Real-Time Age Estimation System Based on Convolutional Neural Network

Inhalt:
The demand for automatic face detection and age recognition has become increasingly significant for the social media industry and society. Most of the previous age estimation models are based on the presumption that the input images for their models are facial images of humans. In the real-world scenario, however, the input might not be qualified enough. Moreover, their estimation performance is yet to be improved. This paper designs a novel system for age estimation based on a convolutional neural network (CNN). First, a Caffe model is used to filter out the input which does not contain a face. Then, a CNN-based model is proposed to estimate ages from the faces detected. The performance of the proposed model is evaluated on the Adience benchmark dataset, surpassing a few previous well-known models. Finally, the model is deployed to the cloud to provide real-time services.