CS1674: Homework 5 - Written

Due: 10/3/2016, 11:59pm

This assignment is worth 15 points.

  1. When matching features across two images, why does it make sense to use the ratio: distance to best match / distance to second best match, as a way to judge if we have found a good match?
  2. How do we use clustering to compute a bag-of-words image representation? Describe the process.
  3. How can we find to which cluster we should assign a new feature, which was not part of the set of features used to compute the clustering?
  4. When is it more efficient to create an inverted file index to match a query image to other images in the database, rather than comparing the query to all database images without an index?
  5. Why do we need to measure both precision and recall in order to score the quality of retrieved results?
  6. Please write pseudocode for the computeBOWRepr function from HW5P.