Age category estimation using matching convolutional neural network

This paper proposes an age category estimation method using matching convolutional neural network (CNN). The proposed method compares an input facial image with target facial images using matching CNN and classifies the input facial image into one of three cases: younger than, similar to, or older than target images. Using a pair of the input facial image and the facial image in each age category, the proposed method dramatically improves the accuracy of the age category estimation. Experimental results show that the proposed method archives 94.5% accuracy of the age category estimation on Adience benchmark with 16 target images per age category whereas most existing methods achieve about 60% accuracy.