Convolutional Neural Network for Ethnicity Classification using Ocular Region in Mobile Environment

Smart Phones are the most widely spread devices over the world. Many people rely on their phones to store sensitive data like financial, personal, or even private photos for family. Thus, the security in mobile environment became one of the essential needs nowadays. In this paper, we presented a new approach to support security on mobile environment using ethnicity biometric. Accordingly, we present six different models that the squeezed models competes to work in mobile environment. These models were designed as a convolutional Neural Network (CNN) to classify five different ethnicities after choosing the desired Region Of Interest (ROI) as extended ocular region from the facial images of the standard dataset FERET. The highest performance CNN model (Model-02) had a classification accuracy of 98.35% with 25, 941 parameters while the squeezed CNN models (Model-05) shows a classification accuracy of 97.35% with just a 8, 117 parameters. The reported results indicated a gain of 68.7% of parameters reduction with 0.52% of loss in accuracy. This will enable the proposed model (Model-05) to work smoothly and efficiently on mobile environment.