Abstract: Optical character recognition strategies are concentrated towards improving recognition efficiencies by adapting post processing techniques. OCR errors of complex scripts are due to improper segmentation as well as inappropriate rendering. Topological features of a script as a global knowledge and geometrical features of isolated patterns as local knowledge are combined together while segmenting the basic unit,syllable. A multilevel segmentation was implemented to adopt the idea of human visual perception capability into the proposed model to use the distinct features exhibited by Devanagari script. The proposed method is based on implementing cognitive approach in segmentation phase by dealing with the syllable as a meaningful unit of information right from segmentation phase itself instead of isolated pattern on document images of Devanagari script.