Crate oar_ocr

OAR OCR

A Rust OCR library that extracts text from document images using ONNX models. Supports text detection, recognition, document orientation, and rectification.

Features

• Complete OCR pipeline from image to text
• Modular components (use only what you need)
• Batch processing support
• ONNX Runtime integration for fast inference

Components

• Text Detection: Find text regions in images
• Text Recognition: Convert text regions to readable text
• Document Orientation: Detect document rotation (0°, 90°, 180°, 270°)
• Document Rectification: Fix perspective distortion
• Text Line Classification: Detect text line orientation

Modules

• core - Core traits, error handling, and batch processing
• predictor - OCR predictor implementations
• pipeline - Complete OCR pipeline
• processors - Image processing utilities
• utils - Utility functions for images and tensors

Quick Start

Complete OCR Pipeline

use oar_ocr::prelude::*;
use std::path::Path;

// Build OCR pipeline
let mut ocr = OAROCRBuilder::new(
    "detection_model.onnx".to_string(),
    "recognition_model.onnx".to_string(),
    "char_dict.txt".to_string(),
).build()?;

// Process single image
let image = load_image(Path::new("document.jpg"))?;
let results = ocr.predict(&[image])?;
let result = &results[0];

// Print results
for region in &result.text_regions {
    if let (Some(text), Some(confidence)) = (&region.text, region.confidence) {
        println!("Text: {} (confidence: {:.2})", text, confidence);
    }
}

// Process multiple images
let images = load_images_batch(&[Path::new("doc1.jpg"), Path::new("doc2.jpg")])?;
let results = ocr.predict(&images)?;
for result in results {
    println!("Image {}: {} text regions found", result.index, result.text_regions.len());
}

Advanced Configuration with Confidence Thresholding

use oar_ocr::prelude::*;
use std::path::Path;

// Build OCR pipeline with confidence thresholding for orientation detection
let mut ocr = OAROCRBuilder::new(
    "detection_model.onnx".to_string(),
    "recognition_model.onnx".to_string(),
    "char_dict.txt".to_string(),
)
// Configure document orientation with confidence threshold
.doc_orientation_classify_model_path("orientation_model.onnx")
.doc_orientation_confidence_threshold(0.8) // Only accept predictions with 80% confidence
.use_doc_orientation_classify(true)
// Configure text line orientation with confidence threshold
.textline_orientation_classify_model_path("textline_orientation_model.onnx")
.textline_orientation_confidence_threshold(0.7) // Only accept predictions with 70% confidence
.use_textline_orientation(true)
// Set recognition score threshold
.text_rec_score_thresh(0.5)
.build()?;

// Process image - low confidence orientations will fall back to defaults
let image = load_image(Path::new("document.jpg"))?;
let results = ocr.predict(&[image])?;

Individual Components

use oar_ocr::prelude::*;
use oar_ocr::core::traits::StandardPredictor;
use oar_ocr::utils::load_image;
use std::path::Path;

// Text detection only
let mut detector = TextDetPredictorBuilder::new()
    .build(Path::new("detection_model.onnx"))?;

let image = load_image(Path::new("image.jpg"))?;
let result = detector.predict(vec![image], None)?;
println!("Detection result: {:?}", result);

// Text recognition only
let char_dict = vec!["a".to_string(), "b".to_string()]; // Load your dictionary
let mut recognizer = TextRecPredictorBuilder::new()
    .character_dict(char_dict)
    .build(Path::new("recognition_model.onnx"))?;

let image = load_image(Path::new("text_crop.jpg"))?;
let result = recognizer.predict(vec![image], None)?;
println!("Recognition result: {:?}", result);

Modules

core

The core module of the OCR pipeline.

pipeline

The OCR pipeline module.

predictor

Predictor implementations for various OCR tasks.

prelude

Prelude module for convenient imports.

processors

Image processing utilities for OCR systems.

utils

Utility functions for the OCR pipeline.

Macros

impl_complete_builder

Comprehensive builder macro for generating common builder method patterns.

impl_config_validator

Macro to implement ConfigValidator with basic validation patterns.

metrics

Macro to create pre-populated StageMetrics with common patterns.

validate_field

Helper macro for field validation.

with_nested

Macro to handle optional nested config initialization in builders.