Summary: III. 2­d elastic models ­ image
matching
III.1 Introduction
ffl In images of an object class there is much more information than shape of con­
tour. Images of object class exhibit many topological and geometric similarities
which are ignored when the contour alone is computed.
ffl Variations between images are often smooth ­ elastic.
With some pulling, stretching and squeezing ­ take one image from the class
and transform it into another.
ffl Z = R ­ The set Z is the entire reference grid.
\Theta ­ smooth one­to­one mappings of R into L.
T (z); z 2 R template image of the object: typically one random image of the
family of images.
ffl Initial ` given ­ in particular scale and translation.
ffl Model I(x) for x only around `R. No background modeling.
Any image of the object is generated as
I(`(z)) = T (z) +N(z); z 2 R; (III.1)
,
21


Source: Amit, Yali - Departments of Computer Science & Statistics, University of Chicago

Collections: Computer Technologies and Information Sciences