Biomedical Engineering Reference
In-Depth Information
system features are built into integrated units. The assumption is that units at one or more
of these levels access memory for known categories of patterns, and the best match deter-
mines the category to which the incoming stimulus will be assigned. Typically, pattern
recognition theories assume that not only are these processes driven by sensory stimu-
lation in a bottom-up fashion, but, also, that they rely on acquired knowledge and that
current expectations will be derived in a top-down driven manner [54].
These theoretical assumptions are so sufficiently broad that they can be applied to any
sensory modality, for example, vision, touch, audition, or olfaction. But there is no doubt
that by far and away the most predominant modality examined in the studies of pattern
recognition is vision, followed by speech recognition. In distinct contrast, little work has
been devoted to the study of how familiar patterns are recognized by touch [54].
Intuitively, one may feel that the emphasis on vision is appropriate because it is the
way objects are typically recognized in the real world. One might believe that there is
little role for object recognition by touch in everyday life. It could be argued, however,
that the importance of touch in everyday object recognition has not been highlighted to
the extent that it deserves and that, in fact, objects are more frequently recognized by
touch than people would believe. People often interact with an object without looking at
it but, in order to do this and if they wish to acquire any form of recognition, they must
first know what the object is. When we dress, we do not have to fixate on buttons to
fasten them; the button and buttonhole can be found by touch. When driving a car, we
shift gears and adjust knobs without looking at them rather than taking our eyes off the
road. Sometimes we are able to locate to certain items by feel alone, such as our wallets
or keys, which exemplifies the fact that recognizing common objects by touch is not only
something that we
can
do, but something that we
often
do on a reflexive basis [54].
Haptics is a multi-disciplinary field which plays a key role in any scientific multi-
disciplinary field that involves touch, neuroscience, virtual reality, and robotics. Before
continuing with our discussion on haptic object recognition, let us first briefly consider
how visual object recognition is performed. One model of object recognition, based on
neuropsychological evidence, provides information that allows us to divide the process
into four different stages [55 - 57].
1. Processing of basic object components, such as color, depth, and form.
2. Grouping similarly shaped components together in order to show their outlines and
segregate each visual form on the basis of figure-ground perception.
3. Storing a visual representation in data memory and then matching it against structural
descriptions.
4. Obtaining semantic attributes that are applied to the visual representation in order to
provide meaning and recognition.
Within each of these stages, there are more specific processes that take place to complete
the different processing components.
Visual recognition processing has been typically viewed as a bottom-up hierarchy in
which information is processed sequentially with increasing complexity. At the bottom
of the processing hierarchy are the lower cortical processors that deal with aspects such
as the primary visual cortex; at the top are the higher-level critical processors such as
the inferotemporal cortex (IT), where recognition is facilitated [58]. One of the more
