CAP 6416 Final Examination Study Guide

(Computers and Vision)

Fall 1994

Document is currently in flux!

Chapter 1 Material

Be able to describe the roles of the various components of vision systems.
Chapter 2 Material
Describe the structure of the eye and what roles the components play in the human vision system (HVS).
Discuss the role of rods and cones in scotopic and photopic vision and color/illumination discrimination, and the distribution of rods and cones in various regions of the retina.
What is the behavior of the HVS w.r.t. actual luminous intensity and subjective brightness.
Discuss the Weber ratio. Be able to explain a statement like the following:
The human vision system follows a Weber law with respect to estimates of object width as a function of object size in the field of view.
Describe the relation of illumination and reflectance to image intensity described in Gonzalez and Woods's simple image model.
Discuss the relation of image size, and dynamic range on the amount of memory required to store and image. How does changing each of the quantities affect the presentation quality of an image?
Discuss four- and eight-connectivity.
What is a connected component?
How can Warshall's algorithm be used to identify connected components in a set of points?
What matrices may be used to describe translation, scaling, and rotation of points in an image?
What are the formulae describing perspective transformations from world coordinate points into points in an image plane?
How can two epipolar images be used to identify world coordinate locations associated with image objects? Describe all the steps associated with this process.
Image Algebra
Discuss the relation of point sets, value sets, and images.
What is a neighborhood?
What is a template?
Formally define what it means for operation * to be induced pointwise on two images.
What is the result of composing an image with a point-mapping function? What is the result of composing a value-mapping function with an image?
How could you use point transformations (as described in chapter 1) to translate, scale, or rotate an image? What image algebra operations would be employed?
Why is an image algebra of any value in developing computer vision programs in the first place?
Chapter 8 Section 4 (Morphology)
Define the operations dilation, erosion, opening, and closing.
Define Minkowski addition and subtraction.
What is the hit-and-miss (a.k.a. hit-or-miss) transform?
Discuss morphological techniques for boundary extraction, region filling, connected component extraction, thinning, skeletonizing.
Draw diagrams to demonstrate the behavior of opening and closing in gray-level images.
Items from Chapter 8 sections 1 to 3
How is a chain code used to represent an object's boundary? How can a unique starting point for a chain code be determined? How would one normalize the chain code for rotation and why is this method not robust?
What requirements must be satisfied by an object to be able to represent its boundary with a signature as described in section 8.1.3.
What is the convex hull of a set?
Discuss the properties of Fourier Descriptors that make them desirable or undesirable as a representation for object boundaries.
Chapter 3 Material
Chapter 4 Material
How is contrast modification used for enhancement? Explain the relation between constrast stretching and dynamic range compression. What is gray-level slicing?
What is an image histogram? Discuss the relation of an image histogram to a probably density function (pdf).
What is histogram equalization? How can one modify the gray values of an image so that its histogram matches an a-priori pdf?
What is local histogram equalization and why is it a reasonable idea?
What are the requirements of an imaging system in order to be able to exploit digital image subtraction as a way of identifying objects or features?
What are the requirements of image averaging, and disucss the statistical properties of averaged images that demonstrate that averaging (in the right setting) can reduce noise.
Discuss the uses of lowpass and highpass filtering and the relationships between spatial convolution kernels for these filters and their frequency domain counterparts. Which is used for smoothing and which for sharpening?
How does high-boost filtering relate to sharpening? What is unsharp masking?
Discuss the relations between Butterworth low- and highpass filters and ideal low- and highpass filters. Where does the "ringing" in the use of ideal filters come from?
What fundamental property of the image model employed by Gonzalez and Woods does homomorphic filtering exploit?
Chapter 7 Material
Discuss the structure of templates that might be employed to detect point, line,and edge features in an image.
Why is it both reasonable and unreasonable to use first derivative operators as edge detectors? What do such detectors identify well and poorly?
Marr and Hildreth
What unique property of the Gaussian makes it a good choice for smoothing images in the spatial domain if frequency information is to be extracted?
What assumptions about regions surrounding edge points do Marr and Hildreth make in order to use the Laplacian to indicate the presence of an edge? (This is required in order to employ Theorem II.) Is this a reasonable assumption?
Define the two terms zero-crossing and zero-crossing segment. Which one is used in order to discover the other, and how are these found in the first place?
In what way do Marr and Hildreth say that multiple filter sizes should be used to detect edges? What is the special role that small spatial channels play?
What important role do Marr and Hildreth assign to the experimental results of Hubel and Wiesel? Why should facts about primate vision systems be important in computer vision? What aspects of the primate vision system are supported by the computational techniques desribed by Marr and Hildreth?
Canny
How does Canny's work differ from that of Marr and Hildreth and how is it similar?
Canny uses images with Gaussian additive noise. Describe how one can add Gaussian noise to an image.
What are the two primary criteria for a good edge detector, and what third criterion does Canny add to these?
What is the role of signal-noise ratio (SNR) to good edge-detector criteria?
How does Canny measure localization of edges?
What is the effect of Gaussian noise on the number of edges reported by a step-filter within one filter width?
What mechanism does Canny use to identify the best edge detector satisfying all his edge detector criteria?
How is Canny's overall approach similar to, yet different from Marr and Hildreth's approach? What is his driving motivation and how is it distinct from theirs?

$Id$