Gender Classification from Human Face Images Using Deep Learning Based on MobileNetV2 Architecture

Abstract

A person’s face provides information about many identifying characteristics, including age, gender, and race. Among these characteristics, gender prediction has drawn much attention due to its many applications and use cases. Using a face to determine an individual’s gender and assign it to the appropriate category of “male” or “female” is called gender recognition. Typically, there are multiple steps in the process, such as face detection and feature extraction to record the distinctive features of the face. Efficient feature extraction for gender classification is possible through a number of approaches, including deep learning-based convolutional neural networks (CNNs), which have shown excellent performance in learning hierarchical representations directly from raw pixel data. Previous attempts at gender recognition have focused on several static physical features, such as fingernails, body shape, hand shape, eyebrow, face, etc. In this study, we propose a gender classification using deep learning from human face images. This study uses a deep learning model based on the MobileNetV2 architecture for gender classification. The model was pre-trained and fine-tuned on the dataset using transfer learning techniques. During training, the model was optimized using Adam optimizer with focus loss function and learning rate scheduler. The dataset name is the largest gender-specific face recognition dataset from kaggle with sample images (man and woman). Experimental results show perfect performance and F1score value of 96%.This means that the model achieves excellent performance in the balance between precision and recall, and it is close to perfect performance on many classification tasks.