Networks Architecture Selection

The convolutional neural network (CNN) was selected as a network model because it produces state-of-the-art recognition results according to the literature. The main merit of CNNs is the fact that they combine the two major blocks of feature extraction and classification into a unified learner, as shown in Figure 8.

**Figure 8.** Block diagram of convolutional neural network (CNN).

In fact, this advantage is very convenient to this application for two reasons: First, CNNs learn directly on raw data (i.e., pixel intensities of the eye images), and thus eliminate the need for manual feature extraction. More importantly, the automation of feature extraction by means of training goes hand in hand with the philosophy of user-specificity because features that best characterize the eye pupil are learned for each individual user. By contrast, the previously discussed template-matching approach is not truly user-specific, as it operates on features that are fixed and handcrafted for all users, which does not necessarily yield the optimal representation of the input image.

Architecture wise, CNNs are simply feedforward artificial neural networks (ANNs) with two constraints:

