Handwritten text documents binarization and skew normalization approaches

Abstract

Handwritten text recognition has been an active research area for many years. Handwritten text recognition needs to perform some preprocessing steps for better recognition. First, we find binary image of given handwritten text document and then after performing the line segmentation task, we normalize it to the segmented lines. There are various normalization tasks such as skew normalization, slant normalization and size normalization. This paper, focuses on the handwritten document binarization and skew normalization and proposes a novel global binarization approach, which is very cost effective. We also propose a new skew normalization approach which is based on orthogonal projection of the segmented line with respect to x-axis. The method has been experimented on various styles of handwritten text documents, and it is found that it detects the exact skew angle, and corrects it efficiently. A comparative study has also been reported to provide a detailed analysis of the proposed methods together with some other existing methods in the literature.