OCR Preprocessor

A C++ tool designed to detect and extract text regions from images using Deep Learning models (DBNet). The tool handles text detection, perspective warping, and binarization to prepare images for OCR.

Features

• Text Detection: Uses ONNX Runtime with DBNet models.
• Geometry Processing: Merges fragmented detections using Stable Component Analysis.
• Image Warping: Unwarps perspective-distorted text regions.
• Binarization: Converts regions to black and white for better OCR accuracy.
• Auto-Cleaning: Automatically clears output directory before saving new results.

Prerequisites

• CMake: Version 3.17 or higher.
• Compiler: C++17 compliant compiler (GCC, Clang, or MSVC).
• Internet: Required during the first build to download ONNX Runtime dependencies automatically.

Building the Project

1. Clone the repository (if applicable).
2. Create a build directory:

   mkdir build
   cd build

3. Configure the project with CMake:

   cmake ..

   Note: This step will download the appropriate ONNX Runtime binaries for your platform (Linux/Windows).
4. Build the executable:

   make

Usage

./build/ocr_preprocessor <image_path> [model_path]

Arguments

• <image_path>: Path to the input image file.
• [model_path]: (Optional) Path to the ONNX model file. Defaults to det_v5.onnx.

Important: Model Compatibility

The default model det_v5.onnx uses a newer ONNX IR version (v10). The ONNX Runtime version used in this project supports up to IR version 9.

Please use the provided det_v5_ir9.onnx model.

Example Command

./build/ocr_preprocessor test1.png det_v5_ir9.onnx

Output Behavior

The tool is configured to always store output in the project root's output folder (ocp/output/), regardless of where you run the command from.

Directory Cleaning: The output/ directory is automatically cleaned (all files deleted) every time you run the tool to ensure you only see results from the latest run.