Text Line Segmentation on Myanmar Handwritten Documents Using Directional Gaussian Filter

Text line segmentation is a crucial step in document processing. Although text line segmentation techniques for handwritten documents in other languages (English, Chinese, Arabic) have developed, there is still room for improving text line segmentation in Myanmar handwritten documents. It is a challenging task to segment text lines in handwritten documents due to variations in handwriting styles, irregular spacing between lines and characters, and touching characters between lines. A method for Myanmar handwritten documents proposes to overcome these problems. The proposed system extracts bounding boxes of characters from text images by applying an anisotropic Gaussian method and connected component analysis (CCA). To extract correcting text lines by touching characters between lines, text lines are extracted by using proposed algorithm based on adaptive blob scale. To evaluate the proposed method, Myanmar handwritten documents dataset is collected from undergraduate students and high school students. According to the experimental results, the proposed technique achieves 97.5% accuracy rate with correct lines.