Abstract:
Estimating age and gender simultaneously from a single facial image is important for many
applications. Although previous deep learning approaches have achieved strong performance
in gender classification, their performance in age estimation remains limited, as they often fail
to capture the shape and structural features crucial for accurate prediction. To overcome
this limitation, this study proposes a dual-branch CNN model for joint age and gender
estimation. The proposed approach takes a raw facial image and generates its corresponding
edge map using the Canny edge detection algorithm. Features from the raw image are
extracted using a pre-trained VGG-16 network, while those from the edge map are obtained
through a customised CNN. The inclusion of edge-based features captures fine structural
details, whereas channel and spatial attention modules enhance discriminative learning across
both branches. Finally, the features from both branches are concatenated and processed
through a series of fully connected layers for classification. In this work, age estimation is
treated as a multi-class classification problem rather than regression, enabling group-wise
age prediction. The proposed approach was trained and evaluated on the publicly available
UTKFace dataset, attaining classification accuracies of 87.88% for gender and 60.00% for
age prediction. Compared to state-of-the-art methods, our approach achieves competitive
accuracy while using less than half the training samples, underscoring its efficiency and novelty.
