Active shape model: Difference between revisions

Revision as of 10:05, 2 March 2013

Revision as of 20:03, 13 December 2013

Line 1:

'''Active shape models (ASMs)''' are [[statistical model]]s of the [[shape]] of objects which iteratively deform to fit to an example of the object in a new image, developed by Tim Cootes and Chris Taylor in 1995.<ref name=Cootes>{{cite journal| author=T.F. Cootes and C.J. Taylor and D.H. Cooper and J. Graham| title=Active shape models - their training and application| journal=Computer Vision and Image Understanding| pages=38–59| year=1995| issue=61}} [http://www.cs.ualberta.ca/~nray1/CMPUT615/Snake/cootes_cviu95.pdf]</ref> The shapes are constrained by the PDM ([[point distribution model]]) [[Statistical Shape Model]] to vary only in ways seen in a training set of labelled examples.

The shape of an object is represented by a set of points (controlled by the shape model). The ASM algorithm aims to match the model to a new image. It works by alternating the following steps:

* Look in the image around each point for a better position for that point

* Update the model parameters to best match to these new found positions

[[File:Operation Of Shape Model In Active Shape Model.jpg|frame|right|Operation of the shape model]]

To locate a better position for each point one can look for strong edges, or a match to a statistical model of what is expected at the

point. The original methodology suggests using the [[Mahalanobis distance]] to detect a better position for each landmark point.<ref name=Cootes/>

The ASM works by alternating the following steps:

* Generate a suggested shape by looking in the image around each point for a better position for the point. This is commonly done using what is called a "profile model", which looks for strong edges or uses the [[Mahalanobis distance]] to match a model template for the point.<ref name=Cootes/>

* Conform the suggested shape to the point distribution model, commonly called a "shape model" in this context. The figure to the right shows an example.

The technique has been widely used to analyse images of faces, mechanical assemblies and medical images (in 2D and 3D).