Automatic courtesy amount recognition for Indian banks’ checks

Preprocessing is an important step for automatic check processing in Indian scenario where there is huge variation in writing style, especially the way in which Courtesy Amount is terminated and the fractional (Paisa) amount is written. Courtesy Amount Recognition (CAR) and Legal Amount Recognition (LAR) form the core of automated check processing system. For CAR identification, a number of approaches have been suggested. However, most of these do not provide strong and comprehensive preprocessing techniques. We propose an algorithm that automatically detects the courtesy amount region, preprocesses this region, segments the courtesy amount into individual characters before feeding it to an ICR engine. For detecting the courtesy-amount region, a Most Probable Region (MPR) is detected, based on configurable rules and semantic analysis. The presented algorithm intelligently removes currency symbols (dasiaRs.psila), terminal characters (dasia/psila, dasia=psila, dasia/-psila), and delimiters (dasia,psila, dasia.psila) with a high degree of accuracy. Results show that the ICR rates have increased from 51% to 90% using our algorithm.