Face Detection, Tracking and Alignment

We detect a human face using Viola & Jones detector enhanced by a post-processing step which uses skin-color cue. During face tracking, we provide two options: a SSD (Sum of Squared Difference) tracker or a Mean-Shift tracker. Both trackers work in real-time. Active Shape Model (ASM) is proposed to align a frontal-view face upon request. We illustrate this system as following:

Step 1: Faces are automatically detected in a video sequence captured by a web-camera. “Detector1” is the original Viola & Jones face detector implemented in Open CV library, and “Detector2” is enhanced by adding a skin-color filtering step to reduce false positives.

Viola & Jones face detector

Step 2: Once a face is identified, a SSD tracker (Tracker1) or a Mean-Shift tracker (Tracker 2) is used to track it in the consecutive frames.

SSD tracker

Mean-Shift tracker

Step 3: To align a human face, we need left-click the video and store the current frame to the right window. Active Shape Model (ASM) is used to locate the face components. Currently, our face alignment algorithm only works on front-view face

Active Shape Model

• Qiang Zhu, Kwang-Ting Cheng, Hongjiang Zhang. SSD tracking using dynamic template and log-polar transformation (Oral session). IEEE International Conference on Multimedia and Expo 2004 (ICME 2004), June 27-30, Taiwan.

Demos