A reasonably efficient and inexpensive (Optical Character Recognition System) OCR are commercially available today to recognize printed text in widely used languages/scripts. The present work is an attempt to use EM (Expectation Maximization) Algorithm for building and incremental training OCR model without overlooking prior knowledge. The aim of proposed research is to implement a generic, robust & high accuracy character recognizer using a single technique (EM Algorithm) that can exploit the inherent characteristics of Telugu. The impact of proposed method (EMTrain++ algorithm) ---- is demonstrated on two languages (English and Telugu).
