# Implementing a Custom IOcrProvider

RadPdfProcessing offers a [default](https://www.telerik.com/document-processing-libraries/documentation/libraries/radpdfprocessing/formats-and-conversion/ocr/ocrformatprovider) implementation for an `IOcrProvider` engine wrapper that `OcrFormatProvider` uses. The `TesseractOcrProvider` uses the [Tesseract 5.2.0 engine](https://github.com/tesseract-ocr/tesseract?tab=readme-ov-file#tesseract-ocr) to extract text from an image.

You can also implement your own `IOcrProvider` that uses a different engine to recognize text from a screenshot.

## Using the Azure AI Vision

The [Azure AI Vision](https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/overview) service provides access to the [Optical Character Recognition](https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/overview-ocr) (OCR) service that extracts text from images. The following example uses the Azure [OCR engine](https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/overview-ocr#ocr-engine) to implement a custom `IOcrProvider` for the RadPdfProcessing library.

1\. Add the following **required** NuGet packages to your project:

* [Azure.AI.Vision.ImageAnalysis](https://www.nuget.org/packages/Azure.AI.Vision.ImageAnalysis)
* Telerik.Documents.Fixed
* Telerik.Documents.Fixed.FormatProviders.Ocr
* Telerik.Documents.ImageUtils
* SkiaSharp

2\. Generate your Azure AI key and endpoint: [Get your credentials from your Azure AI services resource](https://learn.microsoft.com/en-us/azure/ai-services/use-key-vault?tabs=azure-cli\&pivots=programming-language-csharp)

![Azure AI key and endpoint configuration](https://www.telerik.com/document-processing-libraries/documentation/assets/f82eec76e68770e6b04c86db1e95f0dc/azure-ai-key.png)

3\. Create the custom `AzureAIOcrProvider` class that implements the `IOcrProvider` interface:

```C#
public class AzureAIOcrProvider : IOcrProvider
{
    public const string azure_key = "your azure key";
    public const string azure_endpoint = "your azure endpoint";
    bool IOcrProvider.CorrectVerticalPosition => false;

    OcrParseLevel IOcrProvider.ParseLevel { get; set; } = OcrParseLevel.Word;

    string IOcrProvider.GetAllTextFromImage(byte[] imageBytes)
    {
        ImageAnalysisResult imageResult = Analyze(imageBytes);

        StringBuilder stringBuilder = new StringBuilder();

        foreach (DetectedTextBlock block in imageResult.Read.Blocks)
        {
            foreach (DetectedTextLine line in block.Lines)
            {
                foreach (DetectedTextWord word in line.Words)
                {
                    stringBuilder.AppendFormat("{0} ", word.Text);
                }
            }
        }

        string resultText = stringBuilder.ToString();

        return resultText;
    }
    private static Azure.AI.Vision.ImageAnalysis.ImageAnalysisResult Analyze(byte[] imageBytes)
    {
        string endpoint = azure_endpoint;
        string key = azure_key;

        ImageAnalysisClient client = new ImageAnalysisClient(
            new Uri(endpoint),
            new Azure.AzureKeyCredential(key));

        BinaryData binaryData = BinaryData.FromBytes(imageBytes);
        ImageAnalysisResult result = client.Analyze(
            binaryData,
            VisualFeatures.Read,
            new ImageAnalysisOptions { GenderNeutralCaption = false });
        return result;
    }

    Dictionary<Rectangle, string> IOcrProvider.GetTextFromImage(byte[] imageBytes)
    {
        ImageAnalysisResult result = Analyze(imageBytes);

        Dictionary<Rectangle, string> foundObjects = [];

        foreach (DetectedTextBlock block in result.Read.Blocks)
        {
            foreach (DetectedTextLine line in block.Lines)
            {
                foreach (DetectedTextWord word in line.Words)
                {
                    int topLeftX = word.BoundingPolygon.Min(p => p.X);
                    int topLeftY = word.BoundingPolygon.Min(p => p.Y);

                    int bottomRightX = word.BoundingPolygon.Max(p => p.X);
                    int bottomRightY = word.BoundingPolygon.Max(p => p.Y);

                    Rectangle rect = new Rectangle(new Point(topLeftX, topLeftY), new Size(bottomRightX - topLeftX, bottomRightY - topLeftY));

                    foundObjects.Add(rect, word.Text);
                }
            }
        }

        return foundObjects;
    }
}
```

4\. Use the custom implementation with the [OcrFormatProvider](https://www.telerik.com/document-processing-libraries/documentation/libraries/radpdfprocessing/formats-and-conversion/ocr/ocrformatprovider) that RadPdfProcessing offers:

```C#
FixedExtensibilityManager.ImagePropertiesResolver = new Telerik.Documents.ImageUtils.ImagePropertiesResolver();
AzureAIOcrProvider customOcrProvider = new AzureAIOcrProvider();
Telerik.Documents.Fixed.FormatProviders.Ocr.OcrFormatProvider PdfOcrProvider = new OcrFormatProvider(customOcrProvider);

RadFixedDocument document = new RadFixedDocument();
RadFixedPage page = new RadFixedPage();
string imagesFolderPath = @"..\..\..\images";
string[] pdfFilePaths = Directory.GetFiles(imagesFolderPath);
foreach (string imageFilePath in pdfFilePaths)
{
    byte[] bytes = System.IO.File.ReadAllBytes(imageFilePath);
    string text = PdfOcrProvider.OcrProvider.GetAllTextFromImage(bytes);
    page = PdfOcrProvider.Import(new FileStream(imageFilePath, FileMode.Open), null);
    document.Pages.Add(page);
}

PdfFormatProvider PdfProvider = new PdfFormatProvider();
string outputFilePath = "outputAzureAI.pdf";
if (System.IO.File.Exists(outputFilePath))
{
    System.IO.File.Delete(outputFilePath);
}
using (Stream output = System.IO.File.OpenWrite(outputFilePath))
{

    PdfProvider.Export(document, output, TimeSpan.FromSeconds(10));
}
Process.Start(new ProcessStartInfo() { FileName = outputFilePath, UseShellExecute = true });
```

After iterating all images in the specified folder (which contain content in different languages), the provider generates a PDF document with the respective content recognized as text fragments:

![Azure AI OCR result showing recognized text from multiple languages](https://www.telerik.com/document-processing-libraries/documentation/assets/6d6f3f70c4510fd4e48f9717768110ef/azure-ai-result.png)

## See Also

* [OCR Provider Prerequisites](https://www.telerik.com/document-processing-libraries/documentation/libraries/radpdfprocessing/formats-and-conversion/ocr/prerequisites)
* [Using OcrFormatProvider](https://www.telerik.com/document-processing-libraries/documentation/libraries/radpdfprocessing/formats-and-conversion/ocr/ocrformatprovider)
* [Cross-Platform Support - Images](https://www.telerik.com/document-processing-libraries/documentation/libraries/radpdfprocessing/cross-platform/images)
* [What is Azure AI Vision?](https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/overview)
* [Get your credentials from your Azure AI services resource](https://learn.microsoft.com/en-us/azure/ai-services/use-key-vault?tabs=azure-cli\&pivots=programming-language-csharp#get-your-credentials-from-your-azure-ai-services-resource)