Line removal and handwritten word recognition of police accident report forms

摘要

In this paper, we present an approach for line removal and handwritten word recognition using the New York State MV- 104A Police Accident Report (PAR) form. Only limited space is available on the form for written inputs and hence many handwritten characters overlap or touch the form's preprinted lines and characters. The complex structure of the form and the touching characters make line removal and word recognition difficult. We addressed this problem using run-length coding scheme and baseline analysis. Preliminary word recognition tests show promising results for PAR form processing and filled-in data extraction and recognition.