Configurable text stamp identification tool with application of fuzzy logic

Abstract

This paper presents our work on locating and removing unwanted text stamps within archive documents which are being prepared for OCR. Text stamps mainly comprise one or several text lines with a fixed shape, font size and colour, and may appear anywhere on the document with variable orientation and overlap of other text fields. We apply a configurable user interface to register features of a sample stamp (such as corners, font-size and print colour) as a template using fuzzy rules, and then analyse each document image to find matching stamps using fuzzy functions as a classification mechanism. The configurable interface allows the user to decide which and how many features should be used to describe the target stamp. Evaluation was very encouraging. We tested 1,241 specimen index cards from a biological archive card index, and achieved 92-95% correct detection rate and 85-95% complete removal rate. © Springer-Verlag 2004.