ASM/AAM Links


Below are listed links to groups or individuals that I'm aware of doing work on shape/appearance models, and ASM, AAM or related search algorithms.


Mikkel B. Stegmann has a C++ implementation of the AAM and gives examples of its use on medical and meat images.

Extensions to the AAM

Simon Baker and Iain Matthews (and co-workers) at CMU are doing some wonderful work on analysing and improving the AAM update algorithm. They have gone on to develop fast tracking, model building and 3D reconstruction algorithms. Awesome.

Chris Wolstenholme is using wavelet compression to reduce the memory requirements of the appearance model and AAM, with a view to making a 3D AAM of a brain fit into less than a zigabyte.

Ghassan Hamarneh (Chalmers, Sweden) investigated the extension of Active Shape Models to Spatio-Temporal Shapes, and devised a new algorithm for deforming the ST-shape model to better fit the image sequence data.


People/Face Modelling/Tracking/Recognition

Stephen Milborrow has extended the ASM for locating facial features. The site includes code and data.

The Vision Group at Leeds is doing all sorts of interesting things with shape and appearance type models for understanding behaviour.

Genemation Ltd are using appearance models to generate synthetic faces for the computer games industry and for psychological experimentation.

Richard Bowden is working on 3D PDMs to model the body. He has also done work on Non-linear PDM's.

Nick Costen is working on decomposing appearance models of the face into subspaces representing identity, pose, expression, lighting etc.

J�gen Ahlberg is using a modified AAM to match a 3D face model to image sequences.

Dan W. Hansen is using AAMs in an eye tracking tool.


Medical Image Interpretation

The folk at imorphics Ltd are using 3D AAMs in their tools for interpretting 3D medical images.

Milan Sonka and his group have been very active in demonstrating the applicability of ASMs and AAMs to medical image interpretation, using them to solve a variety of problems in different modalities. Steve Mitchell is working on using 2D+time AAMs to segment hearts in MR sequences.

Hans Bosch is using 2D+time models to locate the heart boundaries in echocardiogram sequences. He has done some cool work on applying a histogram normalisation to the intensities to turn the noise distribution into something closer to a gaussian distribution, leading to significant improvements in performance on the nasty noisy echo images.

Quantitiative Analysis of Biomedical Images project uses ASMs to locate and measure structures in 3D images.

Researchers at Image Analysis Section of the Department of Mathematical Modelling, Technical University of Denmark have used both ASMs and AAMs for medical image interpretation.

Angela Caunce used ASMs and a cunning automated training algorithm to locate the major sulci in 3D MR images of the brain.

Carole Twining is using ASMs and AAMs for tracking rats in order to monitor their behaviour in mazes.

Bram van Ginneken has used ASMs for interpretting images of chest radiographs, and has done encouraging work improving the matching performance using more complex classifiers to locate the best positions for the model points.


Automatic Model Building

Automatic model building is all about placing landmarks across sets of images as automatically as possible, so as to generate a training set for an ASM/AAM. It is particularly important for 3D data, where annotating images accurately is hugely time consuming

Kevin Walker worked on the problem, finding matches between distinctive features on each image in the set, finding correspondences on other images and then trying to find a globally consistent correspondence across the whole set.

Alan Brett worked on automatically placing landmarks on sets of surfaces so as to define useful correspondences in 3D data.

Anders Ericsson works on correspondence in 2D and 3D, with applications in medical image interpretation. He also has published test sets and methodologies for assessing 2D automatic landmarking algorithms.

Rhodri Davies and collaborators did some splendid work, developing powerful methods of automatically finding correspondences across sets of 3D surfaces.


Active Blobs

Stan Sclaroff's Group is doing fascinating work with `Active Blobs', which are closely related to AAMs. They are particularly interested in tracking deformable objects.

Shape Statistics

Kanti Mardia, Ian Dryden and colleagues at Leeds are doing excellent work on the statistics of shape.

Fred Bookstein has done ground breaking work on the analysis of shape and the thin-plate spline


Industrial Inspection

Bernard Rolfe and Matthew Doolan have used statistical shape models to examine shape errors in forging.

If you are doing related work, please email me (Tim.Cootes at man.ac.uk)with a brief description of your work and a suitable link so I can include it above.

Tim Cootes

(Sorry about the missing links - I'll add them real soon)

(Thanks to the Danes I met at MIUA'99 for suggesting I create this page)