Automatic Facial Point Detection

Investigators: Dr. Eun-Jung Holden, Professor Robyn Owens

Last updated: 21 November 2001.

[This is an ARC supported project.]

1. Background

Our 3D head tracker and lip contour tracker require automatic location of facial points from first image of the sequence prior to tracking. We have adapted USC face recognition technique where gabor wavelet responses of the facial point locations are used as features for automatic detection.

2. Facial Points to be detected

There are seven facial points to be detected: two outer eye corners, one nostril, left, right, top and bottom of the lips. Each point is represented as 24 complexed values that are Log Gabor wavelet reponses. Then using the model points extracted from the training images are used to find similar points on the testing image. Once likely locations of each facial point are detected, the system uses facial geometry to solve correspondence problem. The testing image and the facial geometry are shown below.

The detection result is shown below.

Home | Department | UWA