Captcha Engine

A high-performance, ONNX-based captcha recognition engine for Rust.

Features

• High Accuracy: Uses a custom-trained CNN-RNN-CTC model (95%+ accuracy)
• Fast Inference: Powered by ort (ONNX Runtime)
• Flexible Deployment:
  • download (default): Tiny binary, downloads model on first run
  • embed-model: Zero-dependency, single binary distribution (bundles model)
• Standard Interface: Simple predict(image) API

Usage

Add to Cargo.toml:

[dependencies]
captcha-engine = { path = "crates/captcha-engine" }
# For embedded model:
# captcha-engine = { path = "crates/captcha-engine", features = ["embed-model"], default-features = false }

Example

use captcha_engine::CaptchaModel;

fn main() -> anyhow::Result<()> {
    // 1. Initialize Model
    // With `embed-model` feature (recommended for plugins):
    let mut model = CaptchaModel::load_embedded()?;

    // OR with `download` feature (default):
    // let model_path = captcha_engine::ensure_model_downloaded("models")?;
    // let mut model = CaptchaModel::load(&model_path)?;

    // 2. Predict
    let image = image::open("captcha.png")?;
    let text = model.predict(&image)?;

    println!("Captcha: {}", text);
    Ok(())
}

Architecture

• Input: 215x80 RGB images
• Model: Custom ResNet + BiLSTM + CTC (ONNX format)
• Output: 6-character alphanumeric string (case-insensitive)