Bill Triggs

Home | LEAR team | Research | Grants | Collaborators | Publications | Software | Events | Student Projects | Contact Address

Home > Student Projects > Pose-invariant Face Detection

MSc Project, 2006-2007

** This project has been taken, it is no longer available **

Vector Boosting for Pose-invariant Face Detection

Supervisor

Bill Triggs (Bill.Triggs@inrialpes.fr)

Summary

Keywords : face detection, image analysis, pattern recognition

Human faces are one of the most important classes of image and video content and it is often necessary to detect them as a first stage of a processing chain for recognition, expression analysis, person counting, etc. Modern face detectors are usually based on statistical learning methods. Computer vision techniques are used to provide a large set of candidate image features, some of which are discriminant for faces, and a machine learning method selects the most useful features and builds a face / non-face classifier based on them. Current detectors are quite good at finding upright well-lit faces seen from the front, but significantly less good at tilted and profile faces under difficult lighting. Also, running a full classsifier at all points of an image is computationally expensive. Usually almost all of the image regions tested are not faces, so algorithms based on "rejection chains" or "rejection trees" have become popular : the early stages of the chain contain classifiers designed to quickly reject the most obvious non-faces with a minimum of computation, and as the chain or tree is descended these "rejectors" progressively become stricter but more expensive, until finally the few remaining regions can be classified as faces.

The project will provide hands-on experience of building a sophisticated modern face detector. It will be based mainly on the ICCV 2005 conference paper Vector Boosting for Rotation Invariant Multi-View Face Detection by C. Huang, H. Ai, Y. Li and S. Lao. This is a rejection tree method that is designed to handle a wide range of head poses, and that gives very good results in practice.

This is a challenging project. The ideal candidate would be a student who programs well and who wishes to go on to do a practically-oriented PhD in visual recognition. The project will begin by building a conventional rejection chain detector as this is a useful intermediate stage for the full tree-based detector. Possible continuations after the MSc would include incorporating face recognition or expression analysis in the detector.

The reference list below gives just some of the recent work on cascade-based detectors. The most important references for this project are [1] and especially [8-9], and [5-7] for details of optimization methods.

References on Cascade Detectors

  1. P. Viola and M. Jones. Robust real-time face detection. International Journal of Computer Vision (IJCV) 57(2) 137-154, 2004. PDF. (Originally appeared in CVPR'01).
  2. P. Viola, M. Jones and D. Snow. Detecting Pedestrians using Patterns of Motion and Appearance. International Journal of Computer Vision (IJCV) 63(2) 153-161, 2005. PDF. (Jointly won best paper prize at ICCV'03).
  3. Qiang Zhu, Mei-Chen Yeh, Kwang-Ting Cheng and Shai Avidan. Fast Human Detection Using a Cascade of Histograms of Oriented Gradients. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2006, pages 1491-1498. PDF. (A Viola-Jones style human detector using a descriptor set originally developed in our group).
  4. P. Viola and M. Jones. Fast and Robust Classification using Asymmetric AdaBoost and a Detector Cascade. Conference on Neural Information Processing Systems (NIPS) 2002. PDF. (Method for optimizing Viola-Jones detection cascades).
  5. Jie Sun, James M. Rehg and Aaron Bobick. Automatic Cascade Training with Perturbation Bias. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2004. PDF. (More on optimizing Viola-Jones detection cascades).
  6. Jianxin Wu, Matthew D. Mullin and James M. Rehg. Linear Asymmetric Classifier for Cascade Detectors. International Conference on Machine Learning (ICML) 2005. PDF. (More on optimizing Viola-Jones detection cascades).
  7. S. Charles Brubaker, Matthew D. Mullin and James M. Rehg. Towards Optimal Training of Cascaded Detectors. European Conference on Computer Vision (ECCV) 2006. PDF. (Yet more on optimizing Viola-Jones detection cascades).
  8. Stan Z. Li, Senior Member, IEEE, and ZhenQiu Zhang. FloatBoost Learning and Statistical Face Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI) 26(9), 2004. PDF. (An alternative cascade learning procedure, here used to learn a tree of face detectors not a linear chain).
  9. Chang Huang, Haizhou Ai, Yuan Li and Shihong Lao. Vector Boosting for Rotation Invariant Multi-View Face Detection. International Conference on Computer Vision (ICCV) 2005. PDF. (More on tree structured face detection cascades).
  10. Simon Baker and Shree K. Nayar. Pattern Rejection. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 1996. PDF. (One of the earliest papers on rejection based classifiers).
  11. François Fleuret and Donald Geman. Coarse-to-fine Face Detection. International Journal of Computer Vision (IJCV) 41: 85-107, 2001. PDF. (An early paper on rejection cascade based face detection).