CSE 408 Fall 2021

Final Exam Topics &

Questions
[lecture number reference may be approximate, you should find the exact reference yourself]
Cheat sheet: 2 pages, i.e., 1 sheet handwritten/printed on both sides, or 2 single-side
sheets.

1. Name and explain the 5 basic tasks of data visualization. [L14, p. 4]

● Overview: Gain an overview of the entire collection

● Zoom: Zoom in on items of interest

● Filter : Filter out uninteresting items

● Details-on-demand: Select an item or group and get details

● Relate: View relationships among items

2. What is the difference between intrinsic and perceived dimension for visualization?
[L14,p.9]
Perceived dimensions are limited by display dimensions. For dimensions <= 4, intrinsic
and perceived dimensions can be same

3. Video content visualization [L14,p.10]– True/False

a. video summarization is a technique for video data visualization, - T
b. Mosaicking combines multiple frames into one shot, - T
c. a key frame is a frame representative of a shot. - T

4. What is interactive content based image retrieval and how would you do it using
current web tools to deal with the semantic gap? [L14,p.35](HW03)
a. What is interactive content based image retrieval? - In content-based image
retrieval, interaction is a complex interplay between the user, the images, and their
semantic interpretations.
b. How would you do it using current web tools to deal with the semantic gap? - The
semantic gap characterizes the difference between two descriptions of an object
by different linguistic representations, for instance languages or symbols. In HW3,
we used tools like auto-tagging to generate captions for our images. It will also use
lower-level features like texture, color, and shape. These features are either used
in combination with interfaces that allow easier input of the criteria or with
databases that have already been trained to match features (such as faces,
fingerprints, or shape matching). However, in general, image retrieval requires
 human feedback in order to identify higher-level concepts.

5. What is automated image captioning or annotation, and what is it used for (HW03).

Automatic image annotation (also known as automatic image tagging or linguistic indexing) is the process by
which a computer system automatically assigns metadata in the form of captioning or keywords to a digital image.
Image captioning has various applications such as recommendations in editing applications, usage in virtual
assistants, for image indexing, for visually impaired persons, for social media, and several other natural language
processing applications.

6. Explain sensitivity of image retrieval to image quality (HW03)

7. True or false:
a. in associative search there is only one category of interest, - T
b. answer search is the simplest type of search. [L14, p. 36] - F, far more complex

8. Provide an example where interactive retrieval is made possible by using ontologies

[L14,p.38]
- Example: Person needs help making a sandwich and needs to see any training videos.
1. Event query: Search up “make a sandwich”
2. Event Query to Concept Matching. This will result in concept based search to
figure out the best match.
a. Run EventNet Ontology (EvenNet are event-specific concepts are the
semantic concepts designed for the events of interest, which can be used
as a mid-level representation of complex events in videos.)
b. Offline Concept Discovery and Training by youtube
c. Concept Library
3. Video corpus will give us the results
4. At this point, videos of “make a sandwich” will come up.

9. Explain three types of user feedback in interactive retrieval. [L14, p. 45]

1) Indicate irrelevant results

2) Manipulate objects to adjust relationships

3) Machine learning so system can learn answers queries it previously couldn’t

answer.

10. Explain how to evaluate MMIS using task driven evaluation. [L14, p. 46-47]

-quality of the final answer

-execution time
-interaction time
-user experience(satisfaction, fatigue, annoyance)

11. Explain task-driven evaluation of MMIS [L14,p.48].

 Task Driven visualization is the evaluation of the application by measuring how well it
performs on one or a set of tasks that are representative of the tasks the application is supposed
to perform. Evaluation of task performance is done using criteria such as:
1) Quality of final answer
2) Execution time
3) Interaction count
4) User experience (satisfaction, fatigue, annoyance etc)
5) Compare against traditional practice (manual or old tool)

12. What is compression ratio? Explain the formula [L15,p.3]

Compression ratio = B0/B1 (B0 is the number of bits before compression and B1 is the number of bits
after compression) a. If compression and decompression process induce no information loss, then
compression scheme is lossless, else lossy

13. In entropy coding, what is the relationship between entropy and the number of bits
required to represent symbols of a given code? [(L09),L15,p.4,6]
Entropy = Σ (pi * log2(1/pi))

For a symbol si, pi is the probability of si occurring in S.

log2(1/pi) denotes the number of bits required to represent si.

14. Draw the coding tree for the word BUBBLE using the Huffman algorithm. What is the
symbol for E? [L15,p.31-36]

https://cmps-people.ok.ubc.ca/ylucet/DS/Huffman.html

15. What is lossy compression? [L16,p.2]

In lossy compression, the compressed data is not the same as the original uncompressed
data, but a close approximation of it. Lossy compression yields higher compression ratios
than lossless compression.
16. Explain the rate-distortion function [L16,p.4]
The rate-distortion theory addresses the problem of minimal number of bits per symbol
(measured by Rate R), that should be communicated over a channel, so that the input
signal can be approximately reconstructed at the receiver without exceeding an expected
distortion (D).

The rate-distortion function relate the both the parameters Rate(R) and Distortion (D). As
the distortion increases the rate decreases. (NEEDS FURTHER EXPLANATION).

17. T/F: vector quantization assigns one symbol for a group of samples [L16,p.12] - T

18. What is transform coding and what is it used for? [L16,p.16-22]

The rationale behind transform coding:

If Y is the result of the linear transform T of the input vector X in such a way that the components of
Y are much less correlated, then Y can be coded more efficiently than X. IF most of the info is
accurately described by the first few components of a Transformed vectors, then the remaining
 components can be coarsely quantized or even set zero with little signal distortion

19. What is the difference between DFT and DCT? [L16,p.44-49]

The role of the DCT is to decompose the original signal into its DC and AC components;
DC(Frequency = 0) AC(Frequency > 0)
The obvious distinction between a DCT and a DFT is that the former uses only cosine functions, while
the latter uses both cosines and sines (in the form
of complex exponentials).

20. Name two differences between DFT and DWT [L16,p.50-52]

In Fourier transform we represent a signal in terms of sinusoids. FT gives a signal that is
localized in the frequency domain. It does not give any information of the signal in the time
domain.

Wavelet transform - Basis functions are small waves located in different times. They are
obtained using scaling and translation of a scaling function and a wavelet function. Therefore WT
is localized in both time and frequency domain.

21. Which step of the JPEG compression causes the information loss? [L17]
a. Quantization
b. Run-length coding
c. Differential Pulse Code Modulation
d. Entropy coding

22. What are the 3 observations that make JPEG compression possible? [L17,p.4]
a. Observation 1: Useful image contents change relatively slowly across the image,
particularly within an 8×8 image block—spatial redundancy
b. Observation 2: Psychophysical experiments suggest that humans are much less
likely to notice the loss of very high spatial frequency components than the loss of lower
frequency components—perception boundaries.
c. Observation 3: Visual acuity (accuracy in distinguishing closely spaced lines) is
much greater for gray (“black and white”) than for color—HVS color perception.

23. Encode the sequence 57, 45, 0, 0, 0, 0, 23, 0, -30, -16, 0, 0, 1,… using RLC [L17,p.15]
(0,57); (0,45); (4,23); (1,-30); (0,-16); (2,1)…EOB

24. What is computer vision – explain difference with computer graphics? [L18,p.4]
Computer vision is:
An emulation of human understanding of images
• A sensor modality for robotics
• Computer emulation of human vision
• CV is the inverse of Computer Graphics
25. Explain the goals of computer vision [L18,p.8]
- Inference* -> computation
- Building machines that see
- Modeling biological perception
*A conclusion reached on the basis of evidence and reasoning. synonyms: deduction,
conclusion, reasoning, conjecture, speculation, guess, presumption, assumption,
supposition, reckoning, extrapolation.
26. If you were the student of prof. Minsky who was asked to link a camera to a computer
and write a program to tell what it sees, how would you go about it with what you know
today? [L18,p.16]
-

27. What is the relationship between epipolar geometry and stereo vision? [L18,p.29]
- Epipolar geometry is the geometry of stereo vision that describes the geometric
relationship between two camera systems

28. In computer vision, what are the difference between object detection,
recognition and identification? [L19,p.46]
1. Object detection: Is answering the question "Where is that object"?
2. Object recognition: Is responding to the question "What is the object in
the image".
3. Object identification: Is asking the question “Who is it?” to identify objects
in images

29. What is image segmentation and what are its goals and applications. [L19,p.6]
Image segmentation is the process of clustering or grouping similar pixels. It has
applications in Computer Vision problems like Object Detection and Object Recognition.

30. Explain three approaches to image segmentation.[L19,p7]

1. Edge based: an edge filter is applied to the image, pixels are classified as edge or
non edge, and pixels which are not separated by an edge are allocated to the
same category.
2. Region based: segmentation algorithms operate iteratively by grouping together
pixels which are neighbors and have similar values and splitting groups of pixels
which are dissimilar in value
3. Thresholding: pixels are allocated to categories according to the range of values
in which a pixel lies.

31. What is mean shift segmentation, what is the principle behind mean shift? [L19,p.11-19]
A local homogenization technique that is very useful for damping shading or tonality
differences in localized objects. It replaces each pixel with the mean of the pixels in a
range-r neighborhood and whose value is within a distance d. The principle behind it is
clustering-based segmentation? Builds on the concept of kernel density estimation?
32. True or false: a. mean shift assumes spherical clusters - True, b. mean shift finds a
variable number of modes -True, c. mean shift cannot be used on color images -False
33. What is the basis for graph-based segmentation? [L19,p.29,35]
The basis for graph-based segmentation seems to be founded in Region-based
segmentation. Graph-based segmentation relies on a weighted graph that measures
similarity or dissimilarity on the weights between nodes.
34. True or false: [L20,p.3]
a. raw data contains information, - T
b. information extracted from raw data is useless, - F
c. machine learning finds patterns in data. - T
d. data patterns enable decision making -T
35. True or false: [L20,p.8-9]
a. decision trees can be represented as rule sets, - T
 b. Statistics is about testing hypotheses, - T
c. Machine learning is about finding the right hypothesis, - T
d. Most machine learning algorithms do not use statistics - F (slide 9)
36. True or false: [L20,p.10]
a. Inductive learning aims to find a concept that fits the data, - T
b. Generalization by inductive learning cannot be posed as a search problem - F
37. Explain classification learning, and the steps to build a classifier.[L20,p.14-15]
Classification learning: learning process where
● Patterns are described as a feature vector
● In statistical classification the pattern to be recognized is classified based on the
distribution of patterns in the space spanned by pattern features
● In order to decide if the pattern has been recognized, a distance function must be
provided

Building process:

38. Explain probabilistic classifiers.[L20,p.17]

39. True or false:
a. Clustering finds categories without knowing which ones they are in advance, True
b. in clustering we maximize intra-class similarity, True
c. In clustering we minimize inter-class similarity. True
40. True or false: [L20,p.18]
a. association learning uses valid rules, T
b. the coverage and accuracy are non-conflicting measures of rule validity, F
c. In decision trees the root node contains all documents. T
41. What is the difference between a Markov process and a Hidden Markov Model?
[L20,p.20]
The hidden Markov model hides information from the user that the Markov process does
not.

42. Given a set of objects, what are the first 2 steps to classify them? [L20,p23-24]

43. What are the applications of K-NN[L20], K-means [L15,L20], and mean-shift
algorithms[L19]?
Application of KNN: Text mining
Application of K-means: Image compression
Application of Mean-shift algorithms: cluster analysis in computer vision and image
processing.

44. What is the difference between mean-shift for clustering and mean-shift for
segmentation?