RECOGNITION-BASED SEGMENTATION OF ONLINE RUN-ON HANDPRINTED WORDS - INPUT VS OUTPUT SEGMENTATION

The performance of two methods for recognition-based segmentation of s trings of on-line handprinted capital Latin characters is reported. Th e input strings consist of a time-ordered sequence of X, Y coordinates , punctuated by pen-lifts. The methods are designed to work in ''run-o n mode'' where there is no constraint on the spacing between character s. While both methods use a neural network recognition engine and a gr aph-algorithmic post-processor, their approaches to segmentation are q uite different. The first method, which we call INSEG (for input segme ntation), uses a combination of heuristics to identify particular pen- lifts as tentative segmentation points. The second method, which we ca ll OUTSEG (for output segmentation), relies on the empirically trained recognition engine for both recognizing characters and identifying re levant segmentation points. The best results are obtained with the INS EG method: 11% error on handprinted words from an 80,000 word dictiona ry.