Gender Detection on Cephalogram Images using Entropy Equalization Technique and Deep Learning Convolutional Neural Networks

One focus of the field of forensic medicine is to identify corpses. One of the most challenging things in identifying a corpse is when only the skull and neck bones remain, like a fire victim. The method commonly used to identify skulls is quantitative analysis or morphometry which is carried out using measurements, projections and angles. However, this method has a weakness because the discriminant formula used has been developed specifically for limited person patterns. Until now, this formula has not received an update regarding the evolutionary patterns that have occurred over several decades even though interracial mating activities have occurred. This research aims to try to conduct gender detection experiments using cephalogram images in relation to forensic purposes. The method used in this research including image entropy equalization and SMOTE for preprocessing steps. Convolutional Neural Network (CNN) model as the predictive model. The proposed method using entropy equalization from this research resulting in 60% of macro-F1 Score performance and 60% weighted-F1 Score. Hence, it can help identify bodies based on skull remains found at the scene or crime scene more correctly.