Detection of Facial Landmarks from Neutral, Happy, and Disgust Facial Images


Automated analysis and recognition of facial expressions of emotions has been recently studied to improve quality of human-computer interaction. In this framework, expression-invariant detection and segmentation of a face are crucial steps for any vision-based interaction scheme. In present study, a new method for detecting facial landmarks from neutral and expressive facial images was designed and described. The candidates for landmarks were formed from edge regions with characteristic pattern of oriented edges extracted from image at several levels of resolution. The impact of expressions on the developed method was tested using dataset including neutral, happy, and disgust images. The results demonstrated a high accuracy in detecting landmarks from neutral images. However, expressions of happiness and disgust had a deteriorating effect on the landmark detection. In general, results showed that expressions affect computer segmentation of a face and are needed to be taken into account while designing fully automated systems of face processing.