Easy ones (screeners) in the context of image / object recognition:

* What is the difference between exact matching, search and 
  classification?

* What is the difference between global and local descriptors?

* What properties make an image / object easy to recognize?

* What are the advantages and drawbacks of deep learned features?

* What are the advantages and drawbacks of binary descriptors?

Harder ones IMO, more open:

* What do you think is the minimum size of an image signature database 
  that could be used to recognize the covers of all CDs sold on Amazon?

* When is sight not the best sense to use to make sense of a situation?

* Is search by music closer to text search or search by image? Why?

More practical ones:

* Do you know Brewer's concepts of "harvest" and "yield"? How do they 
  apply to image search?

* How do you quantize features in large dimensions? Why do you have to?

* What ANN algorithms do you know?

* How do you tune a KD-forest?

* Can you build an image search engine on top of something like Elastic 
  Search? Advantages / drawbacks?