A short quiz to help you see what you know about light and images

Adrian F. Clark

1. Which of the following is NOT a machine learning algorithm?

   • genetic programming
   • back propagation

   • deep learning

   • simulated annealing

   Deep learning is a generic term for a particular type of neural network (one with many hidden layers) but is not itself a ML algorithm. Conversely, multi-layer perceptron neural networks are normally trained using back propagation, while genetic programming and simulated annealing are both ML algorithms in their own rights.

2. Which of the following compression schemes is used in standard definition TV in the UK?

   • H.261
   • MPEG-4
   • H.264

   • MPEG-2

   Standard definition digital TV in the UK (aka Freeview) uses MPEG-2. HD channels on Freeview and all channels on Freesat use MPEG-4, which is a better compression algorithm but more costly in terms of hardware. H.261 and H.264 are equivalent standards for videoconferencing rather than broadcast; they have the added need for their encoder to work at video rate.

3. Which of the following is useful for comparing the similarity of a pair of images?

   • pixel-wise product

   • correlation

   • standard deviation
   • mean value

   Correlation is especially useful for comparing the similarity of a pair of images because it is independent of both differences in brightness (mean grey level) and contrast (standard deviation). However, it is NOT independent of any spatial shift between images. The other approaches listed are all pretty awful for this job.

4. What is the name of the face detection algorithm in digital cameras?

   • AdaBoost
   • Haar

   • Viola-Jones

   • FACEDET

   The algorithm that is almost universally used is due to Viola and Jones (and is patented). Haar features and AdaBoost are algorithms used in Viola-Jones, while FACEDET is something just I made up.

5. In how many different wavebands does a digital camera capture an image?

   • 3

   • 2
   • 1
   • 4

   Digital cameras record in 3 wavebands corresponding to red, green and blue light.

6. Which of the following corresponds to a 3 MP image from a digital camera?

   • 1280 x 1024 pixels

   • 2048 x 1536 pixels

   • 2562 x 2048 pixels
   • 640 x 480 pixels

   2048 x 1536 = 3,145,728 which is about 3 Mpixel.

7. In programs such as Photoshop, what is unsharp masking?

   • using a blurred mask to identify regions of an image
   • taking photographs with the lens out of focus

   • emphasising the edges in an image

   • showing faint lines in a camera's viewfinder to help focussing

   Unsharp masking is a technique used in image manipulation which involves adding a blurred version of the image to it, the aim being to enhance edges in the image (often just before printing).

8. What is chromatic aberration?

   • the difference in focus due to the wavelength of light

   • the inability to focus on highlights from bright objects
   • the curving of straight lines when photographed
   • the difference in focus due to aspherical lenses

   Chromatic aberration arises because the different wavelengths of light are deflected (diffracted) at the interface between air and glass (or plastic) by different amounts. It appears as a coloured halo around objects. Good camera lenses have coatings which reduce the effect of chromatic aberration. As an aside, mirrors do not suffer from chromatic aberration and that is one of the reasons why telescopes tend to use curved mirrors rather than lenses.

9. In photography, which of the following controls the depth of field?

   • pixel size
   • ISO rating

   • aperture

   • shutter speed

   The main thing which determines the depth of field of an image is the aperture: a wide aperture means that a small range of distances are in focus simultaneously and vice versa.

10. Roughly, what is the wavelength of visible light?

    • 5,000 nm
    • 5 nm

    • 500 nm

    • 50 nm

    Visible light has wavelengths of about 400-700 nm.