Human Face Emotions Recognition from Thermal Images Using DenseNet

In the current scenario face identification and recognition is an important technique in surveillance. The a necessary biometric humans. Therefore detection plays major job computer vision applications. Several emotions classification approaches have been presented throughout last few decades of research to improve rate for thermal pictures. However, real-time, lighting conditions might change due several factors, such as different times capture, weather, etc. Due variations intensity, performance facial expression system not good. This paper proposed model human classification. Four main steps were involved this research. Initially, Difference Gaussian (DOG) filter utilized crop input images then normalize using median pre-processing step. Then, Efficient Net used extracting features shape, location, occurrences from images. After that, detect faces by YOLOv4 better Finally, classify on DenseNet into seven happy, sad, disgust, surprise, anger, fear, neutral. method outperforms state-of-art techniques pictures, classifies expressions, according experimentations RGB-D-T database. accuracy, precision, recall, f1-score metrics will be with database assess efficacy methodology. models achieve high accuracy 95.97% Furthermore, outcomes show good precision various tasks.