Bruce A. Draper, Jose Bins, and Kyungim Baek
Department of Computer Science
Colorado State University
AbstractMany modern computer-vision systems are built by chaining together standard vision procedures, often in graphical programming environments such as Khoros, CVIPtools, or IUE. Typically, these procedures are selected and sequenced based on a programmer's intuition rather than sound scientific principles. This paper presents a theoretically sound method for constructing object-recognition strategies by modeling object recognition as a Markov decision process (MDP). The result is a system called ADORE (Adaptive Object Recognition) that automatically learns object-recognition strategies from training data. Experimental results are presented in which ADORE is trained to recognize five types of houses in aerial images, and where its performance can be (and is) compared to optimal.
Keywords: object recognition, reinforcement learning, Markov processes, computer vision.