Page 277 - Electrical Engineering Dictionary

FDMA See frequency division multiple inspection to determine whether products are
access. within acceptable tolerances.

FDTD See finite difference time domain. feature orientation measurement of the
orientation of features, either as part of the
feature a measurable characteristic of an recognitionprocessoraspartofaninspection
object in an image. Simple examples would or image measurement process.
bearea, perimeter, andconvexity. Morecom-
plex features use vectors; examples include
feature recognition the process of lo-
moments, Fourier descriptors, projections,
cating features and determining what types
and histogram based features. Features are
of features they are, either directly or indi-
frequently used to recognize classes of ob-
rectly through the location of sub-features
ject, and sets of simple features can be col-
followed by suitable inference procedures.
lected into a vector for this purpose. Using
Typically, inference is carried out by appli-
both area and perimeter, for instance, one can
cation of Hough transforms or association
quickly distinguish between a circle and a tri-
graphs.
angle. May also refer to a characteristic of
a whole image: such a feature could then be
used in image database analysis. feature size the characteristic size of elec-
tronic components on a die.
feature detection the detection of smaller
features within an image with a view to in- FEC See forward error correction.
ferring the presence of objects. This type
of process is cognate to pattern recognition.
feedback (1) signal or data that is sent
Typically, it is used to locate products ready back to a commanding unit from a control
for inspection or to locate faults during in- process output for use as input in subsequent
spection. A feature can be detected by find- operations. In a closed-loop system, it is the
ing points having optimal response to a given part of the system that brings back informa-
combination of local operations such as con- tion on the process condition under control.
volutions or morphological operators. See
(2) the provision of a path from the output
also object detection.
to the input of a system, such that the output
may be made a function of both the input and
feature extraction a method of trans-
the previous outputs of the system.
forming raw data, which can have very high
dimensionality, into a lower dimensional rep- (3) the technique of sampling the output
resentation that still contains the important of an amplifier and using that information to
features of the data. modify the amplifier input signal. A portion
of the output is “fed back” to the input. Posi-
feature map a fixed geometrical struc- tivefeedbackoccurswhentheoutputisadded
ture (often two dimensional) for unsuper- to the input; negative feedback occurs when
vised learning that maps the input patterns the output is subtracted from the input. Neg-
to different output units in the structure so ative feedback, invented by communications
that similar input patterns always trigger engineer Harold Black in 1928, usually re-
nearby output units topographically. See sults in a gain–bandwidth tradeoff: decreas-
also self-organizing system, self-organizing ing and stabilizing the amplifier gain, while
algorithm. increasing the bandwidth. According to Nor-
bert Wiener, feedback is a method of control-
feature measurement the measurement ling a system by reinserting into it the results
of features, with the aim of recognition or of its past performance.

c
2000 by CRC Press LLC