An Efficient Technique for Detection and Removal of Lines with Text Stroke Crossings in Document Images

Abstract

Precise automatic reading of the characters in a document image is the functionality of Optical Character Recognition (OCR) systems. The overall recognition accuracy can be accomplished only through efficient pre-processing procedures. The recognition of characters in pre-printed document images is a highly challenging task as it desires unique pre-processing methods and it depends on the layout of document. In this paper we propose a pre-processing technique for removal of horizontal/vertical lines in the pre-printed documents. The major challenge involved in removal of the horizontal lines is retention of the pixels overlapped between line and characters in document. The proposed algorithm works in two phases; image enhancement and line detection is made in the first phase and the second phase comprises convolution process using rectangular structuring element for detection of text stroke crossings on lines which are detected in phase one. The output image is further subjected to undergo post enhancement and analysis operations using connected component analysis and area features for removal of broken/dotted line structures. The experimental outcomes achieved are quite satisfactory and consistent enough for subsequent processing of document.