On age prediction from facial images in presence of facial expressions

Predicting age and gender from facial images is a fundamental research problem that has many applications in major research areas. The state-of-the-art online APIs can predict age and gender but their accuracy degrades when emotions are present. In this paper, we present feature-extraction based machine learning models that can predict ages with acceptable accuracy in presence of facial-expressions. After identifying 68 facial landmarks, different distances and ratios (that changes with age and expressions) are selected to predict the age that can overcome the impact of emotions with reasonable accuracy. The experimental results show that while neutralising the effect of emotion, the proposed models can perform better on female images compared to the male image set. And images with disgust and contempt expressions deviate most during prediction. In contrast, predicted age is more accurate for angry expressions. Also for different ethnic groups, the predicted age deviates differently from the actual age.