Robust method for the text line detection and splitting of overlapping text in the Latin manuscripts

Abstract

The paper presents the modified method of the text lines separation in the handwritten manuscripts. Such an approach is required for the medieval text analysis, where multiple text lines overlap and are written at different angles. The proposed approach consists in dividing the bounding boxes into smaller components based on the points of the character curves intersection. The method considers the askew text lines, producing non-rectangular zones between the neighboring lines.