Enhancing Tibetan OCR with Fonts Created from Tibetan Pecha

[tenkal]

Enhancing Tibetan OCR with Fonts Created from Tibetan Pecha

Introduction

As the demand for digitizing Tibetan texts grows, the need for efficient Optical Character Recognition (OCR) tools tailored for the Tibetan language becomes ever more crucial. Tibetan script, particularly the intricate glyphs found in Tibetan pecha manuscripts, presents unique challenges for existing OCR systems. These systems often struggle with the variety of fonts, complex ligatures, and inconsistent glyph spacing present in historical manuscripts.

To overcome these obstacles, synthetic data that accurately captures the diverse styles of Tibetan script is essential. By extracting and reconstructing glyphs from Tibetan pecha manuscripts, we can create custom fonts that reflect the subtleties of these traditional scripts. This synthetic data will not only enhance OCR accuracy but also help preserve Tibetan cultural heritage in the digital age.

Approach to Glyph Cropping and Font Creation

Creating a high-quality Tibetan font from Pecha manuscripts involves several stages, including manual annotation, automated cropping, and font creation. This approach enables us to generate a functional and versatile font that can improve OCR systems and meet various typographical needs.

Glyph Cropping and Manual Annotation

The first step in building the Tibetan font is manually cropping the glyphs. This process begins with the selection of suitable glyph images from Pecha manuscripts.

• Glyph Selection: The required Tibetan glyphs are identified, and images from published Pecha texts containing these glyphs are downloaded.

• Initial Glyph Detection: OCR tools such as Google OCR are used to detect the Tibetan glyphs in these images, providing bounding boxes for each character.

• Variants Collection: Multiple variants of each glyph—typically around 100 per character—are collected to ensure diversity in the representation of each glyph.

• Glyph Extraction and Quality Control: The glyphs are cropped using the bounding boxes, and the best-quality images are selected. These images are then uploaded to GitHub for further review.

• Manual Annotation with Prodigy: The selected glyph images are uploaded to Amazon S3 for further processing with Prodigy. Here, annotators manually annotate the bounding boxes and baseline coordinates for each glyph. This is crucial because the baseline information helps determine the Left Side Bearing (LSB) and Right Side Bearing (RSB) of each glyph.

• Data Finalization: After annotation, the glyph images are cleaned to remove any background noise, ensuring that each glyph is clearly represented and ready for training purposes.

This process results in a curated collection of high-quality glyphs, ready to be used for font creation.

Automated Font Creation from Tibetan Glyphs

Once the manual annotation is completed, we proceed to automate the process of glyph cropping, which speeds up font creation and ensures consistency across large sets of glyphs.

• Glyph Isolation: Using an automated cropping model, glyphs are extracted from high-resolution Pecha manuscript images, and each Tibetan character is saved as a high-quality PNG image.
• Vectorization: These PNG glyphs are then converted into SVG (Scalable Vector Graphics) format. Vectorization is essential because it allows the glyphs to be scaled without losing quality, which is vital when creating a font that may need to be resized for various applications.
• Font Compilation: The vectorized SVG glyphs are compiled into a font file, making them ready for use across any digital platform.
  
  

Creating a Tibetan Font from SVG

Transforming SVG glyphs into a fully functional Tibetan font requires careful attention to the unique features of the Tibetan script. This Python-based solution uses several powerful libraries to facilitate the conversion of SVG files into a complete Tibetan font.

Technical Overview

The process leverages the following Python libraries:

• FontTools: A comprehensive library for working with TTF (TrueType Font) files.
• svg.path: A library designed to interpret and manipulate SVG path data.
• xml.etree.ElementTree: A tool for parsing and managing SVG files.

Key Components

• Unicode and Glyph Naming: Properly mapping Unicode codepoints to glyph names is essential for consistency across systems. The Unicode information is extracted from the filenames of the glyphs, ensuring that each Tibetan character is mapped correctly, making the font compatible with standard Tibetan Unicode representations.
• SVG Path Conversion: The conversion of SVG paths into TrueType-compatible outlines is a crucial step to ensure that the font will render properly. This process uses the SVGPen class, which interprets SVG commands (such as move, line, and curve) and translates them into TrueType font outlines.
  • Path Command Management: The class converts SVG path commands into TrueType contours.
  • Bounding Box Calculations: It ensures that each glyph is properly adjusted to fit within its bounds, preventing clipping during rendering.
  • Outline Conversion: The SVG path data is converted into outlines that can be integrated into a TrueType font file.

This conversion ensures that even complex Tibetan glyphs—often containing intricate curves and details—are faithfully represented in the final font.

Glyph Positioning and Metrics

Proper positioning of Tibetan characters is essential due to the script’s unique structure. For example, Tibetan vowel marks are often placed above or below the base consonants, and the positioning must be precise to avoid overlap.

• Vowel Mark Positioning: Tibetan vowel marks require specific placement, and their position varies relative to the base characters.
• Metrics and Spacing: Precise metrics are necessary to ensure legibility. This includes adjusting the Left Side Bearing (LSB) to prevent glyph overlap and modifying the Advance Width to ensure smooth text flow.
• Handling Punctuation: Tibetan punctuation marks, like the tsek, have specific spacing rules that must be respected to maintain clarity.

Implementation Pipeline

The entire font creation process is streamlined through the following pipeline:

1. File Processing: The system reads and organizes SVG files, extracting character information from each glyph.
2. Transformation: The SVG paths are converted into TrueType outlines, ensuring correct positioning and metrics.
3. Metrics Adjustment: The spacing and alignment of each glyph are fine-tuned to ensure smooth and accurate text rendering.

Font Metadata

Once the font is created, metadata is embedded to define the font’s name, family, and style. This metadata ensures the font is correctly identified and displayed across different systems.

Usage Workflow

To create the font, users simply need to:

1. Prepare SVG files with a consistent naming convention.
2. Organize these files in a structured directory.
3. Run the Python script with specified paths for the SVG files, a base font reference, and the destination path for the final TTF font file.

Conclusion

This approach to creating Tibetan fonts from SVG glyphs offers a robust and efficient method for developing high-quality, OCR-friendly fonts. By addressing the unique challenges of Tibetan script—such as precise glyph positioning, baseline adjustments, and spacing—this solution ensures that the final font faithfully represents traditional Tibetan typography. The flexibility of this process allows for easy customization, making it adaptable to different scripts and typographical needs. Ultimately, this framework serves as a valuable tool for Tibetan digital typeface development, preserving cultural heritage while improving OCR accuracy.