Introduction to basic image analysis - Lecture 2 - Week 1

Question 1

How can we represent an image?

Answer 1

Image function
Landscape
Array of pixels
Image histogram

Question 2

What is an image function?

Answer 2

I from R^2 to R for a greyscale image
I(x,y) gives the intensity at position x, y

A digital image is a discrete version of this function

Question 3

What information does a histogram discard from the image?

Answer 3

Spatial information

Question 4

How are intensity values in grid images represented

Answer 4

A single byte (0=black, 255=white)

Question 5

What is an image transformation?

Answer 5

Applying a function to an image, e.g. adding 20 to every pixel value

Question 6

What is addition in image arithmetic and what does it achieve?

Answer 6

Taking the average over two images in sequence
Reduces noise

Question 7

What is subtraction in image arithmetic and what does it achieve?

Answer 7

Take the difference between two images, need to somehow shift back to [0:255], or take the absolute value

Can detect change over static backgrounds, can find objects, shadows & reflections in real-world scenarios

Question 8

What is noise?

Answer 8

Stochastic (random) process
Identical images might not have identical values, small fluctuations in pixel values

Question 9

How is the Signal to Noise ratio (SNR) calculated?

Answer 9

The max signal / sigma (standard deviation of noise)

Question 10

What is the formula for noise reduction with temporal averaging?

Answer 10

new sigma = old sigma / (sqrt(Number of images averaged))

Question 11

How is local spatial averaging conducted?

Answer 11

By doing a 3x3 convolution with all 1s as the kernel

Question 12

Why does noise conflict with scale?

Answer 12

Small structure might be considered “noise” and removed

Question 13

How is sub-sampling performed

Answer 13

Every other row and column is removed to make an image half the size

If the smallest scale is removed in noise reduction, we don’t need as much spatial resolution:
- Can subsample the image and get the same information in fewer pixels

Question 14

How can sub-sampling be made better?

Answer 14

By filtering the image (e.g. gaussian filtering), then subsampling, a better final image is obtained

Question 15

What is multi-scale processing?

Answer 15

If we noise the image first, then slowly bring back detail, face tracking works better