Soft biometric system using fuzzy logic decision fusion for identification

• Main Author: Ayodeji, Arigbabu Olasimbo
• Format: Thesis
• Language: English
• Published: 2014
• Online Access: http://psasir.upm.edu.my/id/eprint/47972/1/FK%202014%2023R.pdf; http://psasir.upm.edu.my/id/eprint/47972/

Traditional biometrics such as fingerprint, retina and iris are highly accurate and efficient biometric modalities. However, their disadvantages include intrusiveness, inability to work at far distance, and requirement of human cooperation to function effectively. On the other hand, some biometric applications such as identification of humans from distance may not require a high degree of accuracy, yet they cannot tolerate the need for human cooperation and intrusiveness. Therefore, soft biometrics is an alternative biometric modality to perform identification of humans from distance. In this situation, soft biometrics can provide a moderate level of identification when the subjects are not cooperative with acquisition system. In addition, intrusiveness issues can be sufficiently minimized by using soft biometrics, because the attribute can be extracted without interaction with the subjects. In this thesis, the possibility of using multiple soft biometrics for identification is investigated. The thesis shows that when multiple soft biometric attributes are combined together, they can be logically applied to find a specific identity in the database. The main focus is placed on combining face and body related soft biometric attributes like facial shape, skin colour, height, and weight for identification purposes. Here, each attribute performs individual identification process, which includes sub-processes of pre-processing, feature extraction, and template matching. This is followed by decision fusion which combines the identification decisions of all soft biometrics to find a particular target in the list of subjects with the closest resemblance in the database. Two main contributions are presented in this thesis. First, techniques for extracting facial shape, height, and body weight are proposed. Second, the thesis evaluates three match score fusion techniques such as SUM, Adaptive Weighted SUM, and Fuzzy Logic to determine the most reliable fusion technique for the soft biometric identification system. The results demonstrate that soft biometric identification system that utilizes fuzzy decision fusion which is based on facial shape, height, and weight is the most optimum with a rank-1 identification rate of 88%.