Background subtraction

19 important questions on Background subtraction

What two properties do we expect/assume a background to have?

It is static, and its colors are different that the foreground.

Four challenges of background subtraction based on reference model (background image)

  1. Movement
  2. Shadows
  3. Color variations/overlap
  4. Aliasing

How is color variations a challenge in background subtraction?

Colors in foreground are similar to the background
The model cannot distinguish foreground from background
  • Higher grades + faster learning
  • Never study anything twice
  • 100% sure, 100% understanding
Discover Study Smart

How is color overlap a challenge in background separation?

Foreground casts shadows/reflections on background
The model cannot distinguish side-effects from foreground

Name a challenge of outdoor background separation

Change in time changes colors in background
Model incorrectly assumes foreground appears yet it is just that light is changed

What is a challenge of background separation of a sailing boat?

Water moves, dynamic environmental changes.

What is image aliasing?

Averaging pixel values in an image

How can image aliasing be a problem for background subtraction?

Per frame pixel values can be averaged into a different color.
These changes might be slight, yet as well large near edges of highly discrepanzing pixels.
This inconsistently resolves in incorrect separation with naïve separation.

What is the reference image in a background subtraction task?

A static image of the background without the foreground.

How does the naïve background subtraction technique work?

Compute L1 distance between pixel value and reference image.
Label the pixel foreground in case a threshold delta is succeeded/surpassed?

What are the two basic techniques for noise reduction in background subtraction?

Erosion and dilation

How does erosion work?

All 9 neighbors of an active pixel are checked on being activated
If the number of active neighbors is heigher than a threshold, the pixel is maintained active.

When to apply erosion in a background subtraction task?

To improve an initial background subtraction that contains outliers that should not be part of the resulting silhouette

How does dilation work?

For every inactive pixel count the number of active neighboring pixels.
If a certain threshold of active neighbors is surpassed, activate that pixel

How to adapt (improve) for slightly changing background colors?

Increase the allowed deviation in color space
Use a Gaussian (normal) distribution with larger standard deviation.

In background subtraction, why do pixels of the reference image near edges have a higher deviation?

Anti-aliasing

In background subtraction, why do pixels of the reference image that are brighter have a larger standard deviation?

They vary more thus larger deviation.
(Yet I assume in consideration of Hue for HSV, or CIElab this will not be the case, rather inverse???...)

What is a Gaussian Mixture Model?

Classifying/Labeling a pixel based on whether it is at least a certain number of standard devations away from multiple component models. This combination is the mixture.

Why use a mixture of Gaussian models for background subtraction?

Multiple sources can change the projected light at a certain position over time. Think for instance of shadows, traffic lights, shadows

The question on the page originate from the summary of the following study material:

  • A unique study and practice tool
  • Never study anything twice again
  • Get the grades you hope for
  • 100% sure, 100% understanding
Remember faster, study better. Scientifically proven.
Trustpilot Logo