oneocr-cli

The 1ocr command — images, PDFs and whole folders to text, JSON, SVG and semantic XHTML.

Website · All repositories · Live output

---

Requirements

• API: oneocr-api
• JDK 22+ (recommended) or JDK 21 with --enable-preview for runtime
• Windows 11

Build

mvn clean package

Usage

# Basic OCR - generates all formats by default (JSON, XHTML, SVG, TXT)
java --enable-native-access=ALL-UNNAMED -jar target/1ocr-2.0.jar image.jpg

# Verbose mode with informative output
java --enable-native-access=ALL-UNNAMED -jar target/1ocr-2.0.jar -v image.jpg

# Custom output files
java --enable-native-access=ALL-UNNAMED -jar target/1ocr-2.0.jar --json output.json --svg output.svg --xhtml output.xhtml --text output.txt image.jpg 

# Only specific formats (disable defaults)
java --enable-native-access=ALL-UNNAMED -jar target/1ocr-2.0.jar --no-defaults --xhtml document.xhtml image.jpg

# Confidence filtering and verbose output
java --enable-native-access=ALL-UNNAMED -jar target/1ocr-2.0.jar --min-confidence 0.8 --verbose image.jpg 

Default Output Files

When no specific output files are specified, the tool generates:

• image.jpg.oneocr.txt - Plain text
• image.jpg.oneocr.json - JSON with metadata
• image.jpg.oneocr.xhtml - XHTML format
• image.jpg.oneocr.svg - SVG visualization

Features

• JSON export with metadata
• SVG visualization with bounding boxes
• Semantic XHTML5 output (see format analysis for AI processing advantages)
• Plain text extraction
• Command-line interface

Notes

• Windows 11 Snipping tools OneOCR model (win11-oneocr) is the best price/quality/speed trade-off you can get for OCR. It's quality and multi-language OCR capabilities are state-of-art.
• The win11-oneocr - is quiet light-weight, just 50MB dll and 50MB model, it does 1 page in 1 to 2 seconds giving a rate of 30K-50K pages per day using pure CPU (utilizing all cores) without requiring GPU.
• The fact that you don't need a GPU is both a pro and con. Not able to leverage GPU limits scaling.
• This tools module is however just an example implementation of the api, and serves as a fairly decent standalone simple tool. It lacks batch processing which is essential, otherwise for each image you would be re-loading the model each time which has a performance hit if you are planning to do millions of pages OCR with this.
• However there are examples which demonstrate how to load model once and do images in batch thus utilizing full CPU ( see TestJDK21JDK22Plus.java in api module). Batch processing in command line will be implemented soon.
• GraalVM native executable version is planned for future releases to simplify deployment, eliminate JDK dependency.
• To use this, you must first build the api module. We DO NOT provide pre-build jars or exes.

Related Projects

• API Module: oneocr-api - Core Java FFM bindings and native libraries
• XHTML Presentation: semantic_xhtml - Interactive CSS/JS for XHTML format
• Original Research: win11-oneocr - Initial reverse engineering of Windows 11 OneOCR

Project Ecosystem

oneocr-api (Core FFM bindings)
    ↓
oneocr-cli (Command-line tools) → Generates XHTML
    ↓
semantic_xhtml (Presentation layer) → Renders XHTML