RapidOCR implementation for .NET - Cross-platform OCR processing using PaddleOCR ONNX models. Based on RapidAI's RapidOCR
$ dotnet add package RapidOCR.NetCross-platform OCR processing library using PaddleOCR ONNX models, and based on original code from RapidAI's RapidOCR.
The code was optimised to remove dependencies on System.Drawing and OpenCV. The image processing is now done only using SkiaSharp and PContourNet.
Note: This is a fork of the original RapidOcrNet repository and the NuGet package is independently published. It is not an official release from the RapidOCR team.
Install via NuGet Package Manager:
dotnet add package RapidOCR.Net
Or via Package Manager Console:
Install-Package RapidOCR.Net
string targetImg = "image.png";
using (var ocrEngin = new RapidOcr())
{
ocrEngin.InitModels();
using (SKBitmap originSrc = SKBitmap.Decode(targetImg))
{
OcrResult ocrResult = ocrEngin.Detect(originSrc, RapidOcrOptions.Default);
Console.WriteLine(ocrResult.ToString());
Console.WriteLine(ocrResult.StrRes);
Console.WriteLine();
// Draw bounding boxes
foreach (var block in ocrResult.TextBlocks)
{
var points = block.BoxPoints;
using (var canvas = new SKCanvas(originSrc))
using (var paint = new SKPaint() { Color = SKColors.Red })
{
canvas.DrawLine(points[0], points[1], paint);
canvas.DrawLine(points[1], points[2], paint);
canvas.DrawLine(points[2], points[3], paint);
canvas.DrawLine(points[3], points[0], paint);
}
}
using (var fs = new FileStream(Path.ChangeExtension(targetImg, "_ocr.png"), FileMode.Create))
{
originSrc.Encode(fs, SKEncodedImageFormat.Png, 100);
}
}
}
Based on source code originally developed in the RapidOCR project (Apache-2.0 license).
Uses parts of source code originally developed in the PdfPig project (Apache-2.0 license).
The dependency on OpenCV was removed thanks to the PContour library and its C# port.
The models made available are from the PaddleOCR project (Apache-2.0 license) and were converted to ONNX using the Paddle2ONNX tool.