Knowledge representation for 2-D real-time object recognition

A knowledge representation scheme is described that is utilized to store and initiate the various classification rules in a real-time classification system. The principle is to represent the rules explicitly in a symbolic representation, and to link them in a decision network. This scheme provides a flexible tool for setting up hierarchical classification schemes. Other main features of the approach described are the capability of the network to handle uncertainty and the use of context to provide better classification accuracy. The explicit representation of the decision network is automatically compiled into low-level machine-instructions yielding a machine independent representation scheme. The performance of the method is demonstrated by examples which are taken from the problem of real-time classification of wooden boards.<>