Instructions
4 questions.
Write code where appropriate.
Feel free to include images or equations.
Please use only the space provided and keep the page breaks. Please do not make new pages, nor remove pages. The document is a template to help grading.
If you really need extra space, please use new pages at the end of the document and refer us to it in your answers.
Questions
Q1: Imagine we were tasked with designing a feature point which could match all of the following three pairs of images. Which real world phenomena and camera effects might cause us problems? Use the OpenCV function cornerHarris to investigate.
RISHLibrary — Chase — LaddObservatory
Q2: In designing our feature point, what characteristics might we wish it to have? Describe the fundamental trade-off between feature point invariance and discriminative power. How should we design for this trade-off?
Q3: In the Harris corner detector, what do the eigenvalues of the ‘M’ second moment matrix represent? Discuss both how they relate to image intensity and how we can interpret them geometrically.
Q4: Explain the difference between the Euclidean distance and the cosine similarity metrics between descriptors. What might their geometric interpretations reveal about when each should be used? Given a distance metric, what is a good method for feature descriptor matching and why?
