Image Processing Reference
In-Depth Information
human face recognition (without person identity) is not known to sufficient accuracy.
In humans, by using magneto resonance studies, the face identity establishing system
engages a brain region called
fusiform gyrus
. However, it may not exclusively be
devoted to face identification, as other sub-categorization of object tasks activate this
region too.
As pointed out above, humans are experts in face recognition, at an astonishing
maturity level, even in infancy. Our expertise is so far-reaching that we remember
hundreds of faces, often many years later, without intermediate contact. This is to
be contrasted to the difficulty in remembering their names many years later, and to
the hopeless task of remembering their telephone numbers. Yet this specialization
appears to have gone far in some respects and less so in others. We have difficulty
recognizing faces of another ethnic group versus own group [36, 38, 65, 158]. For
an african-american and a caucasian, it is easier to recognize people of their own
ethnicity as compared to cross-ethnic person identification, in spite of the fact that
both groups are exposed to each other's faces. Besides this cross-ethnic bias, hair
style/line is another distraction when humans decide on face similarities, [24, 39,
201]. Recently [24], another factor that biases recognition has been evidenced (Fig.
1.9). Women had systematically higher correct answer frequencies than men in a
series of face recognition tests (Fig. 1.10), taken by an excess of 4000 subjects. A
possible explanation is that face identification skill is more crucial to women than
men in their social functioning.
1.9 Further Reading
Much of the retinal cell connections and cell types were revealed by Santiago Ra-
mon y Cajal [44] at the begining of the 1900s using a cell staining technique known
as Golgi staining. However, the current understanding of human vision is still de-
bated at various levels of details (and liveliness) as new experimental evidence ac-
cumulates. An introductory, yet rich description of this view is offered by [113, 121]
whereas [173] offers discussions of the neuronal processing and organization w.r.t.
the experimental support. The study [109] offers a more recent view of mammalian
vision with extrapolations to human vision, whereas [235] presents the current view
on human color vision. The study in [189] provides support for Hubel and Wiesel's
wiring suggestion to model simple cell responses from LGN responses, whereas that
of [206] offers an alternative view that also plausibly models the various types of
simple cells that have different decreases in sensitivity when stimulated with nonop-
timal directions. The reports in [40, 69], provide a broad overview of the human
facerecognition results. The study of [196] suggested a nonuniform resource allo-
cation to analyze static images in computer vision. In analogy with the cortical cell
responses to moving patterns, one could differentiate resource allocations in motion
image processing too. This can be done by designing filter banks containing elements
that can analyze high spatial frequencies moving slowly, as well as low spatial fre-
quencies moving fast at the cost of other combinations. A discussion of this is given
in [21].
