Using statistical moments as invariants for eye detection

In this paper we address the problem of eye detection in greyscale images. We represent face image using topographic labels to alleviate detection under severe lighting condition. The regions in topographic image are then described using regional invariant moments. The employed moments are invariant to similarity transform. This enables the proposed eye detection method to work under head movement. In detection phase we first provide a candidate list of points with pit label in topographic image. Image at neighbourhood of each pair of pit points are compared with eyes model using their corresponding feature vectors. Using a Bayesian classifier we detect the pair of points with the descriptors most similarity to the eyes. The result of experiments confirms the capability of proposed method for detecting eyes in face images.