Researchers are increasingly using digital images to teach machines how to see and understand our physical world. While conventional wisdom states that a picture is worth a thousand words and seeing is believing, it is now important to ask what a picture is really worth and how much trust we can place in vision alone. This thesis has focused on the advancement of these computer vision systems. The algorithms produced by this research have established new levels of efficiency for retrieving and comparing visual data, as well as defining new machine learning techniques that extend our current capacity for automating complex visual tasks.