Amazon Textract vs IronOCR: .NET OCR Library

The S3-Async Pipeline

Single-image OCR via DetectDocumentTextAsync can pass document bytes directly in the request. Multi-page PDFs cannot. Any PDF requires the full asynchronous pipeline:

// AWS Textract: Multi-page PDF requires S3 + async job
public async Task<string> ProcessPdfAsync(string pdfPath)
{
    // Step 1: Upload to S3 — credentials for two services required
    var key = $"uploads/{Guid.NewGuid()}.pdf";
    using (var fileStream = File.OpenRead(pdfPath))
    {
        await _s3Client.PutObjectAsync(new PutObjectRequest
        {
            BucketName = _bucketName,
            Key = key,
            InputStream = fileStream
        });
    }

    try
    {
        // Step 2: Start async Textract job
        var startResponse = await _textractClient.StartDocumentTextDetectionAsync(
            new StartDocumentTextDetectionRequest
            {
                DocumentLocation = new DocumentLocation
                {
                    S3Object = new S3Object { Bucket = _bucketName, Name = key }
                }
            });

        var jobId = startResponse.JobId;

        // Step 3: Poll every 5 seconds until complete
        GetDocumentTextDetectionResponse getResponse;
        do
        {
            await Task.Delay(5000);
            getResponse = await _textractClient.GetDocumentTextDetectionAsync(
                new GetDocumentTextDetectionRequest { JobId = jobId });
        } while (getResponse.JobStatus == JobStatus.IN_PROGRESS);

        if (getResponse.JobStatus != JobStatus.SUCCEEDED)
            throw new Exception($"Textract job failed: {getResponse.StatusMessage}");

        // Step 4: Paginate through result blocks
        var allText = new StringBuilder();
        string nextToken = null;
        do
        {
            var pageResponse = await _textractClient.GetDocumentTextDetectionAsync(
                new GetDocumentTextDetectionRequest
                {
                    JobId = jobId,
                    NextToken = nextToken
                });

            foreach (var block in pageResponse.Blocks.Where(b => b.BlockType == BlockType.LINE))
                allText.AppendLine(block.Text);

            nextToken = pageResponse.NextToken;
        } while (nextToken != null);

        return allText.ToString();
    }
    finally
    {
        // Step 5: Always clean up S3
        await _s3Client.DeleteObjectAsync(_bucketName, key);
    }
}

// AWS Textract: Multi-page PDF requires S3 + async job
public async Task<string> ProcessPdfAsync(string pdfPath)
{
    // Step 1: Upload to S3 — credentials for two services required
    var key = $"uploads/{Guid.NewGuid()}.pdf";
    using (var fileStream = File.OpenRead(pdfPath))
    {
        await _s3Client.PutObjectAsync(new PutObjectRequest
        {
            BucketName = _bucketName,
            Key = key,
            InputStream = fileStream
        });
    }

    try
    {
        // Step 2: Start async Textract job
        var startResponse = await _textractClient.StartDocumentTextDetectionAsync(
            new StartDocumentTextDetectionRequest
            {
                DocumentLocation = new DocumentLocation
                {
                    S3Object = new S3Object { Bucket = _bucketName, Name = key }
                }
            });

        var jobId = startResponse.JobId;

        // Step 3: Poll every 5 seconds until complete
        GetDocumentTextDetectionResponse getResponse;
        do
        {
            await Task.Delay(5000);
            getResponse = await _textractClient.GetDocumentTextDetectionAsync(
                new GetDocumentTextDetectionRequest { JobId = jobId });
        } while (getResponse.JobStatus == JobStatus.IN_PROGRESS);

        if (getResponse.JobStatus != JobStatus.SUCCEEDED)
            throw new Exception($"Textract job failed: {getResponse.StatusMessage}");

        // Step 4: Paginate through result blocks
        var allText = new StringBuilder();
        string nextToken = null;
        do
        {
            var pageResponse = await _textractClient.GetDocumentTextDetectionAsync(
                new GetDocumentTextDetectionRequest
                {
                    JobId = jobId,
                    NextToken = nextToken
                });

            foreach (var block in pageResponse.Blocks.Where(b => b.BlockType == BlockType.LINE))
                allText.AppendLine(block.Text);

            nextToken = pageResponse.NextToken;
        } while (nextToken != null);

        return allText.ToString();
    }
    finally
    {
        // Step 5: Always clean up S3
        await _s3Client.DeleteObjectAsync(_bucketName, key);
    }
}

Imports System
Imports System.IO
Imports System.Text
Imports System.Threading.Tasks
Imports Amazon.S3
Imports Amazon.Textract
Imports Amazon.Textract.Model

Public Class PdfProcessor
    Private _s3Client As IAmazonS3
    Private _textractClient As IAmazonTextract
    Private _bucketName As String

    Public Async Function ProcessPdfAsync(pdfPath As String) As Task(Of String)
        ' Step 1: Upload to S3 — credentials for two services required
        Dim key = $"uploads/{Guid.NewGuid()}.pdf"
        Using fileStream = File.OpenRead(pdfPath)
            Await _s3Client.PutObjectAsync(New PutObjectRequest With {
                .BucketName = _bucketName,
                .Key = key,
                .InputStream = fileStream
            })
        End Using

        Try
            ' Step 2: Start async Textract job
            Dim startResponse = Await _textractClient.StartDocumentTextDetectionAsync(
                New StartDocumentTextDetectionRequest With {
                    .DocumentLocation = New DocumentLocation With {
                        .S3Object = New S3Object With {.Bucket = _bucketName, .Name = key}
                    }
                })

            Dim jobId = startResponse.JobId

            ' Step 3: Poll every 5 seconds until complete
            Dim getResponse As GetDocumentTextDetectionResponse
            Do
                Await Task.Delay(5000)
                getResponse = Await _textractClient.GetDocumentTextDetectionAsync(
                    New GetDocumentTextDetectionRequest With {.JobId = jobId})
            Loop While getResponse.JobStatus = JobStatus.IN_PROGRESS

            If getResponse.JobStatus <> JobStatus.SUCCEEDED Then
                Throw New Exception($"Textract job failed: {getResponse.StatusMessage}")
            End If

            ' Step 4: Paginate through result blocks
            Dim allText = New StringBuilder()
            Dim nextToken As String = Nothing
            Do
                Dim pageResponse = Await _textractClient.GetDocumentTextDetectionAsync(
                    New GetDocumentTextDetectionRequest With {
                        .JobId = jobId,
                        .NextToken = nextToken
                    })

                For Each block In pageResponse.Blocks.Where(Function(b) b.BlockType = BlockType.LINE)
                    allText.AppendLine(block.Text)
                Next

                nextToken = pageResponse.NextToken
            Loop While nextToken IsNot Nothing

            Return allText.ToString()
        Finally
            ' Step 5: Always clean up S3
            Await _s3Client.DeleteObjectAsync(_bucketName, key)
        End Try
    End Function
End Class

This is the minimum viable implementation for reliable PDF processing — five distinct phases, two AWS service clients, and cleanup logic in a finally block. The complete production version with proper error handling, rate limit retry logic, and timeout management runs 150-300 lines.

Detailed Feature Comparison

AWS Textract Approach

// Every call to this method costs money — per page, permanently
public async Task<string> DetectTextAsync(string imagePath)
{
    var imageBytes = File.ReadAllBytes(imagePath);  // Image leaves your network

    var request = new DetectDocumentTextRequest
    {
        Document = new Document
        {
            Bytes = new MemoryStream(imageBytes)
        }
    };

    var response = await _client.DetectDocumentTextAsync(request);  // per-page charge

    return string.Join("\n", response.Blocks
        .Where(b => b.BlockType == BlockType.LINE)
        .Select(b => b.Text));
}

// Every call to this method costs money — per page, permanently
public async Task<string> DetectTextAsync(string imagePath)
{
    var imageBytes = File.ReadAllBytes(imagePath);  // Image leaves your network

    var request = new DetectDocumentTextRequest
    {
        Document = new Document
        {
            Bytes = new MemoryStream(imageBytes)
        }
    };

    var response = await _client.DetectDocumentTextAsync(request);  // per-page charge

    return string.Join("\n", response.Blocks
        .Where(b => b.BlockType == BlockType.LINE)
        .Select(b => b.Text));
}

Imports System.IO
Imports System.Threading.Tasks

' Every call to this method costs money — per page, permanently
Public Async Function DetectTextAsync(imagePath As String) As Task(Of String)
    Dim imageBytes = File.ReadAllBytes(imagePath)  ' Image leaves your network

    Dim request = New DetectDocumentTextRequest With {
        .Document = New Document With {
            .Bytes = New MemoryStream(imageBytes)
        }
    }

    Dim response = Await _client.DetectDocumentTextAsync(request)  ' per-page charge

    Return String.Join(vbLf, response.Blocks _
        .Where(Function(b) b.BlockType = BlockType.LINE) _
        .Select(Function(b) b.Text))
End Function

Consult the AWS Textract pricing page for current per-page rates. Different API features (basic text detection, table extraction, forms extraction) have different rates. A document containing tables and form fields incurs higher charges than basic text detection, and costs grow with volume with no upper bound and no way to pay ahead.

At high page volumes, three-year total costs can be substantial, and the meter keeps running.

IronOCR Approach

// One license. No per-page cost. Same code handles 1 page or 1,000,000.
IronOcr.License.LicenseKey = "YOUR-LICENSE-KEY";

var text = new IronTesseract().Read("document.jpg").Text;

// One license. No per-page cost. Same code handles 1 page or 1,000,000.
IronOcr.License.LicenseKey = "YOUR-LICENSE-KEY";

var text = new IronTesseract().Read("document.jpg").Text;

Imports IronOcr

' One license. No per-page cost. Same code handles 1 page or 1,000,000.
IronOcr.License.LicenseKey = "YOUR-LICENSE-KEY"

Dim text As String = New IronTesseract().Read("document.jpg").Text

The $2,999 Professional license covers 10 developers, unlimited projects, and unlimited page volume. After year one, the ongoing cost for pages processed is zero. For teams processing significant page volumes, the IronOCR license pays for itself quickly compared to ongoing per-page cloud charges.

The IronOCR licensing page covers tier details, SaaS subscription options for usage-based billing scenarios, and OEM redistribution terms.

AWS Textract Approach

// This code sends PHI, legal documents, financial records — whatever is in
// the file — to Amazon Web Services infrastructure
public async Task<string> ProcessSensitiveDocumentAsync(string documentPath)
{
    var imageBytes = File.ReadAllBytes(documentPath);

    // Data crosses your security perimeter here
    var request = new DetectDocumentTextRequest
    {
        Document = new Document
        {
            Bytes = new MemoryStream(imageBytes)
        }
    };

    // Amazon processes it; you receive text back
    var response = await _client.DetectDocumentTextAsync(request);

    return string.Join("\n", response.Blocks
        .Where(b => b.BlockType == BlockType.LINE)
        .Select(b => b.Text));
}

// This code sends PHI, legal documents, financial records — whatever is in
// the file — to Amazon Web Services infrastructure
public async Task<string> ProcessSensitiveDocumentAsync(string documentPath)
{
    var imageBytes = File.ReadAllBytes(documentPath);

    // Data crosses your security perimeter here
    var request = new DetectDocumentTextRequest
    {
        Document = new Document
        {
            Bytes = new MemoryStream(imageBytes)
        }
    };

    // Amazon processes it; you receive text back
    var response = await _client.DetectDocumentTextAsync(request);

    return string.Join("\n", response.Blocks
        .Where(b => b.BlockType == BlockType.LINE)
        .Select(b => b.Text));
}

Imports System.IO
Imports System.Threading.Tasks
Imports Amazon.Textract
Imports Amazon.Textract.Model

Public Class DocumentProcessor
    Private _client As AmazonTextractClient

    Public Sub New(client As AmazonTextractClient)
        _client = client
    End Sub

    ' This code sends PHI, legal documents, financial records — whatever is in
    ' the file — to Amazon Web Services infrastructure
    Public Async Function ProcessSensitiveDocumentAsync(documentPath As String) As Task(Of String)
        Dim imageBytes = File.ReadAllBytes(documentPath)

        ' Data crosses your security perimeter here
        Dim request As New DetectDocumentTextRequest With {
            .Document = New Document With {
                .Bytes = New MemoryStream(imageBytes)
            }
        }

        ' Amazon processes it; you receive text back
        Dim response = Await _client.DetectDocumentTextAsync(request)

        Return String.Join(vbLf, response.Blocks _
            .Where(Function(b) b.BlockType = BlockType.LINE) _
            .Select(Function(b) b.Text))
    End Function
End Class

AWS offers a HIPAA Business Associate Agreement for covered entities, and GovCloud regions provide FedRAMP High authorization. These frameworks do not change the fundamental architecture: documents leave your infrastructure for every operation. For ITAR-controlled technical data, this is not a compliance nuance — it is a prohibition. For CMMC Level 3 workloads with CUI, cloud transmission requires specific authorizations most defense contractors do not hold. For air-gapped systems — research networks, industrial control environments, classified facilities — Textract is simply unavailable.

AWS Textract is available in six regions: us-east-1, us-west-2, eu-west-1, eu-west-2, ap-southeast-1, and ap-southeast-2. Organizations with data residency requirements outside these regions have no compliant option.

IronOCR Approach

// IronOCR: document bytes never leave this process
public string ProcessSensitiveDocument(string documentPath)
{
    // Processes entirely on local hardware — no network call
    var ocr = new IronTesseract();
    return ocr.Read(documentPath).Text;
}

// IronOCR: document bytes never leave this process
public string ProcessSensitiveDocument(string documentPath)
{
    // Processes entirely on local hardware — no network call
    var ocr = new IronTesseract();
    return ocr.Read(documentPath).Text;
}

' IronOCR: document bytes never leave this process
Public Function ProcessSensitiveDocument(documentPath As String) As String
    ' Processes entirely on local hardware — no network call
    Dim ocr As New IronTesseract()
    Return ocr.Read(documentPath).Text
End Function

Because IronOCR executes locally, it fits naturally into healthcare workflows processing PHI, legal document systems handling privileged communications, financial applications handling payment card images, and defense contractor pipelines processing CUI. There is no external processor to audit, no BAA to negotiate, no data residency constraint to satisfy. The compliance scope is your organization's own infrastructure.

For teams deploying on AWS infrastructure but needing local processing, IronOCR runs on AWS EC2 and Lambda without any dependency on Textract — the processing happens within your own AWS account boundary rather than Amazon's managed service.

AWS Textract Approach

// Full production-grade async processor for Textract PDF handling
public class TextractAsyncProcessor
{
    private readonly AmazonTextractClient _textractClient;
    private readonly AmazonS3Client _s3Client;
    private readonly string _bucketName;
    private readonly TimeSpan _pollInterval = TimeSpan.FromSeconds(5);
    private readonly TimeSpan _maxWaitTime = TimeSpan.FromMinutes(10);

    public async Task<DocumentResult> ProcessDocumentAsync(
        string localFilePath,
        CancellationToken cancellationToken = default)
    {
        var s3Key = $"textract-uploads/{Guid.NewGuid()}{Path.GetExtension(localFilePath)}";

        try
        {
            // Phase 1: Upload to S3
            await UploadToS3Async(localFilePath, s3Key, cancellationToken);

            // Phase 2: Start Textract job
            var jobId = await StartTextractJobAsync(s3Key, cancellationToken);

            // Phase 3: Poll until complete (up to 10 minutes)
            var pollResult = await PollForCompletionAsync(jobId, cancellationToken);

            if (!pollResult.Success)
                throw new Exception($"Textract job failed: {pollResult.ErrorMessage}");

            // Phase 4: Retrieve paginated results
            return await GetAllResultsAsync(jobId, cancellationToken);
        }
        finally
        {
            // Phase 5: S3 cleanup — must succeed or storage costs accumulate
            await DeleteFromS3Async(s3Key, cancellationToken);
        }
    }

    private async Task<(bool Success, string ErrorMessage)> PollForCompletionAsync(
        string jobId, CancellationToken cancellationToken)
    {
        var startTime = DateTime.UtcNow;
        int pollCount = 0;

        while (DateTime.UtcNow - startTime < _maxWaitTime)
        {
            cancellationToken.ThrowIfCancellationRequested();

            var response = await _textractClient.GetDocumentTextDetectionAsync(
                new GetDocumentTextDetectionRequest { JobId = jobId }, cancellationToken);

            pollCount++;

            switch (response.JobStatus)
            {
                case JobStatus.SUCCEEDED: return (true, null);
                case JobStatus.FAILED: return (false, response.StatusMessage ?? "Unknown error");
                case JobStatus.IN_PROGRESS:
                    await Task.Delay(_pollInterval, cancellationToken);
                    break;
                default:
                    throw new Exception($"Unknown job status: {response.JobStatus}");
            }
        }

        return (false, "Job timed out");
    }
}

// Full production-grade async processor for Textract PDF handling
public class TextractAsyncProcessor
{
    private readonly AmazonTextractClient _textractClient;
    private readonly AmazonS3Client _s3Client;
    private readonly string _bucketName;
    private readonly TimeSpan _pollInterval = TimeSpan.FromSeconds(5);
    private readonly TimeSpan _maxWaitTime = TimeSpan.FromMinutes(10);

    public async Task<DocumentResult> ProcessDocumentAsync(
        string localFilePath,
        CancellationToken cancellationToken = default)
    {
        var s3Key = $"textract-uploads/{Guid.NewGuid()}{Path.GetExtension(localFilePath)}";

        try
        {
            // Phase 1: Upload to S3
            await UploadToS3Async(localFilePath, s3Key, cancellationToken);

            // Phase 2: Start Textract job
            var jobId = await StartTextractJobAsync(s3Key, cancellationToken);

            // Phase 3: Poll until complete (up to 10 minutes)
            var pollResult = await PollForCompletionAsync(jobId, cancellationToken);

            if (!pollResult.Success)
                throw new Exception($"Textract job failed: {pollResult.ErrorMessage}");

            // Phase 4: Retrieve paginated results
            return await GetAllResultsAsync(jobId, cancellationToken);
        }
        finally
        {
            // Phase 5: S3 cleanup — must succeed or storage costs accumulate
            await DeleteFromS3Async(s3Key, cancellationToken);
        }
    }

    private async Task<(bool Success, string ErrorMessage)> PollForCompletionAsync(
        string jobId, CancellationToken cancellationToken)
    {
        var startTime = DateTime.UtcNow;
        int pollCount = 0;

        while (DateTime.UtcNow - startTime < _maxWaitTime)
        {
            cancellationToken.ThrowIfCancellationRequested();

            var response = await _textractClient.GetDocumentTextDetectionAsync(
                new GetDocumentTextDetectionRequest { JobId = jobId }, cancellationToken);

            pollCount++;

            switch (response.JobStatus)
            {
                case JobStatus.SUCCEEDED: return (true, null);
                case JobStatus.FAILED: return (false, response.StatusMessage ?? "Unknown error");
                case JobStatus.IN_PROGRESS:
                    await Task.Delay(_pollInterval, cancellationToken);
                    break;
                default:
                    throw new Exception($"Unknown job status: {response.JobStatus}");
            }
        }

        return (false, "Job timed out");
    }
}

Imports System
Imports System.IO
Imports System.Threading
Imports System.Threading.Tasks
Imports Amazon.Textract
Imports Amazon.S3
Imports Amazon.Textract.Model

' Full production-grade async processor for Textract PDF handling
Public Class TextractAsyncProcessor
    Private ReadOnly _textractClient As AmazonTextractClient
    Private ReadOnly _s3Client As AmazonS3Client
    Private ReadOnly _bucketName As String
    Private ReadOnly _pollInterval As TimeSpan = TimeSpan.FromSeconds(5)
    Private ReadOnly _maxWaitTime As TimeSpan = TimeSpan.FromMinutes(10)

    Public Async Function ProcessDocumentAsync(localFilePath As String, Optional cancellationToken As CancellationToken = Nothing) As Task(Of DocumentResult)
        Dim s3Key = $"textract-uploads/{Guid.NewGuid()}{Path.GetExtension(localFilePath)}"

        Try
            ' Phase 1: Upload to S3
            Await UploadToS3Async(localFilePath, s3Key, cancellationToken)

            ' Phase 2: Start Textract job
            Dim jobId = Await StartTextractJobAsync(s3Key, cancellationToken)

            ' Phase 3: Poll until complete (up to 10 minutes)
            Dim pollResult = Await PollForCompletionAsync(jobId, cancellationToken)

            If Not pollResult.Success Then
                Throw New Exception($"Textract job failed: {pollResult.ErrorMessage}")
            End If

            ' Phase 4: Retrieve paginated results
            Return Await GetAllResultsAsync(jobId, cancellationToken)
        Finally
            ' Phase 5: S3 cleanup — must succeed or storage costs accumulate
            Await DeleteFromS3Async(s3Key, cancellationToken)
        End Try
    End Function

    Private Async Function PollForCompletionAsync(jobId As String, cancellationToken As CancellationToken) As Task(Of (Success As Boolean, ErrorMessage As String))
        Dim startTime = DateTime.UtcNow
        Dim pollCount As Integer = 0

        While DateTime.UtcNow - startTime < _maxWaitTime
            cancellationToken.ThrowIfCancellationRequested()

            Dim response = Await _textractClient.GetDocumentTextDetectionAsync(New GetDocumentTextDetectionRequest With {.JobId = jobId}, cancellationToken)

            pollCount += 1

            Select Case response.JobStatus
                Case JobStatus.SUCCEEDED
                    Return (True, Nothing)
                Case JobStatus.FAILED
                    Return (False, If(response.StatusMessage, "Unknown error"))
                Case JobStatus.IN_PROGRESS
                    Await Task.Delay(_pollInterval, cancellationToken)
                Case Else
                    Throw New Exception($"Unknown job status: {response.JobStatus}")
            End Select
        End While

        Return (False, "Job timed out")
    End Function
End Class

This is not boilerplate that can be generated and forgotten. When a Textract job fails mid-flight, the S3 cleanup must still run. When a job times out after 10 minutes, the caller needs a clean error. When the network drops during polling, the retry strategy must not create duplicate jobs. Each of these failure modes requires explicit handling — the structure shown above is the minimum responsible implementation.

Batch processing adds another layer: Textract's default StartDocumentTextDetection TPS limit is 5 requests per second. Processing 100 documents requires a SemaphoreSlim throttle, a rate-replenishment timer, and retry logic for ProvisionedThroughputExceededException.

IronOCR Approach

// IronOCR: same synchronous API regardless of document type or size
public string ProcessDocument(string filePath)
{
    using var input = new OcrInput();

    if (Path.GetExtension(filePath).Equals(".pdf", StringComparison.OrdinalIgnoreCase))
        input.LoadPdf(filePath);
    else
        input.LoadImage(filePath);

    return new IronTesseract().Read(input).Text;
}

// IronOCR: same synchronous API regardless of document type or size
public string ProcessDocument(string filePath)
{
    using var input = new OcrInput();

    if (Path.GetExtension(filePath).Equals(".pdf", StringComparison.OrdinalIgnoreCase))
        input.LoadPdf(filePath);
    else
        input.LoadImage(filePath);

    return new IronTesseract().Read(input).Text;
}

Imports System.IO

' IronOCR: same synchronous API regardless of document type or size
Public Function ProcessDocument(filePath As String) As String
    Using input As New OcrInput()
        If Path.GetExtension(filePath).Equals(".pdf", StringComparison.OrdinalIgnoreCase) Then
            input.LoadPdf(filePath)
        Else
            input.LoadImage(filePath)
        End If

        Return New IronTesseract().Read(input).Text
    End Using
End Function

There is no polling loop, no job ID tracking, no S3 bucket, no result pagination. The same code handles a single JPEG and a 200-page PDF. Processing completes or throws — no intermediate "in progress" state to manage. For batch processing, IronOCR is thread-safe and a single IronTesseract instance handles Parallel.ForEach without locks or semaphores.

The IronTesseract setup guide covers configuration, and the PDF input guide documents page range selection, password-protected PDFs, and stream-based input for PDFs retrieved from databases or HTTP responses.

AWS Textract Approach

Before calling DetectDocumentTextAsync, a developer must:

1. Create an AWS account or obtain access to an existing one
2. Create an IAM user or role with textract:DetectDocumentText and textract:AnalyzeDocument permissions
3. Generate and securely store access key ID and secret access key
4. Configure credential resolution — environment variables, AWS credentials file, or EC2 instance profile
5. If processing PDFs: create an S3 bucket, configure bucket policy, add s3:PutObject and s3:DeleteObject permissions
6. Implement credential rotation policies to meet security standards
7. Store credentials securely in each deployment environment — Docker secrets, Kubernetes secrets, AWS Secrets Manager, or CI/CD pipeline variables

// Every environment needs these configured before this constructor succeeds
public TextractOcrService()
{
    // Reads credentials from environment, ~/.aws/credentials, or IAM role
    _client = new AmazonTextractClient(Amazon.RegionEndpoint.USEast1);
}

// Every environment needs these configured before this constructor succeeds
public TextractOcrService()
{
    // Reads credentials from environment, ~/.aws/credentials, or IAM role
    _client = new AmazonTextractClient(Amazon.RegionEndpoint.USEast1);
}

' Every environment needs these configured before this constructor succeeds
Public Sub New()
    ' Reads credentials from environment, ~/.aws/credentials, or IAM role
    _client = New AmazonTextractClient(Amazon.RegionEndpoint.USEast1)
End Sub

When credentials expire, rotate, or are misconfigured, every OCR call fails with AmazonTextractException carrying ErrorCode == "AccessDeniedException". In a production system, this means implementing specific catch blocks for credential failures and monitoring for IAM policy drift.

IronOCR Approach

// One-time setup at application startup
IronOcr.License.LicenseKey = "YOUR-LICENSE-KEY";

// Or from environment — recommended for deployments
IronOcr.License.LicenseKey = Environment.GetEnvironmentVariable("IRONOCR_LICENSE");

// One-time setup at application startup
IronOcr.License.LicenseKey = "YOUR-LICENSE-KEY";

// Or from environment — recommended for deployments
IronOcr.License.LicenseKey = Environment.GetEnvironmentVariable("IRONOCR_LICENSE");

' One-time setup at application startup
IronOcr.License.LicenseKey = "YOUR-LICENSE-KEY"

' Or from environment — recommended for deployments
IronOcr.License.LicenseKey = Environment.GetEnvironmentVariable("IRONOCR_LICENSE")

The license key is a static string. It does not expire mid-operation, does not require rotation, and carries no permissions to manage. A Docker container that processes documents does not need injected AWS credentials, an IAM role bound to an execution context, or network access to AWS STS for token refresh.

The complete credential overhead reduction when moving from Textract to IronOCR: three NuGet packages removed (AWSSDK.Textract, AWSSDK.S3, AWSSDK.Core), all AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY / AWS_DEFAULT_REGION environment variables removed, and IAM roles and S3 bucket configurations decommissioned. The image input guide and stream input guide cover the full range of input methods that replace Textract's byte-array and S3-object document models.

When the Monthly Bill Becomes a Budget Line Item

Teams that started with Textract at low volume often encounter a specific moment: the AWS bill for OCR processing appears in a quarterly budget review and someone asks whether this cost is fixed. It is not. At high page volumes, annual Textract costs can be substantial — consult the AWS Textract pricing page for current rates. The IronOCR Professional license at $2,999 one-time pays for itself quickly at moderate to high page volumes.

When a Compliance Requirement Blocks Cloud Processing

Healthcare organizations implementing document digitization workflows frequently discover mid-project that HIPAA PHI cannot flow through cloud services without a BAA and additional legal review, or that their security team prohibits cloud transmission entirely. Defense contractors handling technical drawings, specifications, or any CUI face ITAR and CMMC constraints that exclude AWS Textract from consideration. Legal firms processing privileged communications have similar concerns. These are not theoretical compliance edge cases — they appear regularly in procurement reviews, security audits, and contract negotiations. IronOCR processes locally, so the compliance question for document data reduces to whether your own infrastructure is in scope, not whether Amazon's infrastructure is in scope.

When the Async PDF Complexity Exceeds Its Value

The five-phase S3-async pipeline — upload, start job, poll, paginate results, clean up — is not technically difficult to implement. It is difficult to maintain, test, and operate. Every phase is a failure point. S3 upload failures require retry logic. Textract job failures require distinguishing transient from permanent errors. Polling timeouts require timeout handling separate from cancellation. Result pagination requires accumulating state across multiple API calls. S3 cleanup failures require alerting because orphaned objects accumulate costs. Teams that have shipped this pipeline into production typically spend more ongoing engineering time maintaining it than they spent building it. The IronOCR equivalent — input.LoadPdf(path) followed by ocr.Read(input) — eliminates all five phases and their associated failure modes.

When Deployment Environments Lack Internet Access

Docker containers running in isolated network segments, on-premise servers without outbound internet, air-gapped research environments, and industrial systems with strict network controls all share one characteristic: AWS Textract is not available. IronOCR installs as a standard NuGet package and operates without any network calls after installation. Teams running .NET applications in these environments have no Textract option and need a library that processes locally. The Docker deployment guide and Linux deployment guide cover the specific configuration for containerized environments.

When Rate Limit Throttling Disrupts Batch Workflows

The default StartDocumentTextDetection TPS limit is 5 requests per second. DetectDocumentText synchronous calls are also rate-limited. Batch jobs processing hundreds or thousands of documents must implement SemaphoreSlim throttling, exponential backoff on ProvisionedThroughputExceededException, and rate-replenishment timers. AWS supports TPS limit increase requests, but they require justification, review, and are not guaranteed. IronOCR processes as fast as local CPU allows — a 32-core server processes 32 documents concurrently without throttle configuration or service tier negotiation.

Replacing the Block Graph with Direct Collections

Textract represents all results as a flat List<Block> where lines, words, cells, tables, and key-value pairs are distinguished by BlockType and linked by relationship ID arrays. IronOCR provides direct typed collections.

// Textract: filter flat block list by type
var lines = response.Blocks.Where(b => b.BlockType == BlockType.LINE);
var words = response.Blocks.Where(b => b.BlockType == BlockType.WORD);

// IronOCR: direct access to typed collections
var result = ocr.Read(imagePath);
var lines = result.Lines;   // IEnumerable<OcrResult.OcrResultLine>
var words = result.Words;   // IEnumerable<OcrResult.OcrResultWord>
foreach (var word in result.Words)
    Console.WriteLine($"'{word.Text}' at ({word.X},{word.Y}) confidence {word.Confidence}%");

// Textract: filter flat block list by type
var lines = response.Blocks.Where(b => b.BlockType == BlockType.LINE);
var words = response.Blocks.Where(b => b.BlockType == BlockType.WORD);

// IronOCR: direct access to typed collections
var result = ocr.Read(imagePath);
var lines = result.Lines;   // IEnumerable<OcrResult.OcrResultLine>
var words = result.Words;   // IEnumerable<OcrResult.OcrResultWord>
foreach (var word in result.Words)
    Console.WriteLine($"'{word.Text}' at ({word.X},{word.Y}) confidence {word.Confidence}%");

' Textract: filter flat block list by type
Dim lines = response.Blocks.Where(Function(b) b.BlockType = BlockType.LINE)
Dim words = response.Blocks.Where(Function(b) b.BlockType = BlockType.WORD)

' IronOCR: direct access to typed collections
Dim result = ocr.Read(imagePath)
Dim lines = result.Lines   ' IEnumerable(Of OcrResult.OcrResultLine)
Dim words = result.Words   ' IEnumerable(Of OcrResult.OcrResultWord)
For Each word In result.Words
    Console.WriteLine($"'{word.Text}' at ({word.X},{word.Y}) confidence {word.Confidence}%")
Next

The structured results guide covers result.Pages, result.Paragraphs, result.Lines, result.Words, and coordinate access for building layout-aware document processing.

Replacing S3-Staged PDF Processing with Direct LoadPdf

Any Textract code that uploads to S3 before starting a detection job can be replaced with a direct PDF load. No staging bucket, no upload timing, no cleanup logic.

// Textract: upload to S3 → start job → poll → paginate → cleanup (50+ lines)
// IronOCR equivalent:
public string ProcessPdf(string pdfPath)
{
    var ocr = new IronTesseract();
    using var input = new OcrInput();
    input.LoadPdf(pdfPath);
    return ocr.Read(input).Text;
}

// Specific page ranges (no Textract equivalent without async job per range)
public string ProcessPdfPages(string pdfPath, int startPage, int endPage)
{
    var ocr = new IronTesseract();
    using var input = new OcrInput();
    input.LoadPdfPages(pdfPath, startPage, endPage);
    return ocr.Read(input).Text;
}

// Textract: upload to S3 → start job → poll → paginate → cleanup (50+ lines)
// IronOCR equivalent:
public string ProcessPdf(string pdfPath)
{
    var ocr = new IronTesseract();
    using var input = new OcrInput();
    input.LoadPdf(pdfPath);
    return ocr.Read(input).Text;
}

// Specific page ranges (no Textract equivalent without async job per range)
public string ProcessPdfPages(string pdfPath, int startPage, int endPage)
{
    var ocr = new IronTesseract();
    using var input = new OcrInput();
    input.LoadPdfPages(pdfPath, startPage, endPage);
    return ocr.Read(input).Text;
}

Imports IronOcr

Public Class PdfProcessor
    ' Textract: upload to S3 → start job → poll → paginate → cleanup (50+ lines)
    ' IronOCR equivalent:
    Public Function ProcessPdf(pdfPath As String) As String
        Dim ocr As New IronTesseract()
        Using input As New OcrInput()
            input.LoadPdf(pdfPath)
            Return ocr.Read(input).Text
        End Using
    End Function

    ' Specific page ranges (no Textract equivalent without async job per range)
    Public Function ProcessPdfPages(pdfPath As String, startPage As Integer, endPage As Integer) As String
        Dim ocr As New IronTesseract()
        Using input As New OcrInput()
            input.LoadPdfPages(pdfPath, startPage, endPage)
            Return ocr.Read(input).Text
        End Using
    End Function
End Class

Adding Preprocessing for Documents That Produced Low Confidence in Textract

Textract's preprocessing is internal and not configurable. When a scanned document produces poor results, the only options are retrying or accepting low-confidence output. IronOCR exposes the preprocessing pipeline directly.

// For documents that returned low-confidence results from Textract
using var input = new OcrInput();
input.LoadImage("low-quality-scan.jpg");

input.Deskew();              // Fix rotation from scanner misalignment
input.DeNoise();             // Remove scanner noise artifacts
input.Contrast();            // Boost faint text
input.EnhanceResolution(300); // Scale to optimal OCR resolution

var result = new IronTesseract().Read(input);
Console.WriteLine($"Confidence: {result.Confidence}%");

// For documents that returned low-confidence results from Textract
using var input = new OcrInput();
input.LoadImage("low-quality-scan.jpg");

input.Deskew();              // Fix rotation from scanner misalignment
input.DeNoise();             // Remove scanner noise artifacts
input.Contrast();            // Boost faint text
input.EnhanceResolution(300); // Scale to optimal OCR resolution

var result = new IronTesseract().Read(input);
Console.WriteLine($"Confidence: {result.Confidence}%");

Imports IronOcr

Dim input As New OcrInput()
input.LoadImage("low-quality-scan.jpg")

input.Deskew()              ' Fix rotation from scanner misalignment
input.DeNoise()             ' Remove scanner noise artifacts
input.Contrast()            ' Boost faint text
input.EnhanceResolution(300) ' Scale to optimal OCR resolution

Dim result = New IronTesseract().Read(input)
Console.WriteLine($"Confidence: {result.Confidence}%")

The image quality correction guide and image filters tutorial document the full preprocessing pipeline and combinations that work best for specific document types. For confidence score interpretation and per-element confidence access, the confidence scores guide covers the result.Confidence property and per-word confidence values.

Handling the Async-to-Synchronous Pattern Change

Existing Textract code is necessarily async Task<T> throughout because the SDK is async-only. IronOCR operations are synchronous. For application code that already has an async call chain, wrap the IronOCR call in Task.Run to keep the async boundary.

// Preserves async call site for minimal refactoring
public async Task<string> ExtractTextAsync(string path)
{
    return await Task.Run(() => new IronTesseract().Read(path).Text);
}

// Preserves async call site for minimal refactoring
public async Task<string> ExtractTextAsync(string path)
{
    return await Task.Run(() => new IronTesseract().Read(path).Text);
}

Imports System.Threading.Tasks

' Preserves async call site for minimal refactoring
Public Async Function ExtractTextAsync(path As String) As Task(Of String)
    Return Await Task.Run(Function() New IronTesseract().Read(path).Text)
End Function

This is a convenience wrapper, not a requirement. For server-side processing where the calling code is already on a background thread, the synchronous call is preferred directly.