Character Fusion Dataset Generator

A toolkit for generating synthetic text image datasets featuring character fusion and color gradient effects, supporting both Chinese characters and English words. For project "Visible Yet Unreadable: A Systematic Blind Spot of Vision–Language Models Across Writing Systems".

🌟 Features

• Chinese Character Fusion: Generate fused Chinese character images using various blending modes
• English Color Gradient: Create gradient-colored English word images with artistic effects
• Multiple Fusion Modes: Support for left-right split, top-bottom split, diagonal cut, checkerboard, stripes, and alpha blending
• Flexible Input: Process sentences, idioms, or custom character pairs
• Batch Processing: Generate multiple variations and layouts efficiently
• JSON Annotations: Comprehensive metadata for each generated image

🚀 Quick Start

Prerequisites

pip install pillow numpy scipy tqdm

Chinese Character Fusion

# Generate from idioms file
python cin-gen.py --font path/to/chinese/font.otf --idioms_file cin_100.txt --out_dir chinese_output

# Generate from custom sentences
python cin-gen.py --font path/to/chinese/font.otf --sentence "机器学习" "人工智能" --out_dir chinese_output

# Generate all modes for comparison
python cin-gen.py --font path/to/chinese/font.otf --idioms_file cin_100.txt --all_modes --out_dir chinese_output

English Color Gradient

# Generate from word list
python en-color-gradient-fusion.py --words_file en_100.txt --out_dir english_output

# Generate with custom color schemes
python en-color-gradient-fusion.py --words computer security language --color_scheme1 fire --color_scheme2 ocean --out_dir english_output

# Save original parts for comparison
python en-color-gradient-fusion.py --words_file en_100.txt --save_originals --out_dir english_output

📊 Fusion Modes

Chinese Character Fusion

Mode	Description	Visual Effect
lr	Left-Right Split	First char on left, second on right
tb	Top-Bottom Split	First char on top, second on bottom
diag	Diagonal Cut	Diagonal line separation
checker	Checkerboard	Alternating square pattern
vstripes	Vertical Stripes	Vertical stripe pattern
hstripes	Horizontal Stripes	Horizontal stripe pattern
alpha	Alpha Blending	Transparent overlay effect

English Color Schemes

Scheme	Colors	Effect
fire	Red → Orange	Warm flame effect
ocean	Blue → Cyan	Cool water effect
forest	Green → Yellow-Green	Natural forest effect
sunset	Coral → Gold	Sunset gradient
rainbow1	Red → Green	Vibrant transition
neon	Pink → Cyan	Electric neon effect

📁 Project Structure

├── cin-gen.py                    # Chinese character fusion generator
├── en-color-gradient-fusion.py   # English color gradient generator
├── cin_100.txt                   # Sample Chinese idioms (100 entries)
├── en_100.txt                     # Sample English words (100 entries)
└── README.md                      # This file

🎯 Use Cases

• Computer Vision Research: Generate synthetic datasets for text recognition tasks
• Font Style Transfer: Create training data for style transfer models
• Data Augmentation: Expand existing text datasets with artistic variations
• Typography Research: Study character fusion and color gradient effects
• Educational Tools: Demonstrate text processing and image generation techniques

📋 Command Line Options

Chinese Generator (cin-gen.py)

Option	Description	Default
--font	Chinese font path (required)	-
--out_dir	Output directory	heti_out
--cell_size	Cell size in pixels	400
--cols	Grid columns	2
--modes	Fusion modes (comma-separated)	lr,tb,diag
--repeat	Random variations per sample	2
--all_modes	Generate all-mode layout	False
--separate_modes	Generate separate mode images	False
--sentence	Input sentences	-
--idioms_file	Idioms file path	-
--pairs	Custom character pairs	-

English Generator (en-color-gradient-fusion.py)

Option	Description	Default
--font	Font path (optional)	System default
--out_dir	Output directory	color_gradient_fusion_out
--font_size	Font size	120
--canvas_width	Canvas width	800
--canvas_height	Canvas height	400
--color_scheme1	First part color scheme	pure_red
--color_scheme2	Second part color scheme	pure_green
--gradient_direction	Gradient direction	horizontal
--save_originals	Save original parts	False
--words	Input words	-
--words_file	Words file path	-

📈 Output Format

Both generators create:

• Image files: PNG format with artistic effects
• Annotation file: labels.jsonl with comprehensive metadata

Sample Annotation Entry

{
  "image": "example.png",
  "source": "idiom",
  "raw_text": "画蛇添足",
  "pairs": ["画+蛇", "添+足"],
  "modes": ["lr", "tb"],
  "grid": {"rows": 2, "cols": 1, "cell_size": 400, "gap": 14}
}

🔧 Advanced Usage

Batch Processing Multiple Files

# Process multiple idiom files
for file in idioms_*.txt; do
    python cin-gen.py --font font.otf --idioms_file "$file" --out_dir "output_$(basename $file .txt)"
done

Custom Font Integration

# Use different fonts for variety
python cin-gen.py --font NotoSerifSC-Regular.otf --sentence "传统字体"
python cin-gen.py --font SourceHanSansCN-Bold.otf --sentence "现代字体"

🤝 Contributing

Contributions are welcome! Please feel free to:

• Report bugs and issues
• Suggest new fusion modes or color schemes
• Improve documentation
• Add support for new languages

📄 License

This project is open source. Please check individual file headers for specific license information.

🙏 Acknowledgments

• Built with Python PIL, NumPy, and SciPy
• Supports various font formats including TTF and OTF
• Inspired by artistic text effects and computer vision research

---

Generated with ❤️ for text processing and computer vision research