A Comparative Evaluation of Active Appearance Model Algorithms

An Active Appearance Model (AAM) allows complex models of shape and appearance to be matched to new images rapidly. An AAM contains a statistical model of the shape and grey-level appearance of an object of interest The associated search algorithm exploits

A Comparative Evaluation of Active

Appearance Model Algorithms

T.F.Cootes,G.Edwards and C.J.Taylor

Dept.Medical Biophysics,

Manchester University,UK

tcootes@server1.smb.man.ac.uk

Abstract

An Active Appearance Model(AAM)allows complex models of shape and

appearance to be matched to new images rapidly.An AAM contains a sta-

tistical model of the shape and grey-level appearance of an object of interest

The associated search algorithm exploits the locally linear relationship be-

tween model parameter displacements and the residual errors between model

instance and image.This relationship can be learnt during a training phase.

To match to an image we measure the current residuals and use the model

to predict changes to the current parameters.The algorithm converges in a

few iterations.In this paper we describe variations of the basic algorithm

aimed at improving the speed and robustness of search.These include sub-

sampling and using image residuals to drive the shape rather than full appear-

ance model.We show examples of search and give the results of experiments

comparing the performance of the different algorithms.

1Introduction

Model based methods are now widely used in image interpretation.By constraining valid solutions a more robust result can be obtained.Recently models have been developed which represent the full appearance of an object,allowing convincing synthetic images to be generated[3][8][9][11].With such models image interpretation can be interpretted as an optimisation problem in which we seek the parameters which minimise the difference between a synthetic model image and the target image.Typically the models will have50 or more parameters.Optimisation in such a high dimensional space using standard meth-ods is possible but slow[9].However,by exploiting the relationship between parameter displacements and image differences,a fast algorithm can be developed.

An Active Appearance Model(AAM)contains a statistical model of the shape and grey-level appearance of an object of interest,which can befit rapidly to an example in a new image[3].The appearance model,given a good enough training set,can generalise to almost any valid example of the class of objects represented,potentially giving a full photo-realistic approximation.

During a training phase a model instance is randomly displaced from the optimum position in a set of training images.The difference between the displaced model instance and the image is recorded,and linear regression is used to estimate the relationship be-tween this residual and the parameter displacement.

A Comparative Evaluation of Active Appearance Model Algorithms相关文档

最新文档

返回顶部