DeepFaces: Advancing Age and Gender Classification using Facial Images with Deep Features
Journal Title: International Journal of Innovations in Science and Technology - Year 2024, Vol 6, Issue 2
Abstract
In the realm of identity recognition and social interactions, human facial features play a pivotal role. Accurate age estimation and gender classification from facial images have practical implications across various fields, including biometrics, surveillance, and personalized services. This study presents a novel approach that harnesses deep features extracted by the VGG-19 architecture for age and gender prediction, employing a custom convolutional neural network (CNN) for classification. Leveraging the UTKFace dataset, encompassing a diverse collection of facial images with annotated age and gender labels spanning various ages, ethnicities, and gender representations, provides a robust foundation for model training and evaluation. Deep features extracted from the VGG-19 architecture serve as rich representations of facial patterns, enabling our model to discern discriminative cues for age and gender. These deep features are input to CNN model, which is fine-tuned specifically for age and gender classification. The model comprises input layer, Dense layers, incorporating dropout and batch normalization to mitigate overfitting, and Activation Functions Sigmoid for gender classification and SoftMax for Age group classification. The dataset is divided into training and validation sets (70% and 30%, respectively), enabling the model to learn to map VGG-19 features to age and gender labels. To evaluate the performance of the model, metrics like accuracy, precision, recall, and F1-score are employed. The proposed model achieves an impressive 78.67% accuracy in predicting age and 97.02% accuracy in gender classification on the UTKFace dataset, outperforming traditional methods despite challenges posed by variations in lighting, pose, and expression. The robustness of our approach is evidenced by its capability to handle diverse gender representations.
Authors and Affiliations
Ammad Noor, Wakeel Ahmad, Syed M. Adnan Shah
Keywords
Related Articles
- EP ID EP760344
- DOI -
- Views 69
- Downloads 0
Enhanced Skin Cancer Classification with MobileNetV3 and Morphological Preprocessing: A Deep Learning-Based Extension
Skin cancer detection continues to pose challenges due to the visual similarity between the types of lesions and the limitations of traditional diagnostic methods. This study presents an extended and...
Challenges and Practices Identification via Systematic Literature Review in the Design of Green/Energy-Efficient Embedded Real-Time Systems
As most embedded devices are portable, that is they are operated by batteries, early battery exhaustion is likely to cause the failure of the embedded real-time systems (ERTS). Therefore, developers and users enjoy the...
Assessment of ML Classifiers in Complex Human Activity Recognition Using Wearable Sensors Data
Human Activity Recognition (HAR) is essential for understanding daily behavior patterns, and wearable sensor data serves as a reliable source for monitoring complex activities. This study uniquely evaluates the perform...
Optimizing UAV Wing Performance: A Computational Analysis with Computer-Based Algorithms for Composite Material Integration
Introduction/Importance of Study: The aircraft wing, a vital component, demands intricate design to balance lift generation, drag reduction, and weight minimization. In advanced UAVs (Unmanned Aerial Vehicles), priorit...
Catalytic PerformanceofElectro-Oxidative Natural Manganese Sand forAmmonium Nitrogen Removal
The environmental risks associated with ammonium nitrogen (NH₄⁺-N) pollution have led to a growing focus on prevention. Electrochemical advanced oxidation is an effective and ecofriendly method that only requires electr...