CSC 249/449 Computer Vision: Test2 Study Questions
The following are examples of questions that have appeared on previous
second exams. Similar or identical questions may appear on the upcoming exam.
Most of the questions will be on material from after the midterm,
however there is some basic material from before the midterm
that you need to remember, and some questions will be asked on it, though
perhaps not as many as the number of sample questions might indicate.
Some of the questions below may be on areas we have not covered this year.
They will not be asked.
We may have covered areas not represented by any question here.
You are responsible for those areas anyway.
General
-
Describe two specific applications in which computer vision/image processing
techniques have been used successfully. Specify in some detail
(10 points worth) the techniques used in each case.
-
Give two specific computer vision problems in which active research is
being pursued and describe in some detail the principal difficulties.
Image Formation
(No, you are not allowed to forget this.)
-
Define radiance (L) and irradiance (I).
-
Consider a pinhole camera model with an image plane 1 meter
from the center of projection.
Suppose we have a cube 1 meter on a side positioned
so that the optical axis goes through the center of one edge
and out through the center of the diagonally opposite edge.
The nearest edge is one meter from the center of projection.
Sketch the shape of the projection, together with sufficient
dimensional information to draw the shape exactly.
-
Suppose we have a sphere 10cm in diameter
lying on the optical axis of a camera with a focal length of 10cm.
If the sphere is 1 meter from the center of projection (Z distance = 1 meter),
what is the size of the circular disk it projects to?
Now suppose the sphere is no longer on the optical axis, but lies
1 meter from it in the X direction (the Z distance is still 1 meter)
In other words, the sphere lies 45 degrees off the optical axis.
Why is the shape of the projection no longer a circle.
What geometric figure is the projection?
What are its dimensions?
Image Processing
(Can't forget this either. There will not be as many questions as
the relative number under this section might indicate, but you should
be able to answer any of them.)
-
Give two processes that produce "noise" in digital images obtained from real
detectors, and describe methods for dealing with the effects in each case
-
What is the Convolution Theorem? What are some of its practical ramifications
for image processing?
-
What is a thinning algorithm? Where are they useful?
Describe briefly one example of a thinning algorithm.
-
Apply the indicated filter (e.g. a sobel mask) to the image shown below.
What mathematical operation does the filter approximate?
In qualitative terms, what is the effect of applying the filter to an image?
-
Compare 1st and 2nd derivative methods of local edge detection.
What are the advantages and disadvantages of each?
-
Give a specific application for high-pass filtering in image processing.
Give one for low-pass filtering.
-
Give two methods of smoothing image data and compare their performance.
-
In what situations is it useful to apply a smoothing operation to
an image?
Compare averaging and median filtering as methods of smoothing.
What are the advantages and disadvantages of each?
Biology
(Some of these questions were also on the midterm study list, however
some years we don't cover biology until after the midterm, so biology is
fair game)
-
Briefly describe the following structures in the eye and their functional
significance.
- Lens
- Cornea
- Iris
- Retina
- Fovea
- Rod
- Cone
- Ganglion cell
- Optic nerve
- Blind spot
-
Draw a schematic diagram of the human eye showing all the above
-
What is meant by the receptive field of a neuron?
What is the significance of the concept?
-
Comment in detail on evidence/arguments for and against,
the statement "the visual cortex of the brain contains edge detectors".
Is the statement meaningful? Why or why not?
-
Describe briefly what the experiments of Hubel and Weisel showed about
information processing in the cat visual cortex.
Segmentation and Grouping
-
Describe how gray level histograms can be used for image segmentation.
-
Describe in detail how color histograms can be used in an attempt
to segment "same-colored" objects in real color images.
What problems arise? what are some ways of dealing with them?
-
Describe how (pixel-wise) region-growing techniques might be used in a
attempt to segment homogeneous objects in gray-scale or color images.
What problems arise? what are some ways of dealing with them?
-
Describe briefly how conditional statistics can be used for
region segmentation.
-
Describe one of the following local search methods for finding
boundaries.
When is the technique useful?
What are its limitations?
- Boundary adjustment
- Divide and conquer
- Non-linear edge space correlation
- Snakes
- Edgel Linking
- Curve growing
Shape Representation
-
Give two examples of boundary representations and briefly describe pros and
cons for each.
-
Give 2 examples of region representations along with some of their
pros and cons.
-
Define any three of the following (2D) shape descriptors and give an example of
a situation in which each would be useful.
- Moments
- Euler number
- Projection
- Direction histogram
- Templates
- Any other favorite of yours
-
Give two examples of 3-D representations and describe some advantages and
disadvantages of each.
Classification and Matching
-
What is k-nearest neighbor classification?
How does it differ from nearest neighbor classification?
Describe a situation in which it might work better than nearest-neighbor
classification.
-
What is the distinction between analog and symbolic matching
matching processes? Give an example of each.
-
What are some ways that matching is employed in computer vision, and why
is it often a hard problem?
-
Describe graph isomorphism, subgraph isomorphism, and double subgraph
isomorphism. Give an example of a computer vision problem to which
each is appropriate.
-
Construct the association graph for the following matching problem and
indicate the maximal clique. (Recall that the nodes of the association
graph correspond to pairings of nodes which are consistent
with respect to unary relations, and the edges correspond to mutually
consistent pairings with respect to higher order relations)
In the following figure, P and Q represent unary relations, and the arrows
represent a single binary relation.
Connectionism
(I don't always cover this)
-
What does the perceptron convergence theorem state?
-
What are some of the limitations of 1-layer perceptrons as pattern
recognition devices?
-
What does the back propagation algorithm accomplish and what is its
significance?
-
What is a relaxation process?
Give an example of a computer vision problem that can be formulated
as a relaxation process, and describe, briefly, the formulation.
Bayesian Reasoning
(Or this)
-
Suppose that a test has been developed for a certain disease which has
a false positive rate of @10 sup -4@ and a false negative rate of @10 sup -2@.
That is,
- p(positive | not infected) = .0001
- p(negative | not infected) = .9999
- p(positive | infected) = .99
- p(negative | infected) = .01
If the incidence of infection in the population is .00001 (one in
one hundred thousand - this is a rare disease), compute the following.
1. p(infected | positive) (the probability that a person testing positive
is actually infected).
2. p(infected | negative) (the probability that a person testing negative
is actually infected and thus escapes detection).
-
Suppose that a test has been developed for a manufacturing defect, which has
a false positive rate of .01 and a false negative rate of .0001.
That is,
- p(positive | not flawed) = .01
- p(negative | not flawed) = .99
- p(positive | flawed) = .9999
- p(negative | flawed) = .0001
If the incidence of defects produced by the manufacturing process
is .001 (one in one thousand), compute the following.
1. p(flawed | positive) (the probability that a part testing positive
is actually flawed).
2. p(flawed | negative) (the probability that a part testing negative
is actually flawed and thus escapes detection).
Object Recognition
-
Describe in some detail how principle components techniques can be used
to recognize and distinguish faces.
What limitations are associated with the method?
Given these limitations, can you think of a practical application
where this method might be useful?
-
Describe in some detail how principle components techniques could be used
in a system to recognize and distinguish rigid 3D objects in gray-scale
images.
What conditions must be met in order for the system to work?
What are the resulting limitations on the performance of the system?
Given these limitations, can you think of a practical application
where this method might be useful?
Color
-
Suppose that colors are represented as points (r,g,b) in 3-space
where r, g, and b represent the responses of the red, green, and
blue cones respectively, and are scaled to lie between 0 (minimum)
and 1 (maximum).
Explain why some points in the space @[0,1] times [0,1] times [0,1]@
do not correspond to any physically realizable stimulus.
-
How is it possible for colors to be spectrally distinct (i.e. contain
different wavelength distributions), and yet appear identical to the
eye? Be specific.
-
Explain the distinction between the opponent color and the red/green/blue
methods of describing color.
-
What is your favorite color?
Motion
-
What is meant by the 2-D motion field?
What is meant by optical flow, and how is it related to the motion field?
-
Briefly describe differential and correlation-based methods of determining
the motion field in an image sequence. What are the advantages and
disadvantages of each method?
- What are two applications of motion analysis in machine perception?
Stereo
-
What is meant by the ``correspondence problem'' and why is it difficult?
-
What is an epipolar line? How is it determined?
How can the concept be used to simplify the correspondence problem in
stereo vision?
-
What is the camera calibration problem?
If the internal camera parameters are known, how many calibration
parameters need to be determined to solve the 2-camera stereo vision
problem. How can "calibration objects" be used to address this problem,
and what minimum conditions must be fulfilled to solve it?
Texture
-
What is visual texture?
-
What is a texel and how are they used?
-
What is the difference between statistical and structural texture models?
-
What is a gray-level co-occurance matrix?
How can they be used for texture discrimination?
-
Briefly describe two methods of statistical texture classification
System Design
-
A group of primatologists wants to study the details of the daily movement,
activities, and and interactions of a group of 6 chimpanzees living on
"chimp island" - a natural, though somewhat open habitat about 50 meters
in diameter, bounded on all sides by water, in the San-Diego zoo.
Since they don't want to sit all day every day recording the second-by second
positions and activities of the chimps, they have come to you, a computer
vision expert, for automated assistance.
They are interested in both compiling statistics about the movement and
location of individuals, and in the frequency and locations of different
interactions and activities (feeding, sleeping, grooming, fighting, etc.)
They are willing to help in labeling
relevant activities, even to the point of answering a few hundred
quick questions per day of data (what's she doing here?),
but they don't want to sit through 24 hours of video to do it.
Ultimately they want an automated database that they can use to
find out how many hours a day chimp Jane sleeps and where,
histogram preferred eating locations,
obtain statistics on who grooms whom, etc.
Their equipment budget is an exceedingly generous $50K.
What help can you give them?
P.S. If you want to "bar-code" the chimps, you
have to do it in a way that will bother neither the chimps, nor the
visitors to the zoo.
-
The security staff at a National Lab, in the wake of a scandal involving
missing computer drives, has asked for your
advice on improving physical security at a small weapons design facilty
(about 100 total employees) through automated video surveillance.
They already have significant conventional security measures in place.
The building is secured by a human staff
and computer logging equipment, and individual rooms and zones are all
card-keyed and logged. They have some video monitoring as well,
though not enough
to cover the entire facility at high resolution, nor do they have the
personnel to man such a degree of scrutiny for rare events.
What they want, ideally, is an automated surveillance system that will detect
"suspicious events" (like someone putting a laptop in an odd place, or
pocketing removable media, or going to an unusual location),
but they realize this may be difficult.
At a minimum, they would like the system to archive and index video of
all activity that might be "significant" so that they can investigate
later. Naturally they would like closeups of significant events
(e.g. unauthorized hard copying of papers or computer files, or unusual
entrances and activities).
What help and advice can you give them? Can you think of any way
of helping them improve physical security through automated video
surveillance? Their budget is effectively unlimited
(at least up to a few million dollars), but they need a system
in place in 3 months.
P.S. If your advice is "forget it", you've got to carefully back this up with
some technical rational.
Back to
vision course main page