This paper presents a new technique for creating efficient and compact
models from data, called matching pursuit filters. The design of a ma
tching pursuit filter is based on an, adapted wavelet expansion, where
the expansion is adapted to-both the data: and the pattern recognitio
n problem being addressed. This contrasts with most adaptation Schemes
, where the representation is a function of the data, but not of the p
roblem to be solved. This approach does not decompose the images in th
e training set individually, but rather determines the expansion by si
multaneously decomposing all the images. Because it uses two-dimension
al wavelets as the building blocks for the decomposition, the represen
tation is explicitly two-dimensional and is composed of local informat
ion. Matching pursuit filters can be trained to detect, recognize, or
identify objects and have been applied to recognizing faces and detect
ing objects in infrared imagery.