Applying CNN to Classify Gender for Identity Verification System

Computers have difficulty recognizing and distinguishing between faces, especially when it comes to gender and age. Gender is important for emotional recognition and security reasons and can help identify individuals who look alike. This research proposes a binary gender recognition system that consists of two stages: face detection and gender classification. The face detection component is compared to three models: Python face recognition library, SSD MobileNet, and EfficientNet. The gender classification component uses two models: custom CNN and VGG Face models. Based on this research, the final prototype is accessible via a web application and is capable of performing image-based face detection and processing each known face to determine its gender (male or female). The face identification model yields varying results, with the EfficientNet model outperforming the others after the model is deployed. The VGG Face model also performs well, with an average accuracy of 99% for faces taken under optimal settings (training and testing datasets) as well as for specific images collected outside of the dataset. However, for masked faces and captures at low brightness levels, it still results in less accurate detection and classification for several types of images.