Comparison of Three Different CNN Architectures for Age Classification
Murat Aydoğdu, Vakkas Celik, M. Fatih Demirci · 2017
As one of the powerful tools of machine learning, Convolutional Neural Network (CNN) architectures are used tosolve complex problems like image recognition, video analysisand natural language processing. In this paper, three differentCNN architectures for age classification using face images arecompared. The Morph dataset containing over 55k images isused in experiments and success of a 6-layer CNN and 2 variantsof ResNet with different depths are compared. The images in thedataset are divided into 6 different age classes. While 80% of theimages are used in training of the networks, the rest of the 20% isused for testing. The performance of the networks are comparedaccording to two different criteria namely, the ability to makethe estimation pointing the exact age classes of test images andthe ability to make the estimation pointing the exact age classesor at most neighboring classes of the images. According to theperformance results obtained, with 6-layer network, it is possibleto estimate the exact or neighboring classes of the images withless than 5% error. It is shown that for a 6 class age classificationproblem 6-layer network is more successful than the deeperResNet counterparts since 6-layer network is less susceptible tooverfitting for this problem.