Classification of printed text on raster documents

Abstract

When highlighting the logical structure of documents, a number of properties are used, one of which is the bold style of text words. In documents, headings, defined words, and column names in tables are often highlighted in bold. This paper proposes a method for classifying text by boldness, which consists of a sequence of steps. The first step is binarization of the entire image. The purpose of this step is to separate the image pixels into text and background pixels. The second step is to evaluate each word. The result is returned a value characterizing the thickness of the main stroke of the character in the given word. At the last step, the ratings are clustered into two clusters: bold text and regular. The proposed method was implemented and tested on three data sets, and the source code was published in an open repository.