GRAPHICAL-DATA-PROCESSING RESEARCH STUDY AND EXPERIMENTAL INVESTIGATION
Abstract
The report describes the continuing development of preprocessing, classification, and context analysis techniques for hand-printed text, which are advancing at an accelerating pace. Experiments were continued with the Piecewise-Linear learning machine, using the outputs of two preprocessors: the PREP 24A simulation of the 1024-image optical preprocessor, and the CALMMASK preprocessor, which employs both edge-detecting and corner-detecting masks. A new low test error rate for classification was achieved on hand-printed alphabets of FORTRAN characters. Statistics of the performance of the learning machine during a single testing iteration are presented, and shed light on several important questions, such as the distribution of rankings of the desired character category when it is not in first place. A discussion of the preprocessing methods used in the topological approach to preprocessing and classification is begun. The initial development of a FORTRAN syntax analyzer is described. A milestone was reached with the passage of a small sample of actual FORTRAN text from a coding sheet through the scanning, preprocessing, classification, and syntax-analysis programs.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1967
- Accession Number
- AD0650926
Entities
People
- J. H. Munson
- P. E. Hart
- R. O. Duda
Organizations
- SRI International