120 likes | 132 Views
Efficient Deformable Template Matching for Face Tracking. Jue Wang and Jiun-Hung Chen CSE/EE 577 Spring 2004. Goal. Original frame. Five face components are precisely located. Applications: face animation, expression analysis, pre-processing for face recognition……. Related Works.
E N D
Efficient Deformable Template Matching for Face Tracking Jue Wang and Jiun-Hung Chen CSE/EE 577 Spring 2004
Goal Original frame Five face components are precisely located Applications: face animation, expression analysis, pre-processing for face recognition……
Related Works Active Shape Model Shuicheng Yan, Ce Liu, Stan Z. Li, HongJiang Zhang, Heung-Yeung Shum, Qiansheng Cheng: Face alignment using texture-constrained active shape models. Image Vision Comput. 21(1): 69-75 (2003) • Cons: • High accuracy • Pros: • Runs slow • Face detection needed for initialization of the template I have a video of 1,0000 frames, can I do it faster?
Our System EM matching algorithm Distance transform Very fast but errorous component detection
Face Component Detection • Haar features for rapid component detection Paul Viola and Michael Jones. Robust Real-time Object Detection. ICCV2001. Positive samples of left eye Negative samples of left eye
Detection Results • Fast detection, but…
Template Design Translation: Spring cost for deformation: Why this form? -Fast computing using distance transform is available!
EM Algorithm Update other parts Distance transform Update central part Finding peaks Update other parts
Examples EM algorithm Iteration 1 Iteration 3 Iteration 2 Iteration 4 Iteration 5
Conclusion • A fast and efficient deformable template matching algorithm for face tracking • Not real-time, 2 seconds per frame • Still too slow for you? • Template is simple, may not be easy to incorporate statistical learning things