Brian Amberg @ Uni-Basel

Contact

Address:	Brian Amberg University of Basel Computer Science Department Bernoullistrasse 16 CH - 4056 Basel, Switzerland
eMail:	brian.amberg@unibas.ch
Phone:	++41 61 267 0552
Fax:	++41 61 267 0559

Publications

Reviewed Publications

• Weight, Sex, and Facial Expressions: On the Manipulation of Attributes in Generative 3D Face Models    [The slides ]  [The video ]
  Brian Amberg, Pascal Paysan and Thomas Vetter
  IN: ISVC'09, 5th International Symposium on Visual Computing, Las Vegas, USA, November 2009
• Face Reconstruction from Skull Shapes and Physical Attributes 
  Pascal Paysan, Marcel Lüthi, Thomas Albrecht, Anita Lerch, Brian Amberg, Francesco Santini, and Thomas Vetter
  IN: Proceedings DAGM'09: 31st Pattern Recognition Symposium, Jena, Germany, September 2009 (DAGM Best Paper Award)
• A 3D Face Model for Pose and Illumination Invariant Face Recognition 
  Pascal Paysan, Reinhard Knothe, Brian Amberg, Sami Romdhani and Thomas Vetter
  IN: AVSS'09, Proceedings of the 6th IEEE International Conference on Advanced Video and Signal based Surveillance for Security, Safety and Monitoring in Smart Environments, Genova, Italy, September 2009
• On Compositional Image Alignment with an Application to Active Appearance Models 
  Brian Amberg, Andrew Blake and Thomas Vetter
  IN: CVPR'09, IEEE International Conference on Computer Vision and Pattern Recognition, Miami, USA, June 2009
• Expression Invariant Face Recognition with a Morphable Model 
  Brian Amberg, Reinhard Knothe and Thomas Vetter
  IN: FG'08, IEEE International Conference on Automatic Face and Gesture Recognition, Amsterdam, The Netherlands, Sep. 2008
• Reconstructing High Quality Face-Surfaces using Model Based Stereo 
  Brian Amberg, Sami Romdhani, Andrew Fitzgibbon, Andrew Blake, Thomas Vetter
  IN: ICCV 2007, Rio de Janeiro, Brasil, Oct. 2007
• Optimal Step Nonrigid ICP Algorithms for Surface Registration 
  Brian Amberg, Sami Romdhani, Thomas Vetter
  IN: CVPR 2007, Minneapolis, USA, June 2007

Contest Entries

• SHREC'08 Entry: Shape Based Face Recognition with a Morphable Model 
  Brian Amberg, Reinhard Knothe and Thomas Vetter
  IN: SMI'08, Shape Modeling International, NY, USA, June 2008

Talks

Invited External Talks

• On Compositional Image Alignment  Video 1  Video 2  Video 3  
  Brian Amberg, Andrew Blake, Thomas Vetter
  IN: UCL Vision Seminar 2009, London, UK, 22 July 2009
  and also: MSRC, Cambridge, UK, August 2009
• Inverse Methods for Image Understanding [PPTX] [PPSX] 
  Brian Amberg
  IN: Microsoft Research Summer School 2009, Cambridge, UK, 30 June 2009
• Expression Invariant Face Recognition with a Morphable Model
  Brian Amberg
  IN: ICVSS, International Computer Vision Summer School, Sicily 17 July 2008
  I got the Best Student Award there.

Other Talks

I should put up some of my older talks

Current Projects

Registration

I am currently developing 3D-3D registration methods based on mesh representations of the data. These are a family of nonrigid ICP algorithms, which differ in the regularisation employed (deformation gradient, deformation laplacian, statistic based). For many regularisations the ICP step is convex, leading to a natural iterative algorithm.

Stereo Fitting

I have implemented a stereo fitting system, which is quite invariant against light changes. This is a straightforward application of the morphable model.

AAM Fitting

I am currently working on efficient 2D and 3D AAM fitting methods in the monocular case. The results got their own website on compositional fitting.

3D-3D Fitting

A simple 3D-3D fitting method shows how important good data modelling is. Following are some applications of 3D-MM based on 3D data.

Face Recognition

The advent of cheap face scanning systems makes biometrie with these sensors feasible. I find it very interesting that it is possible to achieve extremely high recognition rates with shape-only methods, even though humans are able to recognize people even from shape-reduced comic style depictions. Practically, fitting the morphable model to 3D data is more robust and faster than fitting to images, which makes these methods suitable for large-scale applications.

Expression Invariant Face Recognition

In this project we investigate expression-identity seperating models. It turns out, that even with a simple linear model we achieve good expression invariant face recognition rates.

My Slides on Expression Invariant Face Recognition with a Morphable Model  are here.

Software

Latex Poster Template

A comfortable latex poster template makes it easier to go from paper to poster, as elements of the paper can be directly reused. No more need for these ugly "I printed my powerpoint presentation on 24 pages and pinned it to the wall" posters.

Matlab fast image interpolation methods

Rotating a mandrill 36 times by 10 degrees using nearest neighbour, biliniear and bicubic interpolation
Nearest Neighbour	Bi-Linear	Bi-Cubic

Matlabs interp2 method is very slow when interpolating images, especially multi-plane images. Here is an efficient mex implementation for nearest neighbour, bi-linear, and bi-cubic interpolation.

3D Volume Interpolation with ba_interp3

ba_interp3 is a fast interpolation routine for 3D volumes. It can be used as a drop in replacement for matlabs interp3 function. It supports nearest neighbour, tri-linear, and tri-cubic interpolation.

Restrictions: The only restriction is, that the data has to be on a regularly spaced grid (i.e. coming from meshgrid.

Enhancements: The first advantage of ba_interp3 is that it is a lot faster:

Additionally, it can interpolate layered data. If you interpolate a matrix of size MxNxO x L1xL2x...xLk then each of the L1*L2*...*Lk layers is interpolated simultaneously. Think of this as in RGB images, where there would be a single additional dimension of size three.
An volume of 3x3 Matrices (i.e. the volume derivative) would have two additional dimensions of size three.

Download

The code is available at ba_interp3 Interpolation Code

Chord charts

A pet project of mine is this huge database of automatically generated and rated chord-charts for many stringed instruments in different tunings.

Teaching

A Ruby Course

A Ruby Course for people who know a bit of programming. This course was given at the sommercampus 2005 of the university of freiburg. It was a 4 * 4 hours course, of which the slides cover approximately 3 * 4 + 1 hours. For the last day I propose to use the ruby quizzes. The slides are published under the GFDL. Feel free to use the slides under these conditions, but I'd like if you'd drop me a note such that I see if they have been useful.