An Efficient Support Vector Machine with Radial Basis Function for Passive Continuous Authentication System
K. Juliana Gnana Selvi, Shyamala Susan · CiiT international journal of fuzzy systems · 2018
Advancement in communication technology and computer applications in our day-to-day life highlights the strong need for user-friendly systems to secure our assets and protect our privacy without losing our identity. Generally, use a password to access a computer, Personal Identification Number (PIN) to access ATM, cryptographic techniques to access and view files and dozen others to access the internet and so on. These conventional methods of identification can be easily guessed, observed or forgotten. Hence the alternative option like biometric-based recognition for reliable and robust human identification has given more importance in the wide range of commercial sectors. In existing system, face recognition is performed through hybrid swarm intelligence approach which uses both PSO and ABC. Hybrid ABC optimization algorithm was used to train a weighting mask for assisting the face recognition process. In this system training process can be finished offline, because of this it achieves less computation time. However it does not produced satisfactory recognition accuracy. In order to improve the recognition accuracy the proposed system introduced a Radial Basis Function-Support Vector Machine (RBF-SVM) based biometric authentication system. In this proposed system preprocessing is performed by using median filter. Then the skin regions are detected from preprocessed images. And face detection is performed by using boosted classifiers with Haar-like features. The same boosted classifiers are also used for eye detection. In this eye detection the left eye and right eye features extracted to locate multifaces in the result. To ensure the high security eye pupil is detected by using circle Hough transforms. The extracted features are given to RBF-SVM. It classifies the input images as matched or non-matched. The experimental results show that the proposed system achieves better performance compared with existing system in terms of accuracy, precision and recall.