Automated Acquisition of Object Recognition Strategies for Image Exploitation

摘要

This effort attempts to solve a crucial problem of knowledge-based sceneinterpretation by building proper, more efficient, recognition strategies. The proposed system will automatically learn object recognition strategies with the goal of learning how to recognize objects from a combination of training images and a library of visual sources. This project will incorporate two types of learning techniques, Hypothesis Generation Learning, and Hypothesis Verification. Recognition graphs will represent three control strategies: an exhaustive exploration algorithm, a DNF generalization algorithm, and a graph optimization algorithm.