OCR language support reference and tips for different scripts

01 Supported languages overview

Tesseract.js inherits the powerful multilingual capabilities of the Tesseract OCR engine, supporting text recognition in over 100 languages. Below are the most commonly used languages and their characteristics.

Each language has a corresponding trained model file that is automatically downloaded on first use (typically a few MB). Once downloaded, models are cached by the browser and don't need to be re-fetched.

• English (eng) — Latin script, highest recognition rate
• Chinese Simplified (chi_sim) — Simplified Chinese characters for mainland documents
• Chinese Traditional (chi_tra) — Traditional Chinese characters for HK/TW documents
• Japanese (jpn) — Hiragana, Katakana, and Kanji characters
• Korean (kor) — Hangul syllabic script
• French (fra) — Latin script with accented characters
• Spanish (spa) — Latin script with special characters ñ, ¿, ¡
• German (deu) — Latin script with umlauts ä, ö, ü, ß
• Russian (rus) — Cyrillic script
• Arabic (ara) — Right-to-left writing system

02 CJK character recognition tips

CJK characters pose a far greater challenge for OCR engines than Latin scripts due to their massive character sets and complex strokes. Achieving good results requires attention to several key factors.

First, image resolution is critical. Because CJK strokes are dense, low-resolution images easily cause strokes to merge or blur. Ensure text height in the image is at least 30 pixels.

Second, correctly distinguishing between Simplified and Traditional Chinese is essential. Using the wrong model not only lowers accuracy but can produce many incorrect characters. For Japanese documents with heavy Kanji usage, consider loading both Japanese and Chinese models.

Finally, vertical text layouts (common in traditional Chinese and Japanese typesetting) may yield less accurate results than horizontal text. When possible, rotate images to horizontal orientation before processing.

03 Handling mixed-language documents

In practice, many documents use multiple languages. Technical documents often mix Chinese and English, academic papers may include Latin or Greek symbols, and business documents may span several languages.

Tesseract.js allows loading multiple language models simultaneously to handle mixed-language documents. Simply select all relevant languages in the language picker. For example, for a Chinese-English document, select both "English" and "Chinese Simplified".

Be aware that the more language models loaded, the slower the processing and the higher the memory usage. Select only the languages actually present in the document and avoid loading unnecessary models. Typically 2-3 languages is the ideal balance.

04 Accuracy optimization techniques

Regardless of the language used, the following optimization tips will help you significantly improve OCR recognition accuracy.

Image quality is the primary factor affecting recognition. Ensure images are sharp, text edges are crisp, and there's sufficient contrast between background and text. If the original image quality is poor, preprocess it with an image editing tool first.

Text orientation and alignment also matter. Ensure text in the image is horizontal with no visible tilt. Even a small skew angle (2-3 degrees) can significantly impact accuracy. Most image editors provide rotation and correction features.

• Use images at 300 DPI or higher resolution
• Ensure high contrast between text and background (dark text on light background is ideal)
• Crop unnecessary margins and non-text areas
• Correct skewed images to keep text horizontal
• Avoid heavily compressed JPEGs (compression artifacts interfere with recognition)
• For blurry images, try sharpening before recognition