A multistrategy machine learning approach to object recognition

This work describes an innovative approach that combines machine learning and vision into an integrated system. The system is called ORACLE: Object Recognition Accomplished through Consolidation Learning Expertise. It uses two machine learning techniques known as explanation-based learning and conceptual clustering, combined in a synergistic manner, which provide effective object recognition. The learning system can automatically acquire new object models when unknown objects of interest are encountered in the environment. In addition, the object models are dynamic and qualitative in nature so they can be refined by the learning techniques during the recognition-learning cycle of the ORACLE system. Finally, the ORACLE system employs knowledge-based reasoning to effectively handle problems such as imprecise segmentation, image noise, and occlusion.