Recognition of poor quality words without segmentation

A method for recognition of poor quality documents is presented. The method is based on extraction of independent and robust features of 'objects' within a word without segmenting touching objects. A number of alternatives for each object are found. A lexical lookup method is used to verify the alternatives. The method has been applied to seven different poor quality facsimile messages in order to observe its effectiveness. The facsimile messages were first processed using a commercial OCR software and only the unrecognized words were selected and processed using the developed method. Different improvement rates were observed for the facsimile messages. The improvement ranged from 21.57% to 100%. The results obtained from these facsimile messages suggest that the developed method is capable of recognizing poor quality documents whilst avoiding segmentation of touching characters.