| Course#: 526 U1090 | Date: Dec. 9, 1997 |
| Your ID#: | Instructor: Chiou-Shann Fuh |
| Your Name: |
Hint: Answers can always be found from the list below:
4-connected,
8-connected,
affine transformation,
antiextensive,
arc segmentation,
arcs,
area,
area of the ellipse,
ascending reachability,
associative,
autobinomial form,
autocorrelation function,
autoregression models,
autoregressive moving average,
average gray level,
background normalization,
bands,
Bayes decision rules,
Bayes gain,
Bayesian approach to Hough transform,
Bayesian fitting,
binary machine vision,
border tracking,
boundary extraction,
boundary sequence,
bounding rectangle,
bounding second derivatives,
box filter,
breakpoint optimization,
busyness,
centroid,
centroid-linkage criterion,
centroid-linkage region growing,
chain rule,
finding circles,
circularity measure,
closing,
closing characterization,
clustering,
coarse textures,
commutative,
concave,
conditional dilation,
conditional independence assumption,
conditioning,
conditioning and labeling,
connected,
connected components,
connected components analysis,
connected components labeling,
connected shrink,
connectivity,
connectivity number operator,
convex,
convolution,
co-occurrence matrix,
co-occurrence of primitives,
co-occurrence probabilities,
co-occurrence statistics,
corner detection,
corner finding,
corner points,
counting,
covariance,
cross-correlation,
crossing number,
curvature,
cutset,
decision rule,
decision rule construction,
decision rule error,
decision tree,
decision tree construction,
density of texture primitives,
descending components,
descending reachability,
deterministic Bayes rule,
deterministic decision rules,
deterministic maximin rule,
diagonal projections,
digital image,
digital transform method,
dilation,
directional derivative edge finder,
directional derivative edge operator,
directional derivatives,
discrete Chebyshev polynomials,
discrete Gauss-Markov field,
discrete orthogonal polynomials,
discriminant function,
distance transform,
distance transformation,
dominant points,
duality,
economic consequence,
economic gain matrix,
edge detection,
edge linking,
edgeness per unit area,
edges,
ego-motion polar transform,
eigenvalues,
eigenvectors,
ellipse,
endoskeleton,
entropy purity function,
equal-probability-of-ignorance,
equivalence table,
erosion,
error of commission,
error estimation,
error of omission,
exclusive,
exoskeleton,
expand,
expected gain,
expected profit,
extensive operators,
extracting,
extrema per unit area,
extremal points,
extremum sharpening,
facet model,
fair game assumption,
false-alarm rate,
false-detection rate,
false-identification error,
fast dilation,
feature extraction,
feature selection,
features,
fill,
flat,
focus-of-attention rules,
Fourier transform,
fractal dimension,
fractal signature,
fractal surface,
fractal surface area,
Frei and Chen edge detector,
Frei and Chen gradient masks,
Gauss-Markov model,
Gauss filter,
general maximin rule,
generalized closing,
generalized co-occurrence,
generalized distance transform,
generalized gray level spatial dependence,
generalized opening,
generation of synthetic texture,
genus,
global properties,
gradient edge detector,
gradient-based facet edge detection,
gradient descent,
gradient edge detection,
gradient edge detectors,
gradient inverse weighted,
gradient magnitudes,
granularity,
gray level,
gray level co-occurrence,
gray level difference distribution,
gray level difference probability,
gray level primitives,
gray level properties,
gray level variance,
gray levels,
gray scale,
gray scale closing,
gray scale dilation,
gray scale erosion,
gray scale opening,
grouping,
grouping operation,
grow,
Hadamarad transform,
hereditary,
high Laplacian magnitude,
hillside,
histogram,
histogram mode seeking,
hit-and-miss transform,
hold-out method,
hole,
holes,
homogeneous texture,
homomorphism,
horizontal projection,
Hough transform,
hybrid-linkage combinations,
hybrid-linkage region growing,
hysteresis smoothing,
idempotency,
identity gain matrix,
image,
image formation,
image region,
image segmentation,
image texture gradients,
impulse response function,
increasing operators,
incremental change along the contour line,
incremental change along the tangent line,
inference rules,
inflection point,
influence zones,
instantaneous rate of change,
integrated directional derivative,
integrated first directional derivative,
intensity image,
ISODATA,
isodata segmentation,
isotropic derivative,
iterated facet model,
iterative endpoint fit and split,
iterative rule,
kernel,
Kirsch compass masks,
Kullback information distance,
labeling,
labeling operation,
Laplacian,
Laplacian of the Gaussian kernel,
least-squares curve fitting,
least-squares fitting of a line,
least-squares fitting problem,
least-squares procedure,
likelihood ratio test statistic,
line detection,
line fitting,
linear shift-invariant operators,
linear shift-invariant neighborhood operators,
local tangential deflection,
long-term memory,
macrotexture,
major axis,
major axis length,
mark-interior/border-pixel operator,
Markov chain,
Markov mesh models,
Markov random field,
match,
matching,
mathematical morphology,
max Roberts gradient,
maximin decision rule,
maximum likelihood decision rule,
maximum-likelihood test,
measurement,
measurement-space-clustering image segmentation,
measurement-space-guided spatial clustering,
medial axis,
medial axis transform,
median operator,
median root image,
microtexture,
midrange estimator,
minimum mean square noise smoothing,
Minkowski addition,
Minkowski subtraction,
minor axis,
minor axis length,
misdetection rate,
misidentification error,
mixed second moment,
mixed spatial gray level,
morphological pattern spectrum,
morphological sampling theorem,
morphological skeleton,
motion-based segmentation,
multiband images,
multidimensional measurement-space clustering,
multiplicative maximum,
nearest neighbor rule,
nearest neighbor,
neighborhood operators,
neural networks,
Nevatia-Babu 51#15 compass template masks,
Newton method,
noise cleaning,
non-minima-maxima operator,
nonrecursive neighborhood operators,
number of shortest paths,
numeric,
opening,
opening characterization,
order statistic approach,
orientation angle,
orientation of ellipse,
orientation of major axis of ellipse,
pair relationship operator,
peak,
peak features,
peak noise,
peak noise removal,
perimeter,
pit,
pixel,
point spread function,
Poisson line model,
Poisson line process,
position invariant,
power spectrum,
Prewitt edge detector,
primitive spatial event,
principal-axis curve fit,
prior probability,
projection,
projection segment,
projection segmentation,
prominent corner point,
quick Roberts gradient,
radius of fusion,
random mosaic models,
range image,
ravine,
reachability operator,
recursive morphological erosion,
recursive morphology,
recursive neighborhood operators,
region growing,
region-growing operator,
region properties,
region-shrinking operator,
region-of-support,
regions,
relational homomorphism,
relative extrema density,
relative extrema operator,
relative extrema primitives,
relative height,
relative maxima,
relative minima,
reserved judgment,
ridge,
ridge and ravine continua,
Roberts edge detector,
Roberts gradient,
Robinson compass masks,
robust estimation,
robust line fitting,
rotated ellipse,
rule-based segmentation,
run-length encoding,
Rutovitz connectivity number,
saddle,
second column moment,
second diagonal moment,
second mixed moment,
second moments,
second-order approximation,
second-order column moment,
second-order mixed moment,
second-order row moment,
second row moment,
segmentation,
segmentation tree,
segmentations,
segmented,
selected-neighborhood averaging,
separability,
separated,
separation,
shape properties,
shared-nearest-neighbor idea,
sharpening,
shift-invariant operator,
short-term memory,
shrinking,
sigma filter,
signature,
signature analysis,
signature segmentation,
single-linkage criterion,
single-linkage region growing,
skeleton,
slant transform,
slope,
sloped facet model,
Sobel edge detector,
solid angle,
spatial clustering,
spatial gray level dependence,
spatial gray level differences,
spatial moments,
spatial relation,
spatial relationships,
split and merge,
split-and-merge algorithm,
splitting algorithms,
splitting and merging,
spoke filter,
statistical pattern recognition,
step edge,
finding straight-line segments,
structural pattern recognition,
structuring element,
symbolic,
symbolic image,
symmetric,
symmetric axis,
synthetic texture,
tangent line,
tangential angle deflection,
template matching,
texel identification problem,
textural edgeness,
textural energy,
textural plane,
textural primitive,
textural surface,
texture analysis,
texture features,
texture primitives,
texture segmentation,
thickening,
thinning,
thinning operator,
threshold decision,
thresholding,
top hat transformation,
top surface,
topographic primal sketch,
tracking,
trimmed-mean operator,
Tukey's biweight,
two-dimensional extrema,
type I error,
type II error,
umbra,
umbra homomorphism theorem,
uniform bounded-error approximation,
uniform error estimation,
units,
unsharp masking,
variogram,
vector dispersion,
vertical projection,
Wallis neighborhood operator,
weighted-median filter,
within-group variance,
Yokoi connectivity number,
zero-crossing edge detector,
zone of influence.
1. (10%)
(a) Chapter titles covered so far include: Computer Vision: Overview,
Binary Machine Vision: Thresholding and Segmentation,
Binary Machine Vision: Region Analysis, ,
, , ,
2. (10%)
(a)
4#4: row, 5#5: column
23#23: grayscale intensity, 2#2 spans each gray level value e.g. 3#3
4#4: operator counts the number of elements in a set
5#5 is called .
(b) Two techniques to find a threshold that minimizes a criterion function is:
minimizing within-group variance and minimizing .
(c) distinguishes pixels that have higher gray values from
pixels that have lower gray values. Pixels whose gray values are high enough
are given the binary value 1.
Pixels whose gray values are not high enough are given the binary value 0.
(d) of a binary image consists of the connected components
labeling of the binary-1 pixels followed by property measurement of the
component regions and decision making.
(e) The only properties a pixel has are its position and its .
3. (10%)
(a) 6#6: region
6#6 is the region's .
(b)
7#7
and
8#8 is the position of
the .
(c)
9#9 is the
of the region.
(d)
10#10 is the
of the region.
(e) 11#11: set of pixels in designated spatial relationship e.g. 4-neighbors
12#12 is the .
4. (14%)
(a) Statistical pattern recognition begins with units, such as
or projected segments, on which a variety of measurements
have been made.
Each unit has an associated measurement vector.
(b) The is designed optimally
to assign each unit to a class or a category on the basis of its
measurement vector.
Optimally can mean, for example, with the smallest classification error for
a given set of measurements and for a given computational complexity.
(c) Statistical pattern recognition techniques include:
1. and extraction techniques either to reduce the number
of measurements to be made or to reduce the dimensionality of the vectors
representing the measurements made to the decision rule,
2. Decision rule construction techniques,
3. Techniques for the estimation of .
(d) When it is important to be right the largest possible fraction of the time,
which economic gain matrix should be used?
(e) Given that an object is good, the conditional probability that it is
detected as bad is .
Given that an object is bad, the conditional probability that it is
detected as good is .
5. (16%)
(a) The four most important operations in mathematical morphology are
, , ,
(b) Binary dilation is the same as ;
binary erosion is similar to .
(c)
13#13: dilation
is .
14#14: dilation is .
6. (14%)
(a) Neighborhood operators whose output is only a function of an input image
neighborhood related to the output pixel position are
called .
Neighborhood operators whose output depends in part on previously generated
output values are called .
(b) 36#36: weight function: kernel or mask of weights
9#9: domain of 36#36
15#15
is the of 22#22 with 36#36.
16#16
is the of 22#22 with 36#36.
(c) Operators that have a domain are usually defined in
terms of arithmetic operations, such as addition, subtraction, or computation
of minima or maxima.
Operators that have a domain are defined in
terms of Boolean operations, such as AND, OR, or NOT, or table-look-up
operations.
(d) Its action on identical neighborhood spatial configurations is the same
regardless of where on the image the neighborhood is located.
This kind of neighborhood operator is called .
7. (16%)
(a)
17#17 are masks used for the .
(b)
18#18 are masks used for the .
(c)
19#19 are masks used for the .
(d)
20#20 are masks used for the .
(e) uses neighborhood spatial coherence and neighborhood
pixel value homogeneity as its basis.
(f) The operator that computes the equally weighted average is called the
. The operator is important because of the ease with which
it can be made to execute quickly.
(g) takes linear combinations of the sorted values of all the
neighborhood pixel values.
(h) The fixed-point result image of a median filter is called the
.
8. (10%)
(a) The principle states that the image can be thought of
as an underlying continuum of piecewise continuous gray level intensity
surface.
The observed digital image is a noisy, discretized sampling of a
distorted version of this surface.
(b) are defined to occur at points for which the first
derivative is zero and the second derivative is negative.
(c) The looks for high values of estimated first
derivatives.
The looks for relative maxima in the value of the first
derivative taken across a possible edge by looking whether the second
derivative crosses zero.
(d) A digital occurs on a digital image when there is a
simply connected sequence of pixels with gray level intensity values that
significantly higher in the sequence than those neighboring the sequence.
Significantly higher or lower may depend on the distribution of brightness
values surrounding the sequence, as well as on the length of the sequence.