Knowledge-Based Chemical Symbol Recognition System

The ultimate goal is to develop and commercialize a functional, low-cost system for machine reading and recognition of chemical structure diagrams. The specific Phase I goal is to demonstrate the feasibility of developing such a system by using a knowledge-based approach which provides a logical method for expanding the chemical symbol recognition domain and for handling exceptions to recognition rules. Prior work on system design, algorithm development and literature research supports our contention that developing a useable chemical reader is feasible. The chemical reader will find application as an intelligent input device for entry of chemical structure data into chemical information systems. Yearly, thousands of structures must be entered into large databases by specially trained persons. In addition, searches of these databases by chemistry researchers often require input of structural information. The chemical reader will enable entry of structural information via universally recognized diagrams instead of using system specific codes. We have described a knowledge-based approach to developing the chemical reader. This approach has been successfully used by others in systems designed to recognize line drawings. In the chemical reader, system a knowledge base will be built which contains rules incorporating knowledge of chemistry and diagramming conventions. The rules will use facts obtained from feature extractors operating on pattern primitives derived from 2D gray level images of structure diagrams. The images can be acquired using either optical scanners or CCD camera/frame grabber devices. The result of processing the structure diagram will be a connection table, the standard representation for chemical structures.