Rule-Based, Probabilistic, Symbolic Target Classification by Object Segmentation
David P. Casasent, Abhijit Mahalanobis · 1987
Optical symbolic processing applications in pattern recognition rather than logic operations [1,2] are considered in this paper. The database we employ is summarized in Section 2. Optical correlators represent one of the most powerful functions possible and preferable for realization on optical systems. We thus retain this architecture as the fundamental level-one symbolic processor to be used [2,3]. We utilize the attractive aspects of distortion-invariant iconic optical matched spatial filter (MSF) filters [4] in this work. We increase the flexibility, capacity and performance of such filters by using segments of the input object as separate filters (Section 3). The correlation outputs for these object sections represent the symbolic description of the input to be processed. A hierarchical set of rules for symbolic output processing and substitution is then employed (Section 4). The symbolic substitution used, the interactive expert system nature of the rule design, and the confidence of each rule are then detailed (Section 4). Tests of the system are then presented (Section 5).