FRCNN-GNB: Cascade Faster R-CNN With Gabor Filters and Naïve Bayes for Enhanced Eye Detection

Research into biometric identification technologies has evolved in recent years, as most secure facilities and applications are now based on digital technology. Among the available biometric identification technologies is eye detection. The relevance and impact of the use of eye detection in a variety of biometric authentication systems are very high. The main problems associated with the accuracy of eye detection methods are occlusion or reflections from glass. In view of this, we propose a hybridized and enhanced eye detection method that uses a faster region-based convolutional neural network with Gabor filters and naive Bayes (FRCNN-GNB) model to address the problems associated with eye detection schemes. The proposed method consists of four components: convolution layers, a region proposal network, a detection network, and a decision model. The enhancement method is based on a cascade Faster R-CNN with Gabor filters and the naive Bayes model, in which the initial bounding boxes of the eye region are detected using Faster R-CNN and the decision step is carried out using Gabor filters and the naive Bayes model to determine which of the bounding boxes belong to the eye region. Experiments on the proposed FRCNN-GNB eye detection scheme are performed on the CASIA-IrisV4 database, and show that the accuracy in terms of eye detection is 100%. The results of the study demonstrate the efficiency of the proposed solution.