A novel document image segmentation method using medial axis transform

Abstract

A novel document image segmentation technique based on medial axis transform (MAT) proposed. The reconstruction ability of the MAT and the distance values of medial axes are utilized in the method. Firstly, the MAT is applied both to the foreground and to the background of an input document. After eliminating the unwanted medial axes of the background with some rules defined in this study, the components of the document can be grouped into homogeneous regions by reconstructing the background using the remaining medial axes. After the main regions of the input document are extracted, the distance values of the medial axes are used as features to classify text and graphic regions. Experimental results on a variety of binary and color documents indicate that the proposed method is robust to documents with any fonts, languages, and orientations, and irregular regions.