Method and evaluation of character stroke preservation on handprint recognition
Michael D. Garris · 1995
A new technique for intelligent form removal has been developed along with a new method for evaluating its impact on optical character recognition.The form removal technique automatically detects the dominant lines in an image and erases them while preserving as much of the overlapping character strokes as possible.This method of form removal relaxes the recognition system's dependence on rigid form design, printing, and reproduction by automatically detecting and removing some of the physical structures (lines) on the form.The line detection and removal technique operates on loosely defined zones in which no image deskewing is performed.The technique was tested on a large number of randomly-ordered handprinted lowercase alphabet fields, as these letters (especially those with descenders) frequently touch and extend through the line along which they are written.It is shown that intelligent form removal can improve lowercase recogrtition by as much as 3%, but this net increase in performance is insufficient to understand the impact on the recognition.There is expected to be trade-offs with the introduction of any new teclmique into a complex recogrtition system.A new statistical analysis was designed to evaluate the impact of intelligent line removal on optical character recognition.This evaluation method compares the statistical distributions of individual confusion pairs between two systems and automatically determines the significant improvements and the significant losses in performance.In order for system developers to continue to reduce error rates, sophisticated analyses like this become necessary to understand the real impact a modification has on recogrtition performance.For example, this method of evaluation should be very useful in squeezing higher performances out of voting systems.The statistical analysis pre- sented in this paper was used to evaluate the new line removal technique and the results are reported.