Brainscape's Knowledge GenomeTM

Browse over 1 million classes created by top students, professors, publishers, and experts.


See full index

COMP 558 Computer Vision > Quiz 1: Lectures 2 to 5 > Flashcards

Quiz 1: Lectures 2 to 5 Flashcards

1
Q

What does SLR camera stand for?

A

Single Lens Reflex camera

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Name a few elements that form a digital camera

A
  • lenses
  • mirrors
  • sensors
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Explain the Pinhole Camera Model

A

imagine there is a point out in the scene at 3D position (x,y,z). That point projects through a small aperture (pinhole). The 3D point then projects uniquely to a 2D position (x,y) on the sensor.

The colour that the 3D point has on the 3D space, will also appear in the 2D point.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

What is a pixel?

A

A pixel is a picture element.

It is formed by 3 8bits. Each 8bit value ranges from 0 to 255 and is associated with the RGB values.

In total, a pixel is 24 bits or 3 bytes.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

What is a megapixel?

A

1 million pixels

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

Sometimes we talk about the definition of the pictures taken by a camera, what does 720p mean?

A

The size of an image 720p is more or less 1 megapixel. This image is a grid with 720 rows and 1280 columns.

p stands for “progressive scan”. (Compared to the “interlaced” method that was used at the beginning.)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Sometimes we talk about the definition of the pictures taken by a camera, what does 1080p mean?

A

The size of an image 1080p is more or less 2 megapixel. This image is a grid with 1080 rows and 2000 columns. It is considered HD (high definition).

p stands for “progressive scan”. (Compared to the “interlaced” method that was used at the beginning.)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

RGB values

A

An RGB value ranges from 0 to 255.

It is a way of ordering “intensities” where 0 is the darkest and 255 the brightest.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Observe the image below, can you find the RGB values for the colours black, white, red, green, blue, cyan, magenta and yellow?

A
  • Black (0,0,0)
  • White (255,255,255)
  • Red (255, 0, 0)
  • Green (0, 255, 0)
  • Blue (0, 0, 255)
  • Cyan (0, 255, 255)
  • Magenta (255, 0, 255)
  • Yellow (255, 255, 0)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

What does it mean for an image to be saturated?

A

Images whose colours are very close to the colours in the corners of the following cube.

(Colors are close to black, white, red, green, blue, cyan, magenta, yellow)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

Light

A

A beam of white light consists of superimposed beams of “coloured” light.

Light consists of electromagnetic waves from 400 to 700 nm.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Spectrum

A

How much light is emitted/absorbed/transmitted/reflected at each wave length.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

Emission Spectrum

A

Distribution of the power of the light emitted by a light source at each wavelength

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

Absorption Spectrum

A

indicates fraction of the light that arrives at a surface or volume and is absorbed as a function of wavelength.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Transmission Spectrum

A

Indicated the fraction of the light that is reflected back in function of wavelength

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Sensors in a camera

A
  • The sensors in a camera form a 2D array
  • Every grey square is basically colour blind
  • To detect colour, there is a colour filter before the sensor.
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
17
Q

Bayer Pattern

A

The pattern of the filters placed before the sensors is called Bayer pattern. Noticed that in this pattern, every 2x2 sensor filter there is twice as many green than red and blue filters.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
18
Q

How do we call the filter array for the sensors?

A

CFA: colour filter array

It is responsible for the colour sensitivity of the sensors.

Each coloured filter allows certain wavelengths to pass which is then transmitted to the sensor.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
19
Q

Exposure time

A

Modifies the amount of light absorbed by the sensor.

The longer the exposure, more light is absorbed and so the higher the intensity.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
20
Q

Can we modify the sensitivity of the sensor in a camera?

A

When the sensitivity of a sensor is modified, this is called gain or ISO.

This could help when you are shooting in dim conditions for example. A higher gain will allow you to get better intensity of the colours.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
21
Q

Aperture

A

The aperture is the space through which light passes to reach the sensor.

It is through this space that the image is projected into the sensor.

The aperture size can be modified. The bigger it is, the better the colour intensity.

22
Q

What affect the RGB values of an image?

A
  • exposure time
  • aperture
  • gain
23
Q

What creates noise in the pictures?

A

Noise in the pictures might come from random photons emitted by the light source. Notice, this variation can be significant when the light levels are low.

Another possible source of noise is the sensor.

24
Q

Analog-to-Digital Conversion

A

The signal (including noise) is digitized, namely quantized and then coded in binary.

A single sensor (which corresponds to a pixel) measures some amount of light energy and maps it to a real number intensity. This real number is then digitized (a number between 0 and 255) and stored as a 12-bit number.

25
Q

RAW Image Format

A

RAW image format is an image that keeps the 12-bit value obtained during the Analog-to-digital conversion.

This means that the details of a RAW file are not standard and thus are proprietary.

26
Q

JPEG, JPG

A

Compressed images that do not keep the 12-bit value but instead an 8-bit value.

27
Q

Demosaicing

A

Notice that because of the Bayer Pattern, there are colour gaps in our image. To solve this we untangle the RGB values of an image an attempt to create 3 images: one red, one green and one blue.

28
Q

White balancing ?

A

???

29
Q

Gamma Correction

A

To display the RGB values to a TV or a screen, the RGB values have to be raised to the power of (1/Gamma) (a number closer to 2).

This is because the intensity for the spectrum emitted at each display pixel is surprisingly not proportional to the RGB values.

On the other hand, the intensity is proportional to the RGB values raised to Gamma.

R^(1/gamma) G^(1/gamma) B^(1/gamma)

30
Q

RGB histograms

A

Shows the frequencies (number of times some colour appear) of each R, G and B values in an image.

There is also a histogram for an average of RGB at each point which is called luminance.

luminance = 0.3 * R + 0.6 * G + 0.1 * B

31
Q

Image Noise Example 1

A
32
Q

Image Noise Example 2

A
33
Q

Draw the RGB image pipeline

A
34
Q

Match the Terminology with the Topics

(Ignore the colour cube)

A
  • Analogue to digital (2)
  • Aperture (1), gain (2), exposure time (1,2)
  • Bayer Pattern (2)
  • Colour histogram (3)
  • Demosaicing (2,3)
  • Gamma Correction (3)
  • JPG (3)
  • Noise (1, 2)
  • RAW image (2,3)
  • Spectra (emitted, reflected, transmitted) (1, 2)
  • sRGB (3)
  • White balance (3)
35
Q

sRGB ?

A

?

36
Q

Lecture 3 todo

A
37
Q

Lecture 4 todo

A
38
Q

Problem: Fitting Points to a line

A

Suppose that you have a set of noisy points {(x_i, y_i)} that roughly fall along a line in 2D plane (x,y). You would like to fing the equation of that line.

To do this you can use two methods:

  • Linear Regression
  • Total Least Squares
39
Q

Linear Regression

A

You have two data variables x & y and n_i which represents the noise. You want to find a linear relationship between samples (x_i, y_i).

Notice that the noise, is only considered on y-axis. In other words, the noise is the vertical distance from the point to the line. Thus, we a linear represnetation of the line is:

y_i = mx_i + c + n_i

We want to find m and c such as the sum_i n_i^2 is minimized namely such as they minimize:
sum_i (y_i - mx_i -c)^2

40
Q

Linear Regression:

How do you find the m and c such as it minimizes:

A
  • Take the partial derivatives wrt m and c
  • Set them to 0
  • (See notes of September 21st)
  • You then get two equations and two unknowns: m and c
  • You can now solve for m and c
41
Q

Linear Regression:

Write the problem in the images using matrices.

A
42
Q

Linear Regression:

Re state the linear regression problem in terms of matrices.

A

Given a matrix A with m>n and a non-zero m-vector b, find the values of vector u that minimizes:

|| Au - b||_2 where || * ||_2 = L_2 norm

Notice: Minimizing L_2 norm is equivalent to minimizing the sum of squares of the elements of the vector Au - b. In other words, L_2 norm is just the square root of the sum of squares of the elements of Au - b

43
Q

Linear Regression:

By using the matrix representation of the problem, derive the general solution.

A
44
Q

Linear Regression:

For the matrix representation of the problem, state how you solve for u

A
45
Q

Pseudoinverse

A

(A^T A)^(-1) A^T

46
Q

Total Least Squares

A

We have a set of points in a 2D plance. We want to fir a line to the set of points.

In this case, the noise or error is the distance between the point and the line. (This distance is perpendicular to the line).

The perpendicular distance from a point to the line is:

|x_i cos(teta) + y_i sin(teta) - r|

where r = xcos(teta) + ysin(teta)

So we try to minimize

sum_i (x_i cos(teta) + y_i sin(teta) - r)^2

47
Q

Total Least Squares:

How do we minimize this:

A

Notice that sin and cos are not linear so we can’t just simply solve for them.

Replace cos(teta) and sin(teta) by a and b respectively.

Now, find values for a, b and r such that the sum is minimized.

To do so:

  1. Find the derivative of the sum with a and b wrt r and set it to 0

Notice, we get the mean values for x and y which means that it lies on the line such as: ax + by = r

Substitute r into the sum which gives us the image

48
Q

Total Least Squares:

Write the problem using matrices

A

[a b] A^T A [a b]^T

Notice that:

a = cos(teta)

b = sin(teta)

49
Q

Total Least Squares:

How do you minimize?

A
  1. Comput the eigenvectors and eigenvalues of A^T A
  2. Take the unit eigenvector which has the smallest eigenvalue
  3. Take teta such that [cos(teta), sin(teta)]^T is this eigenvenctor
50
Q

Read Vanishing points detection (Lecture 5)

A
51
Q

True or False

A surface reflectance spectrum is more saturated if it reflects light only in a small range of wavelengths.

A

False

Actually, if a surface reflects over a small range of wavelengths, this means that it will reflect little light overall.

In order for a surface to reflect saturated colour, the surface must reflect longer wavelengths and this over a wider range of wavelengths.

For example, if it a surface reflects a 100% of wavelengths in the range 560 nm to 700 nm, this surfce would appear as saturated.

COMP 558 Computer Vision (5 decks)

Brainscape helps you reach your goals faster, through stronger study habits.
© 2024 Bold Learning Solutions. Terms and Conditions