingerprint alignment on match-on-card
    (Academic Conferences and Publishing International, 2015-03) Mlambo, S; Nelwamondo, Fulufhelo V; Mathekga, Mmamolatelo E
    Smart cards are one of the most commonly used technologies that are integrated with fingerprints to develop higher security. One of the important uses of smart cards is the development of authentication systems for an individual. The demand of these cards, has led to the need for further improvements on smart cards combined with fingerprint biometrics. Due to the insufficient memory space and few instruction sets in Java smart cards, developers and programmers are faced with implementing efficient applications with limited memory space. This paper presents the application of an improved Hough Transform-based fingerprint alignment on a Java match on card. Experiments conducted determine the performance of Hough Transform based fingerprint alignment algorithm on Java smart card in terms of accuracy of alignment, memory usage and computing time. The algorithm was tested on a public database FVC2004 DB4 because this database contains rotated and translated images that will show results of alignment from the algorithm. Minutiae points were extracted from mentioned databases and used to compute alignment between two fingerprint images of the same finger that are captured at different instances. In addition, the experiments were conducted on two types of smart cards, Java Card Platform for Government ID, contact (ISO/IEC 7816) and contactless (ISO/IEC 14443). The performance measures considered in this research are memory usage and computing time since smart cards have limited resources which affect the performance of its applications. The results show that if the memory required in the alignment process is reduced, the whole matching process will also be reduced. The required memory for the entire process of alignment varies with the number of minutiae points that need to be aligned. As a result, if few minutiae points i.e. less than 20 are used, the accuracy of alignment decreases. Conversely when the number minutiae points have increased, alignment results becomes more accurate but requires more memory for processing. The average computing time that were taken to compute alignment parameters from two sets of minutiae points were depended on the number of minutiae points and the implementation of instruction sets. With sufficient amount of minutiae points the accuracy becomes more correct, however it requires more time to compute alignment parameters and process all minutiae points.
  • Thumbnail Image
    Item
    Brushless doubly fed machine magnetic field distribution characteristics and their impact on the analysis and design
    (Institute of Electrical and Electronics Engineers, 2019-12) Mathekga, Mmamolatelo E; Ademi, S; McMahon, RA
    This paper contributes to the characterisation of the brushless doubly fed induction machine (BDFIM), which is attractive as a variable speed generator in applications (off-shore wind turbine) with minimum maintenance requirements. The BDFIM has two three-phase stator windings of different pole numbers housed within the same stator slots and a short-circuited rotor winding capable of coupling fields of different pole numbers. The stator windings and rotor winding create a magnetic field distribution with a range of characteristics different to those of conventional induction machines. This paper presents an analysis to identify the field characteristics and discusses their impact on the analysis and design of the BDFIM. The characteristics are determined from an analysis of the sum of two rotating sinusoidal field waveforms and confirmed by comparison with time-stepping finite element results and measured magnetic flux density data.
  • Thumbnail Image
    Item
    Comparison of effective Hough transform-based fingerprint alignment approaches
    (IEEE, 2014-08) Mlambo, CS; Nelwamondo, Fulufhelo V; Mathekga, Mmamolatelo E
    In this paper, two effective and mostly used Hough Transform (HT) based fingerprint alignment approaches are compared, namely; Local Match Based Alignment (LMBA) and Discretized Rotation Based Alignment (DRBA). The comparison was performed by considering different conditions of minutiae points, which are rotation, translation and the number of minutiae points. In addition, this research reports the advantages of understanding the quality and relationships between the wide varieties of existing HT based fingerprint alignment methods. Minutiae points extracted from fingerprints of FVC2000 database were used on the experiments to compare these approaches. The results revealed that LMBA approach performs better than the DRBA approach on minutiae points set with larger rotation and small number of points. The DRBA approach was found to perform better with minutiae points with large amount of translation, and the computational time was less than that of LMBA approach. However, the memory usage required in DRBA is greater than memory required in LMBA.
  • No Thumbnail Available
    Item
    Ear recognition for young children
    (2019-11) Ntshangase, Cynthia S; Mathekga, Mmamolatelo E
    Recognition of children using biometrics is a current and critical world issue that needs to be addressed, for the safety and well-being of our children. There has been extensive research with regards to adult biometric recognition. It is interesting to note that for children very little work has been done. This paper presents work that has been done on ear recognition for young children (individuals from age zero to 18 years). It comprises of the significance of using biometrics to recognize children, the advantages of using ear recognition and related work that has been done on ear recognition for children. In closure it seeks to unravel unanswered questions and looks to future prospects.
  • Thumbnail Image
    Item
    Fighting identity theft with advances in fingerprint recognition
    (CSIR, 2015-10) Mathekga, Mmamolatelo E
    The ease with which the green South African ID book could be forged has led to many instances of identity fraud, costing retail businesses millions in lost revenue on fraudulently created credit accounts. This has led the government, through the Department of Home Affairs, to look into using biometrics as a means to confirm identity, as this has proven to be a more reliable means of confirming identity. To support the government in its efforts to use biometrics, which are metrics related to human characteristics, the CSIR undertook to develop biometric research capabilities. With fingerprints being one of the most commonly used biometrics, most of the initial efforts have been directed at developing capability in biometric recognition using fingerprints. Initial efforts were directed at developing algorithms for performing key fingerprint image processing operations that are required in automated fingerprint recognition systems. The algorithms were then packaged into a software library and researchers continue to make improvements. The talk will give an overview of the algorithms developed and a comparison of their performance to those in the commercial software market.
  • Thumbnail Image
    Item
    Fingerprint segmentation: an investigation of various techniques and a parameter study of a variance-based method
    (ICIC International, 2011-09) Msiza, IS; Mathekga, Mmamolatelo E; Nelwamondo, Fulufhelo V; Marwala, T
    Fingerprint image segmentation plays an important role in any fingerprint image analysis implementation and it should, ideally, be executed during the initial stages of a fingerprint manipulation process. After careful consideration of various fingerprint segmentation approaches, this manuscript focuses on a block-wise method that is based on the gray-level variance of the image. Because the method of interest is subjected to a number of variable parameters, this document then presents a formal study of these parameters, using a carefully chosen set of experiments. This series of experiments is conducted on database Db1 a of the 2002 version of the Fingerprint Verification Competition (FVC2002).
  • Thumbnail Image
    Item
    An improved Hough transform-based fingerprint alignment approach
    (IEEE, 2014-11) Mlambo, CS; Nelwamondo, Fulufhelo V; Mathekga, Mmamolatelo E
    An improved Hough Transform based fingerprint alignment approach is presented, which improves computing time and memory usage with accurate alignment parameter (rotation and translation) results. This is achieved by studying the strengths and weaknesses of existing Hough Transform based fingerprint alignment algorithms, and combining the strengths to an improved approach. The results of alignment parameters are checked manually for each image after alignment. The experimental results indicated that the improved approach improves accuracy at a less computing time and memory usage. The experiments were performed using FVC2000 and FVC2004 databases as there are different impressions of fingerprints that represent possible fingerprint rotations and translations.
  • Thumbnail Image
    Item
    Similarity score computation for minutiae-based fingerprint recognition
    (IEEE, 2014-09) Khanyile, NP; De Kock, A; Mathekga, Mmamolatelo E
    This paper identifies and analyses the factors that contribute to the similarity between two sets of minutiae points as well as the probability that two sets of minutiae points were extracted from fingerprints of the same finger. Minutiae-based fingerprint matching has been studied extensively in the literature, however, there is still a need for major improvement especially when it comes to comparing partial fingerprints. This paper looks at existing similarity measures; discusses their performance at discriminating between minutiae points from fingerprints of the same finger and of different fingers. The matching problem has been broken down into smaller subproblems which are easier to define and solve. Each of the scores discussed are analyzed and tested to see if they are able to deal with each of the matching subproblems. Results show that most scores in the literature fall in one of two ends of matching; good at discriminating impostor matches, or good at discriminating genuine matches. The authors propose a score which bridges these two types of scores and enables optimal impostor and genuine comparisons.
  • Thumbnail Image
    Item
    Singular point detection algorithm based on the transition line of the fingerprint orientation image
    (2009-11) Mathekga, Mmamolatelo E; Msiza, IS
    A new algorithm for identifying and locating singular points on a fingerprint image is presented. This algorithm is based on properties of the fingerprint orientation image, including a feature defined as a transition line. The transition line is shown to provide an efficient means of both tracking and classifying singular points without having to search through the entire fingerprint image. The algorithm is shown to be precise and its performance is evaluated using three measures, namely, the detection accuracy (DA), the detection time (DT) and a Humanversus-Algorithm (HvA) score. Using a dataset from the Fingerprint Verification Competition 2002 (FVC 2002), the performance of the algorithm is: DA = 89.6 %, DT = 0.48 s and HvA = 0.92.
  • Thumbnail Image
    Item
    A study of Hough Transform-based fingerprint alignment algorithms
    (Foundation of Computer Science, 2014-10) Mlambo, CS; Mathekga, Mmamolatelo E; Nelwamondo, Fulufhelo V
    This paper classifies existing Hough Transform fingerprint alignment algorithms and compares their performance to determine the one that gives optimal alignment results (translation and rotation). The classification is performed by considering the implementation of each algorithm. The comparison is performed by considering the alignment results computed using each group of algorithms when varying number of minutiae points, rotation angle, and translation. In addition, the memory usage, computing time and accuracy are taken into consideration. The experiments were performed on a small database where fingerprints were captured in different orientations and locations and on the public database FVC2004. Three classes of Hough Transform-Based approaches were classified as the Local Match Based Alignment(LMBA), Discretized Rotation Based Alignment(DRBA) and Matching Pair Based Alignment (MPBA). The results revealed good accuracy on all three approaches, however, the computing time and memory usage affected the performance of each approach. The LMBA approach perform better than the DRBA and the MPBA approaches on minutiae points set with larger rotation and small number of points. The DRBA approach was found to perform better with minutiae points with large amount of translation, and the computational time was less than that of LMBA and the MPBA approaches. However, the memory usage required in DRBA and MPBA for the accumulator array is greater than memory required in LMBA.
  • Thumbnail Image
    Item
    Towards a complete rule-based classification approach for flat fingerprints
    (IEEE Xplore, 2014-12) Webb, L; Mathekga, Mmamolatelo E
    Biometrics, particularly fingerprints, are becoming widely used in the information security field due to the fact that they cannot be lost or stolen as easily as a password or Personal Identification Number (PIN). A challenge with the use of fingerprints is the time taken to search through a large database to find a matching entry. For this reason, fingerprint classification is used to divide fingerprints into different bins to reduce search time. However, many fingerprint classification methods that use the number and types of singular points fail when fingerprints are captured in such a way that one or more of the singular points are missing. This is often the case when using flat fingerprints, which do not contain as much area as rolled fingerprints. This work implements an algorithm which includes new rules to account for more instances of flat fingerprints with missing singular points, specifically when the delta of a Right Loop or Left Loop fingerprint is not captured, when one of the loops of a Whorl fingerprint is not captured, and when no singular points are captured. The algorithm was tested on 833 flat fingerprint images from the FVC2002 Database 1 and 809 flat fingerprint images from the FVC2004 Database 1 and achieved accuracies of 91.1% and 91.8% respectively, which is far higher than two previous rule-based approaches applied on the same images. The additional rules thus show an improvement over previous works, specifically when applied to flat fingerprints.