A Malayalam OCR System Using Column-Stochastic Image Matrix Approach

Indian languages especially South Indian languages have several distinct characteristics that are exploited for the development of a robust optical character recognition system (OCR). This paper addresses the problem of segmentation of printed Malayalam characters, a fairly complex task, along with their characterization through non-trivial dominant Eigen values of column-stochastic image matrices. Rectangular image matrices obtained after digitalization, segmentation and normalization are converted to column-stochastic square matrices. Non trivial dominant Eigen values of such matrices have proved to be unique for characterization of printed Malayalam characters. Further, a novel segmentation algorithm has been proposed and tested. Results and analysis presented indicate effective performance of the OCR system.