An Adaptive Linguistic Model for Pattern Representation and Recognition. Direct Products of Relational Systems. Notes on a Combinational Strategy for Enumeration Problems.
Abstract
An adaptive pattern representation and recognition system for application to mechanized interpretation of pictorial data from aerial reconnaissance photographs is described. This system differs from conventional pattern recognition systems in that: It generates its own features. These features are generated in a decidable subject of the predicate calculus from which they are translated into computer programs. The formulae which define features are generated using quantifiers and logical connectives from a set of atomic predicates which correspond to information about the metric on the sampled input picture. The recognition procedure selects sub-regions of input pictures and identifies them (i.e. scene analysis). It has provision for an unusual amount of interaction with the trainer to accelerate convergence. The relation between the syntax (formula structure) and semantics (the classes of geometrical figures which satisfy these formulae) is well understood and utilized for rapid convergence. Other papers included extend the system and treat a class of combinatorial problems. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1971
- Accession Number
- AD0716856
Entities
People
- David Rothenberg
- Erik Ellentuck