Character segmentation using visual interword constraints in a text page

Character segmentation is a critical preprocessing step for text recognition. In this paper a method is presented that utilizes visual inter-word constraints available in a text image to split word images into smaller image pieces. This method is applicable to machine-printed texts in which the same spacing is always used between identical pairs of characters. The visual inter-word constraints considered here include information about whether a word image is a sub-image of another word image. For example, given two word images A and B, which are "mathematical" and "the". If the short word image B is found to be a sub-image of the long word image A, the longer image A is split into three pieces, A 1 , A 2 and A 3 , where A 2 matches B, A 1 corresponds to "ma", and A 3 corresponds to "matical". The image piece A 1 can be further used to split A 3 into two parts, "ma" and "tical". This method is based purely on image processing using the visual context in a text page. No recognition is involved.