A Study on Image Based Gender Classification Using Convolutional Neural Network

Frontal face image based gender detection is an important module for many computer vision applications. In this paper, we presented a convolutional neural network (CNN) based gender classification model. The faces in image were first detected and located based on the histogram of oriented gradient (HOG) features. Then the 68 facial landmarks were extracted using the ensemble of regression tree (ERT) method. Based on the landmarks, the face was aligned and scaled to fixed size. The aligned and scaled face image was feed to the RestNet. Finally, the fully connected layers were built to classify the gender. The area under curve (AUC) on testing dataset achieved 0.95. In addition, the misidentified cases were analyzed to further understand the classification model.