Eye Iris Classification Using the Alex Net Model with Absolute Activation Function

Abstract

Eye iris classification and recognition are important in pattern recognition applications. Traditional image segmentation techniques and classification using machine learning techniques still have not achieved the best results. Different image-capturing methods, low-resolution photos, images with a high noise ratio, and other factors contribute to the extraction of low-quality features. This study proposed the Alex net model with absolute rectified linear unit (absRelu) and rectified linear unit (Relu) activation functions. First, the images are read and resized, then passed to convolutional neural networks (CNN) and absRelu or Relu layers to extract the best features. The next step is size reduction using Maxpooling techniques. The output is passed to the dense layers and then to the Softmax classifier. The proposed model was evaluated using three datasets: MMU, AMF, and IITD. Data augmentation was applied to increase the number of samples. For the MMU dataset, the proposed model with absReLu and ReLu yields 0.9866% and 0.9778% in sequence. For the AMF dataset, the proposed model with absRelu and Relu yields 1.0000% and 0.9981% in sequence. With the ITTD dataset, the best results for both absRelu and Relu are 0.9958% and 1.00%, respectively.