A Framework for Multi-Domain Sketch Recognition
Christine Alvarado, Michael Oltmans, Randall Davis · 2002
Introduction We present a multi-domain architecture for sketch recognition systems that we believe will make these systems both easier to construct and more robust in operation. At the heart of our approach is a method that combines high-level, domain-specific information with low-level, domain-independent recognizers. We aim to build a recognition framework that can be applied naturally and e#ciently to a variety of domains, yet takes advantage of the power that comes from context. We are motivated by the success of using domain-knowledge for speech understanding, and inspired by the design of the Hearsay-II System (Erman et al. 1980). Hearsay-II combined knowledge at various levels of the speech interpretation process, including the syllable, word, and phrase levels, to generate and choose from multiple interpretations of a spoken utterance. We believe a similar architecture can be e#ective in sketch understanding. There are a variety of issues to be addres