Entropy-Based Approach for Enabling Text Line Segmentation in Handwritten Documents

Determining text and non-text regions in an unconstrained handwritten document image is a challenging task. In this article, we propose a novel approach based on entropy for enabling the text line segmentation. A document image is divided into multiple blocks and entropy is calculated for each block. Entropy would be higher in the text region when compared to that of non-text region. Separator points are introduced accordingly to separate text from non-text part. Further correspondence between these separators would enable text line segmentation. The proposed algorithm works with an order of O (m x n) in worst case and requires less buffer space, since it is based on unsupervised learning. Benchmark ICDAR-13 dataset is used for experimentation and accuracy is reported.