Impact of pose and glasses on face detection using the red eye effect

In current image-processing algorithms for face detection performance is not completely reliable, especially in situations with variable lighting, and with low-resolution images. One possible approach to implement face detection is the use of the "redeye" effect: the reflection produced by human eyes when exposed to co-axial infrared (IR) light. We investigated the effectiveness of the red-eye technique for variability in: skin tone, eye color, pose, angle of IR illumination, scene illumination, and the effect of shine from glasses. Algorithms were developed to detect eye locations from a single IR image. Image processing steps involved: normalization, blurring, dynamic threshold calculation, and candidate eye position validation. Average eye position estimation accuracy approaches 80 to 85 percent.