Simulation Method of Feature Selection for Unconstrained Handprinted Characters.
Abstract
A theory of character recognition and associated methodology which is expected to lead to a machine algorithm for recognizing unconstrained handprinted characters is reviewed and expanded to encompass the previously excluded Arabic numerals. A computer character generator package developed for this and future work is described. By way of example, its use in generating the test stimuli for this work is explained. The character pair 2-Z is selected for study in this work. A systematic method of describing the essential difference between the characters is employed. Three psychophysical experiments, labeling, reaction time and goodness are described and used to determine a quantitative rule for distinguishing between 2s and Zs in neutral context. This thesis serves as a guide for initial investigation of other intercharacter boundary rules. The collection of rules should eventually describe a general character recognition algorithm. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1977
- Accession Number
- ADA054755
Entities
People
- Robert Tornow Babcock
Organizations
- Massachusetts Institute of Technology