Exam - 2018 Flashcards
What is an image histogram?
An image histogram is a plot of the relative frequency of occurrence of each of the
permitted pixel values in the image, against the values themselves. For a grayscale
image, the histogram can be constructed by simply counting how many times each
gray value, a number between 0 and 255, occurs within the image.
The image
histogram is actually a bar graph, where on the x-axis, one represents the admissible
range of intensity values (e.g., from 0 (black) to 255 (white)), and on the y-axis, we
show the number of times each value actually occurs.
What is a moving average filter, how does it work and where do we need
it? [6p]
A moving average filters is used for suppression of random noise in one-dimensional
time-varying signals.
A very simple way to eliminate this random noise is to average a few consecutive
measurements in time domain. This is the principle of a moving (or running) average
filter. The approach works surprisingly well, based on the assumption that if we
measure many times the same quantity, the useful signal will tend to accumulate,
whereas the noise, being random, will tend to cancel itself.
The moving average filter calculates each point in the output signal, y[i], by averaging
M consecutive input samples.
(how does it work) → Temperatures measured whilst the temperature in the
room isn’t changing.
We apply a 3-point moving average filter (M = 3) to this vector at index 3.
Y[3] = 1/m * (X[3] + X[4] + X[5]) = ⅓ * (18 + 18.2 + 17.8) = ⅓ * 54 = 18
The parameter in this method is M, the number of measurements in the average,
called history length. The level of noise reduction is equal to the square root of this
number [3].
For example, a 100-point moving average filter reduces the noise by a
factor of 10. Intuition would say that the higher the M, the better the filtering effect.
However, as M increases, the noise becomes lower, but the signal starts to get
deformed.
In musical notation, each symbol is a note, that has to be played as a
sound with a certain frequency and duration. Imagine that we want to
build a system that reads this musical notation using a camera, and plays
the recognized notes accordingly.
What classification solution do you suggest for the reading of the notes?
Explain which features and what type of classifier would you use. [6p]
Here more answers are possible. Template matching using cropping every
note and comparing with a database of all notes, or blob detection and a
rule based classifier using as features the coordinates of the BLOBs, or
edge detection, or neural networks with the 7 invariant moments of Hu as
input features.
Explain the essence of the Bayes theorem and give an example where we
can we use it in classification. [6p]
Its essence is that it provides a mathematical rule, explaining how our existing beliefs,
denoted by P(B), should change in the light of new evidence, denoted by A.
Usually the probabilities P(A|B), P(B) and P(A) are known. P(B|A) is not known,
and can be calculated by using this Bayes rule.
For example, a naïve Bayes classifier can be used to classify faces based on
their features, such as eyebrow thickness, vertical position at the eye center
position and arches, nose vertical position and width, mouse vertical position,
width, height of upper and lower lips, the shape of the chin, face width at nose
position and halfway between nose and eyes. We need to calculate P(It is mr.
X|features).
Bayes theorem is applied. and P(features|it is mr X) P(it is mr. X) and P(features) are known from observations.
This ROC graph shows the results of testing three classifiers A, B and C.
What can you say about these three classifiers? Which one is the best?
Which one is the second best? [3p]
we can say that classifier A performed
better than classifiers B and C on the same data set.
However, one
cannot say anything about which one is better, B and C. Classifier C has a higher true positive rate, but it also has a higher false positive rate, and
we don’t know what is better; this depends only on the relative
importance the user gives to both metrics.
Imagine that you have to design a neonatal incubator, which is a rigid box-like
enclosure, in which a baby can be kept in a controlled environment for
observation and care. The goal is to keep the air temperature inside the
chamber between 35 and 36 degrees Celsius.
a) Enumerate at least four stakeholders in this project. [3p]
the baby, the family, the developers, the hospitals and doctors, the regulators,
the insurance companies
Imagine that you have to design a neonatal incubator, which is a rigid box-like
enclosure, in which a baby can be kept in a controlled environment for
observation and care. The goal is to keep the air temperature inside the
chamber between 35 and 36 degrees Celsius.
Specify two good requirements. Specify one bad requirement. [3p]
Many answers are possible. Important is to formulate correct, testable
requirements. For example: The system shall keep the temperature inside between 35
and 36 degrees.
A bad requirement is: The system shall be easy to use. This is difficult to test.
Imagine that you have to design a neonatal incubator, which is a rigid box-like
enclosure, in which a baby can be kept in a controlled environment for
observation and care. The goal is to keep the air temperature inside the
chamber between 35 and 36 degrees Celsius.
Generate a set of test cases using Boundary Value Analysis to test the
functionality of the system. [6p]
The valid range is [35-36]. BVA requires to test for inputs from the boundaries
and in the middle of the interval.
