Face parts extraction windows based on bilateral symmetry of gradient direction

Abstract

We propose a simple algorithm to determine face parts extraction window in face image. We utilize bilateral symmetries between and within face parts. We also use knowledge about size and locationship of face parts. First, we examine bilateral symmetries around vertical orientation edge, then obtain symmetry measures. The symmetry measures are projected onto y-axis to produce histogram of the measures. We estimate height of face parts regions by frequency of the histogram. Face parts region, which contains maximum frequency of the histogram, becomes a candidate of face parts region that includes eyes and eyebrows. Secondly, the measures that exist within the height of the face parts region are projected onto x-axis to estimate width of face parts region. We determine face parts extraction windows by the estimated height and width. Finally, we detect irises in the candidate of face parts region that includes eyes and eyebrows, using circular mask.