CS1674: Homework 4 - Written

Due: 9/26/2016, 11:59pm

This assignment is worth 15 points. I expect it will take you 1-2 hours to complete.

  1. How are feature distinctiveness and feature repeatability at odds with each other? Why is each of those properties desirable (in moderation)?
  2. Describe the process of extracting the locations of corners from images, starting with the use of image gradients.
  3. Why do we say that the Harris corner detector is not scale-invariant?
  4. How can we compute a histogram of gradient orientations, when generating a SIFT descriptor?
  5. What is the effect of gradient magnitude on the histogram of gradient orientations, in the computation of the SIFT descriptor?
  6. Why do we have to normalize the SIFT histogram a second time, after clipping all values to be at most 0.2?