A Robust Eyeballs Feature Extraction Algorithm Using Improved Generalized Symmetry Transform

摘要

Eyes or eyeballs extraction from faces is important in facial feature extraction that is crucial in human-computer interaction and other related applications. According to the eyeballs circle symmetry characteristic in face image, original generalized symmetry transform is modified. A robust algorithm proposed in this paper extracts eyeballs from single face images in three steps: (1) using snake and a Gaussian preattentive filter to generate a facial organ mask, (2) using symmetry characteristic to find some candidate points of eyeballs, (3) validating these candidate points by the eye position information. This approach is tested on the ORL database, in which eye feature is not easy to extract for the reason of glasses or closed eyes, with the exact extraction rate 88.5%. It is shown in our experiments that this approach is robust to noise and some interference from glasses.