An efficient indexing and retrieval of iris biometrics data using hybrid transform and firefly based K-means algorithm title

• Main Author: Khalaf, Emad Taha
• Format: Thesis
• Language: English
• Published: 2019
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://umpir.ump.edu.my/id/eprint/29262/1/An%20efficient%20indexing%20and%20retrieval%20of%20iris%20biometrics%20data%20using%20hybrid%20transform%20and%20firefly%20based%20K-means%20algorithm%20title.wm.pdf; http://umpir.ump.edu.my/id/eprint/29262/

The explosive increase in the number of biometric images saved in most databases has made image indexing mandatory. These processes could influence the speed of data access as well as support their retrieval. Hence, researchers are focusing on how to determine suitable image features to be used for clustering and index, with an efficient searching process. The existing methods are unable to extract sufficient number of the most important features of iris image for clustering and indexing processes. However, one of the weaknesses of clustering is the process of extracting the most important features. A combination of three transformation methods, namely, Discrete Cosine Transformation (DCT), Discrete Wavelet Transform (DWT), and Singular Value Decomposition (SVD) for analyzing the iris image and for extracting its local features have yet to be utilized for image clustering and indexing. Another problem related to clustering is when choosing the initial centroids for each cluster randomly. To overcome this disadvantage, the Fireflies Algorithm (FA) was used because it has the ability to perform global searches and has quick convergence rate to optimize the initial clustering centers of the K-means algorithm, using a kind of weighted Euclidean distance to reduce the defects made by noise data and other uncertainties. This thesis presents a new method to extract the most relevant features of iris biometric images for indexing the database within minimum time and search area. The enhanced method combines three transformation methods for analyzing the iris image and extracting its local features. It uses a weighted K-means clustering algorithm based on the improved FA to optimize the initial clustering centers of K-means algorithm, known as Weighted K-means clustering-Improved Firefly Algorithm (WKIFA). For searches and retrieval, an efficient parallel technique has been presented by dividing the group of features into two b-trees based on index keys. Searches within a group can be done using a half-searching algorithm to improve the response time for data retrieval. The system has been tested on publicly available databases. The experimental results showed that the indexing system has a considerably low penetration rate of 0.98%, 0.13%, and 0.12%, and lower bin miss rate of 0.3037%, 0.4226%, and 0.2019% compared to the existing iris databases of the Chinese Academy of Science - Institute of Automation (CASIA), University of Bath (BATH), and Database of Indian Institute of Technology Kanpur (IITK), respectively. Results of the improved WKIFA showed that it was more effective for the clustering stage of the system. It even outperformed the traditional K-mean, by reducing the penetration rates to 0.131%, 0.088%, and 0.108%, and improving the accuracy by reducing the bin miss rate to 0.2604%, 0.309%, and 0.1548% of the aforementioned databases, respectively. Analysis of time complexity of retrieval showed that the computational complexity was reduced to O (log n), which was better than the existing methods.