<![CDATA[Image-based facial age estimation is becoming an important component in biometrics and digital dermatology, but many deep learning approaches still rely on global facial features, making them less sensitive to micro changes on the skin surface, particularly on Asian faces which have distinct ageing patterns. This research offers a novel contribution by integrating SE-ResNeXt50 with skin texture analysis to produce more accurate and interpretable age estimations. The dataset used is APPA-REAL, which consists of 7,612 age-labeled Asian face images. After face detection, skin area cropping, size standardisation, and intensity normalisation, visual features were extracted using SE-ResNeXt50, which utilises a channel attention mechanism through Squeeze-and-Excitation blocks to amplify subtle ageing signals. In parallel, this study adds skin texture analysis based on quantitative indicators, namely wrinkle index, tone unevenness, shine proxy, and brightness, to represent the skin microstructure correlated with ageing. The performance of the method is evaluated using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The results show that the combination of an attention-based deep network and skin texture indicators can improve the consistency of age prediction and provide a clearer basis for interpreting changes in skin texture on Asian faces. This finding strengthens the potential for developing an age estimation system that is not only precise but also relevant for digital skin monitoring applications and ageing evaluation.]]>
Copyrights © 2026
