Eye detection in a face image using linear and nonlinear filters

Two methods of eye detection in a face image are described: the face is first detected as a large flesh-colored region, and anthropometric data are then used to estimate the size and separation of the eyes.

Abstract

This paper describes two methods of eye detection in a face image. The face is first detected as a large flesh-colored region, and anthropometric data are then used to estimate the size and separation of the eyes. When a linear filtering method, using filters based on Gabor wavelets, was then applied to detect the eyes in the gray-level image of the face, the detection rate was good (80% on one dataset, 95% on another), but there were many false alarms. A nonlinear filtering method was therefore developed to detect the corners of the eyes in the color image of the face. This method gave a 90% detection rate with no false alarms.