Image Processing Reference
In-Depth Information
40
40
0
0
-40
0
40
-40
0
40
Fig. 9.16. Left : the retinotopic sampling grid with axes graded in pixels. Right : a few receptive
fields are represented as sets of concentric circles. Adapted from [205].
9.8 Face Recognition by Gabor Filters, an Application
We discuss here a saccadic search strategy, a general-purpose attentional mechanism
that identifies semantically meaningful structures in images by performing “jumps”
( saccades ) between relevant locations [205]. The saccade paths are chosen automat-
ically and rely on apriori knowledge of facial features that are modeled by means of
Gabor filters discussed in the previous sections. Additionally, here they are applied
to a log-polar grid , that is, together with Gabor filter responses, called the retino-
topic sensor , because each sampling point is not a gray value but is instead an array
of responses coming from Gabor filters designed in the log-polar frequency plane .
The usefulness of the concept to complex cognitive tasks is demonstrated by facial
landmark detection and identity authentication experiments over the M2VTS and
Extended M2VTS (XM2VTS) databases.
The saccadic search strategy and face recognition are based on a sparse retino-
topic grid obtained by log-polar mapping [196], also called log-z mapping :
ξ =log x 2 + y 2
η =tan 1 ( y, x )
(9.54)
that we will return to in Chap. 11 in the context of other image processing tasks. The
log-polar grid, which is in the spatial domain, is used to sample the original image
but not for extracting the grayvalues. At each log-polar grid point in the image, a
local Gabor decomposition of the image is performed to the effect that they mimic the
simple cells of the primary visual cortex having the same receptive field but different
spatial directions and frequencies [113].
The log-polar grid used in the example results discussed here is shown in Fig.
9.16. The simple cells are modeled by computing a vector of 30 Gabor filter re-
Search WWH ::




Custom Search