DEEP LEARNING FOR AGE GROUP CLASSIFICATION SYSTEM

Abstract

Age Group Classification (AGC) is a difficult task due to variations in human genders, expression, races, poses and so on. The age group analysis is used in multimedia forensic investigation for crime scenes. In this study, an efficient method for AGC is presented. AGC system uses mainly two stage; preprocessing and classification. The preprocessing stage consists of face region detection, gamma correction, Difference Of Gaussian (DOG) filter and normalization. Then the preprocessed images are given to Visual Geometric Group (VGG) 16. The convolution and max-pooling layers in VGG 16 architecture is used to resize the image. The REctified Linear Unit (RELU) is used as an activation function in each convolution and max-pooling layer. The sigmoid layer is used for the AGC into adolescence, adult and senior adult. The system uses MORPH database for the performance evaluation. AGC system produces the classification accuracy of over 90% for all age groups using VGG16 architecture.

Keywords

• age group classification, neural network, vgg16, morph database