A short quiz on intermediate-level vision

Adrian F. Clark

1. Which of the following schemes is unsuitable for matching arbitrary features in images?

   • cross-correlation
   • image differencing

   • the Laplacian

   • matched filtering

   The Laplacian emphasizes isolated points in images, and is not suited to matching any other feature in an image.

2. Which of the following is not a valid equation for a straight line?

   • the equation $x cos A + y sin A = r$

   • the equation $y^2 + x^2 = r^2$

   • the equation $x = m y - c$
   • the equation $y = c + x / m$

   The cos-and-sin expression listed is the parametric form used by the Hough transform, while the second and third expressions can be manipulated into the form $y = m x + c$, which is the normal equation of a straight line. However, the other one cannot: it is the equation of a circle of radius r.

3. What is the underlying principle of the FAST corner detector?

   • it runs quickly
   • it produces fewer false negatives than alternative algorithms
   • it produces fewer false positives than alternative algorithms

   • there is a contiguous span of more than half a circle of pixels that are either light or dark

   FAST works on the principle that more than half the pixels around a circle are all either light or dark. It does run pretty quickly (but that isn't a principle) but notoriously produces oodles of false positives.

4. In the non-parametric line Hough transform, what is the accumulator array indexed by?

   • end-points

   • gradient and intercept

   • gradient magnitude and direction
   • radius and angle

   The non-parametric Hough transform estimates the gradient and intercept directly, so we use these as the indices of the accumulator array.

5. How many thresholds are employed in hysteresis thresholding?

   • 3

   • 2

   • 1
   • 4

   Hysteresis thresholding employs a pair of thresholds.

6. The Canny edge detector represents more or less the state of the art in detecting edges in images. Why is its output still of limited use?

   • it is too slow
   • the edges are still too noisy to be useful
   • it doesn't find all edges

   • edges are of limited value in practice

   In most cases, identifying the edges of objects is useful for segmenting them from the background; but edges are not much help in (say) identifying the object. For that task, it is much better to use the corners of features in the image.

7. Why does the Canny edge detector perform non-maximum suppression?

   • to stop the edges being maxima
   • to stop the edges being too bright
   • to quantise the direction of the edge

   • to make each edge only one pixel wide

   Non-maximum suppression is intended to make each edge segment only one pixel wide.

8. Why does the first stage of Canny involve convolution with a blurring mask?

   • To reduce the effects of noise

   • To detect edges
   • To perform non-maximum suppression
   • To reduce the number of pixels

   Blurring smooths out an image, reducing the effects of noise.

9. What is the purpose of the lower threshold in hysteresis thresholding?

   • Edge segments with values below this threshold are joined to edge segments known to be correct
   • To identify dark regions in images

   • Pixels with values below this threshold are not parts of edges

   • To identify dark lines in images

   Pixels whose values lie below this threshold are rejected as edges, while pixels whose values lie above this threshold are potentially edges, depending on the values of their neighbbouring pixels.

10. How should peaks chosen from the Hough accumulator array?

    • Smallest radius first

    • Largest first

    • Smallest first
    • Nearest to origin first

    The peak with the most lines in the accumulator contributing to it is the most significant line, and this is the pixel with the largest value.