David Cristinacce - Automatic Face Editing MSc Project

Automatic Face Editing MSc Project

A proposed six month MSc research project, suitable for Manchester University MSc students completing in September 2006 or September 2007. See the downloadable demo to get a quick idea of what this project is about.

Overview

The aim of the MSc project is to develop methods of automatically editing human face images. For example, the following face has been edited by fitting a model to the face, then changing the model parameters to change the expression from neutral to smiling.

Original Image	Model Fitted to Image	New Synthetic Smile Image

A downloadable demo of this method is now available.

Aims

• Allow the user to interactively alter the parameters of the face (e.g. expression, identity, sex, age etc)
• Improve our current methods for predicting expression changes, without altering identity
• Possibly evaluate methods for automatically classifying faces by expression, sex, age etc

Approach

Statistical methods of modelling the visual appearance of objects have been developed at Manchester University over the last 15 years. The main approach is the Appearance Model method, which learns the shape and texture variation of an object class from a set of labelled training examples[1].

Given a statistical model of an object (in this case the human face) it is necessary to match the model to unseen examples of the object. This is achieved using the Active Appearance Model algorithm, which aims to minimise the difference between the Appearance Model texture and the target image[2]. Given the best match of the Appearance Model to a target image, the parameters of the model can be varied to make small (but realistic) changes to the appearance of the object.

The project will use automatic global and local detection methods to fit appearance models to face images[3], then constrain the parameter changes that a human operator can make, to produce realistic expression changes. Therefore the project will combine combine computer vision and computer graphic techniques.

Examples of appearance models built from human faces and more information on Appearance Modelling and Active Appearance Modelling can be found at Tim Cootes website. The pages describing Appearance Modelling are here.

References

1. A.Lanitis, C.J. Taylor, T.F. Cootes "Automatic interpretation and coding of face images using flexible models", in Pattern Analysis and Machine Intelligence vol 19, issue 7, pp. 743-756 1997. PDF
2. T.F.Cootes, G.J. Edwards and C.J.Taylor "Active Appearance Models", in Proceedings of European Conference on Computer Vision 1998 pp. 484-498. PDF
3. D.Cristinacce, T.F.Cootes and I.Scott "A Multi-Stage Approach to Facial Feature Detection", in Proceedings of British Machine Vision Conference 2004 pp. 277-286. PDF

Project Timetable

1. Literature review and familiarisation with technical background and available code resources.
2. Collection of data + building of new models using current methods.
3. Extension of current methods.
4. Write-up dissertation.

Requirements

• Enthusiasm for research
• Coding skills in c++ a big advantage, but possible to work with Matlab too
• Mathematics especially linear algebra (not vital, but useful)
• Computer graphics experience (not vital, but useful)
• Interest in human faces e.g. from a psychology perspective

Supervisor

Dr David Cristinacce (webpage)
Imaging Science and Biomedical Engineering (ISBE)
Faculty of Medical and Human Sciences (MHS)
Email: david.cristinacce at manchester.ac.uk