Comparison of Visual and Logical Character Segmentation in Tesseract OCR Language Data for Indic Writing Scripts.