Fig. 12From: Facial expression recognition using optimized active regionsThe structure of proposed CNN. There are three convolution layers, three sub-sampling layers and two fully connected layers in itBack to article page