Disclosed is a method for controlling (constructing) an electronic device. The present control method is provided to automatically extract people and the characteristics (gender, age, etc.) of the people from albums including photos and videos. Two approaches are presented. First, a convolutional neural network (CNN) simultaneously predicts age/gender from all photos and additionally extracts facial expressions suitable for face identification. At this time, Mobilenet is modified and pre-learned to perform face recognition to help to recognize age and gender. Second, extracted faces are grouped by using hierarchical agglomerative clustering (HAS) technology. The age and gender of the people included in each cluster are measured by using predictive calculation for individual photos. The present face clustering quality is very inexpensive, but is comparable to that of a state-of-the-art neural network. Moreover, this approach is characterized by performing more accurate image-based age/gender recognition as compared to models already opened to the public. The method can improve face clustering and age and gender prediction.
展开▼
