ocrs

Ocrs

ocrs is a Rust library and CLI tool for extracting text from images, also known as OCR (Optical Character Recognition).

The goal is to create a modern OCR engine that:

• Works well on a wide variety of images (scanned documents, photos containing text, screenshots etc.) with zero or much less preprocessing effort compared to earlier engines like Tesseract. This is achieved by using machine learning more extensively in the pipeline.
• Is easy to compile and run across a variety of platforms, including WebAssembly
• Is trained on open and liberally licensed datasets
• Has a codebase that is easy to understand and modify

Under the hood, the library uses neural network models trained in PyTorch, which are then exported to ONNX and executed using the RTen engine. See the models section for more details.

Status

ocrs is currently in an early preview. Expect more errors than commercial OCR engines.

Language Support

ocrs currently recognizes the Latin alphabet only (eg. English). Support for more languages is planned.

CLI installation

To install the CLI tool, you will first need Rust and Cargo installed. Then run:

$ cargo install ocrs-cli --locked

To enable support for reading images from the system clipboard, add the clipboard feature:

$ cargo install ocrs-cli --locked --features clipboard

CLI usage

To extract text from an image, run:

$ ocrs image.png

When the tool is run for the first time, it will download the required models automatically and store them in ~/.cache/ocrs.

If ocrs was installed with the clipboard feature, you can extract text from an image on the system clipboard with:

$ ocrs --clipboard
$ ocrs -c  # Short form

Additional examples

Extract text from an image and write to content.txt:

$ ocrs image.png -o content.txt

Extract text and layout information from the image in JSON format:

$ ocrs image.png --json -o content.json

Annotate an image to show the location of detected words and lines:

$ ocrs image.png --png -o annotated.png

Library usage

See the ocrs crate README for details on how to use ocrs as a Rust library.

Models and datasets

ocrs uses neural network models written in PyTorch. See the ocrs-models repository for more details about the models and datasets, as well as tools for training custom models.

Development

To build and run the ocrs library and CLI tool locally you will need a recent stable Rust version installed. Then run:

git clone https://github.com/robertknight/ocrs.git
cd ocrs
cargo run -p ocrs-cli -r -- image.png

Common development tasks are defined as recipes in the justfile and run with just. Run just --list to see them all.

Testing

Ocrs has unit tests for the code that runs before and after ML model processing, plus E2E tests which exercise the whole pipeline, including models.

After making changes to the code, run unit tests and lint checks with:

just check

You can also run standard commands like cargo test directly.

Run the E2E tests with:

just test-e2e

For details of how the ML models are evaluated, see the ocrs-models repository.