// RapidOcrNet: 4 external files required before any OCR can execute
var engine = new RapidOcrEngine(new RapidOcrOptions
{
    DetModelPath = "./models/det.onnx",     // ~3 MB — downloaded from GitHub
    ClsModelPath = "./models/cls.onnx",     // ~1 MB — downloaded from GitHub
    RecModelPath = "./models/rec_en.onnx",  // ~2-10 MB — language-specific download
    KeysPath     = "./models/en_keys.txt"   // character dictionary — language-specific
});

// RapidOcrNet: 4 external files required before any OCR can execute
var engine = new RapidOcrEngine(new RapidOcrOptions
{
    DetModelPath = "./models/det.onnx",     // ~3 MB — downloaded from GitHub
    ClsModelPath = "./models/cls.onnx",     // ~1 MB — downloaded from GitHub
    RecModelPath = "./models/rec_en.onnx",  // ~2-10 MB — language-specific download
    KeysPath     = "./models/en_keys.txt"   // character dictionary — language-specific
});

// IronOCR: install the NuGet package, write one line
IronOcr.License.LicenseKey = "YOUR-LICENSE-KEY";
var text = new IronTesseract().Read("document.jpg").Text;

// IronOCR: install the NuGet package, write one line
IronOcr.License.LicenseKey = "YOUR-LICENSE-KEY";
var text = new IronTesseract().Read("document.jpg").Text;

dotnet remove package RapidOcrNet
dotnet remove package Microsoft.ML.OnnxRuntime

dotnet remove package RapidOcrNet
dotnet remove package Microsoft.ML.OnnxRuntime

// Before (RapidOCR.NET)
using RapidOcrNet;

// After (IronOCR)
using IronOcr;

// Before (RapidOCR.NET)
using RapidOcrNet;

// After (IronOCR)
using IronOcr;

using RapidOcrNet;

// Startup validation — written because a missing model crashes at runtime, not at install
private static void EnsureModelsPresent(string modelDir)
{
    var required = new[]
    {
        Path.Combine(modelDir, "det.onnx"),
        Path.Combine(modelDir, "cls.onnx"),
        Path.Combine(modelDir, "rec_en.onnx"),
        Path.Combine(modelDir, "en_keys.txt")
    };

    var missing = required.Where(f => !File.Exists(f)).ToList();
    if (missing.Any())
        throw new FileNotFoundException(
            $"Missing model files: {string.Join(", ", missing)}\n" +
            "Download from: https://github.com/RapidAI/RapidOCR/releases");
}

// Engine factory — called once at startup, held for lifetime of service
public RapidOcrEngine CreateEngine(string modelDir)
{
    EnsureModelsPresent(modelDir);
    return new RapidOcrEngine(new RapidOcrOptions
    {
        DetModelPath = Path.Combine(modelDir, "det.onnx"),
        ClsModelPath = Path.Combine(modelDir, "cls.onnx"),
        RecModelPath = Path.Combine(modelDir, "rec_en.onnx"),
        KeysPath     = Path.Combine(modelDir, "en_keys.txt"),
        UseGpu       = false,
        NumThreads   = Environment.ProcessorCount
    });
}

using RapidOcrNet;

// Startup validation — written because a missing model crashes at runtime, not at install
private static void EnsureModelsPresent(string modelDir)
{
    var required = new[]
    {
        Path.Combine(modelDir, "det.onnx"),
        Path.Combine(modelDir, "cls.onnx"),
        Path.Combine(modelDir, "rec_en.onnx"),
        Path.Combine(modelDir, "en_keys.txt")
    };

    var missing = required.Where(f => !File.Exists(f)).ToList();
    if (missing.Any())
        throw new FileNotFoundException(
            $"Missing model files: {string.Join(", ", missing)}\n" +
            "Download from: https://github.com/RapidAI/RapidOCR/releases");
}

// Engine factory — called once at startup, held for lifetime of service
public RapidOcrEngine CreateEngine(string modelDir)
{
    EnsureModelsPresent(modelDir);
    return new RapidOcrEngine(new RapidOcrOptions
    {
        DetModelPath = Path.Combine(modelDir, "det.onnx"),
        ClsModelPath = Path.Combine(modelDir, "cls.onnx"),
        RecModelPath = Path.Combine(modelDir, "rec_en.onnx"),
        KeysPath     = Path.Combine(modelDir, "en_keys.txt"),
        UseGpu       = false,
        NumThreads   = Environment.ProcessorCount
    });
}

using IronOcr;

// Non model validation, no path configuration, no GPU flags
// IronTesseract is thread-safe; create one per thread or on demand
IronOcr.License.LicenseKey = "YOUR-LICENSE-KEY";
var ocr = new IronTesseract();

using IronOcr;

// Non model validation, no path configuration, no GPU flags
// IronTesseract is thread-safe; create one per thread or on demand
IronOcr.License.LicenseKey = "YOUR-LICENSE-KEY";
var ocr = new IronTesseract();

using RapidOcrNet;

public class InvoiceTextExtractor
{
    private readonly RapidOcrEngine _engine;

    public InvoiceTextExtractor(string modelDir)
    {
        // Three separate ONNX models run in sequence on every call
        _engine = new RapidOcrEngine(new RapidOcrOptions
        {
            DetModelPath = Path.Combine(modelDir, "det.onnx"),   // Stage 1: detect text regions
            ClsModelPath = Path.Combine(modelDir, "cls.onnx"),   // Stage 2: classify direction
            RecModelPath = Path.Combine(modelDir, "rec_en.onnx"),// Stage 3: recognize characters
            KeysPath     = Path.Combine(modelDir, "en_keys.txt")
        });
    }

    public string ExtractInvoiceText(string imagePath)
    {
        var result = _engine.Run(imagePath);

        // Blocks are unordered — must sort by vertical position, then horizontal
        var orderedBlocks = result.TextBlocks
            .OrderBy(b => b.BoundingBox.Top)
            .ThenBy(b => b.BoundingBox.Left)
            .ToList();

        // Manual assembly — no paragraph or line structure
        return string.Join(Environment.NewLine,
            orderedBlocks.Select(b => b.Text));
    }
}

using RapidOcrNet;

public class InvoiceTextExtractor
{
    private readonly RapidOcrEngine _engine;

    public InvoiceTextExtractor(string modelDir)
    {
        // Three separate ONNX models run in sequence on every call
        _engine = new RapidOcrEngine(new RapidOcrOptions
        {
            DetModelPath = Path.Combine(modelDir, "det.onnx"),   // Stage 1: detect text regions
            ClsModelPath = Path.Combine(modelDir, "cls.onnx"),   // Stage 2: classify direction
            RecModelPath = Path.Combine(modelDir, "rec_en.onnx"),// Stage 3: recognize characters
            KeysPath     = Path.Combine(modelDir, "en_keys.txt")
        });
    }

    public string ExtractInvoiceText(string imagePath)
    {
        var result = _engine.Run(imagePath);

        // Blocks are unordered — must sort by vertical position, then horizontal
        var orderedBlocks = result.TextBlocks
            .OrderBy(b => b.BoundingBox.Top)
            .ThenBy(b => b.BoundingBox.Left)
            .ToList();

        // Manual assembly — no paragraph or line structure
        return string.Join(Environment.NewLine,
            orderedBlocks.Select(b => b.Text));
    }
}

using IronOcr;

public class InvoiceTextExtractor
{
    private readonly IronTesseract _ocr = new IronTesseract();

    public string ExtractInvoiceText(string imagePath)
    {
        // Single call — detection, recognition, reading order all internal
        var result = _ocr.Read(imagePath);
        return result.Text; // Already in reading order
    }

    public IEnumerable<string> ExtractInvoiceParagraphs(string imagePath)
    {
        var result = _ocr.Read(imagePath);
        // Structured paragraphs with coordinates — no sorting or assembly needed
        foreach (var page in result.Pages)
            foreach (var paragraph in page.Paragraphs)
                yield return paragraph.Text;
    }
}

using IronOcr;

public class InvoiceTextExtractor
{
    private readonly IronTesseract _ocr = new IronTesseract();

    public string ExtractInvoiceText(string imagePath)
    {
        // Single call — detection, recognition, reading order all internal
        var result = _ocr.Read(imagePath);
        return result.Text; // Already in reading order
    }

    public IEnumerable<string> ExtractInvoiceParagraphs(string imagePath)
    {
        var result = _ocr.Read(imagePath);
        // Structured paragraphs with coordinates — no sorting or assembly needed
        foreach (var page in result.Pages)
            foreach (var paragraph in page.Paragraphs)
                yield return paragraph.Text;
    }
}

using RapidOcrNet;

public class DocumentRouter
{
    private readonly string _modelDir;

    public DocumentRouter(string modelDir) => _modelDir = modelDir;

    // Must create separate engine instances per configuration
    // Each engine holds ~300-500 MB of loaded model weights
    private RapidOcrEngine BuildEnglishEngine() =>
        new RapidOcrEngine(new RapidOcrOptions
        {
            DetModelPath = Path.Combine(_modelDir, "det.onnx"),
            ClsModelPath = Path.Combine(_modelDir, "cls.onnx"),
            RecModelPath = Path.Combine(_modelDir, "en_rec.onnx"),
            KeysPath     = Path.Combine(_modelDir, "en_keys.txt")
        });

    private RapidOcrEngine BuildChineseEngine() =>
        new RapidOcrEngine(new RapidOcrOptions
        {
            DetModelPath = Path.Combine(_modelDir, "det.onnx"),
            ClsModelPath = Path.Combine(_modelDir, "cls.onnx"),
            RecModelPath = Path.Combine(_modelDir, "ch_rec.onnx"),  // separate download
            KeysPath     = Path.Combine(_modelDir, "ch_keys.txt")   // separate download
        });

    public string ProcessDocument(string imagePath, string language)
    {
        // Rebuild engine for each language — model reload cost on every switch
        using var engine = language == "chinese"
            ? BuildChineseEngine()
            : BuildEnglishEngine();

        var result = engine.Run(imagePath);
        return string.Join("\n", result.TextBlocks
            .OrderBy(b => b.BoundingBox.Top)
            .Select(b => b.Text));
    }
}

using RapidOcrNet;

public class DocumentRouter
{
    private readonly string _modelDir;

    public DocumentRouter(string modelDir) => _modelDir = modelDir;

    // Must create separate engine instances per configuration
    // Each engine holds ~300-500 MB of loaded model weights
    private RapidOcrEngine BuildEnglishEngine() =>
        new RapidOcrEngine(new RapidOcrOptions
        {
            DetModelPath = Path.Combine(_modelDir, "det.onnx"),
            ClsModelPath = Path.Combine(_modelDir, "cls.onnx"),
            RecModelPath = Path.Combine(_modelDir, "en_rec.onnx"),
            KeysPath     = Path.Combine(_modelDir, "en_keys.txt")
        });

    private RapidOcrEngine BuildChineseEngine() =>
        new RapidOcrEngine(new RapidOcrOptions
        {
            DetModelPath = Path.Combine(_modelDir, "det.onnx"),
            ClsModelPath = Path.Combine(_modelDir, "cls.onnx"),
            RecModelPath = Path.Combine(_modelDir, "ch_rec.onnx"),  // separate download
            KeysPath     = Path.Combine(_modelDir, "ch_keys.txt")   // separate download
        });

    public string ProcessDocument(string imagePath, string language)
    {
        // Rebuild engine for each language — model reload cost on every switch
        using var engine = language == "chinese"
            ? BuildChineseEngine()
            : BuildEnglishEngine();

        var result = engine.Run(imagePath);
        return string.Join("\n", result.TextBlocks
            .OrderBy(b => b.BoundingBox.Top)
            .Select(b => b.Text));
    }
}

using IronOcr;

public class DocumentRouter
{
    // One instance handles all languages — language is a property, not a constructor param
    private readonly IronTesseract _ocr = new IronTesseract();

    public string ProcessDocument(string imagePath, string language)
    {
        // Language switch requires no model reload, no rebuild
        _ocr.Language = language switch
        {
            "chinese"  => OcrLanguage.ChineseSimplified,
            "japanese" => OcrLanguage.Japanese,
            "arabic"   => OcrLanguage.Arabic,
            "russian"  => OcrLanguage.Russian,
            _          => OcrLanguage.English
        };

        return _ocr.Read(imagePath).Text;
    }
}

using IronOcr;

public class DocumentRouter
{
    // One instance handles all languages — language is a property, not a constructor param
    private readonly IronTesseract _ocr = new IronTesseract();

    public string ProcessDocument(string imagePath, string language)
    {
        // Language switch requires no model reload, no rebuild
        _ocr.Language = language switch
        {
            "chinese"  => OcrLanguage.ChineseSimplified,
            "japanese" => OcrLanguage.Japanese,
            "arabic"   => OcrLanguage.Arabic,
            "russian"  => OcrLanguage.Russian,
            _          => OcrLanguage.English
        };

        return _ocr.Read(imagePath).Text;
    }
}

using RapidOcrNet;

public class BatchOcrProcessor
{
    private readonly string _modelDir;

    public BatchOcrProcessor(string modelDir) => _modelDir = modelDir;

    // Thread-pool processing — each thread needs its own engine copy
    // 4 threads × 300-500 MB model footprint = 1.2-2 GB RAM minimum
    public Dictionary<string, string> ProcessBatch(IReadOnlyList<string> imagePaths)
    {
        var results = new System.Collections.Concurrent.ConcurrentDictionary<string, string>();

        Parallel.ForEach(imagePaths, new ParallelOptions { MaxDegreeOfParallelism = 4 },
            imagePath =>
            {
                // Each thread must create its own engine — not safe to share
                using var engine = new RapidOcrEngine(new RapidOcrOptions
                {
                    DetModelPath = Path.Combine(_modelDir, "det.onnx"),
                    ClsModelPath = Path.Combine(_modelDir, "cls.onnx"),
                    RecModelPath = Path.Combine(_modelDir, "rec_en.onnx"),
                    KeysPath     = Path.Combine(_modelDir, "en_keys.txt")
                });

                var result = engine.Run(imagePath);
                results[imagePath] = string.Join("\n",
                    result.TextBlocks
                          .OrderBy(b => b.BoundingBox.Top)
                          .Select(b => b.Text));
            });

        return new Dictionary<string, string>(results);
    }
}

using RapidOcrNet;

public class BatchOcrProcessor
{
    private readonly string _modelDir;

    public BatchOcrProcessor(string modelDir) => _modelDir = modelDir;

    // Thread-pool processing — each thread needs its own engine copy
    // 4 threads × 300-500 MB model footprint = 1.2-2 GB RAM minimum
    public Dictionary<string, string> ProcessBatch(IReadOnlyList<string> imagePaths)
    {
        var results = new System.Collections.Concurrent.ConcurrentDictionary<string, string>();

        Parallel.ForEach(imagePaths, new ParallelOptions { MaxDegreeOfParallelism = 4 },
            imagePath =>
            {
                // Each thread must create its own engine — not safe to share
                using var engine = new RapidOcrEngine(new RapidOcrOptions
                {
                    DetModelPath = Path.Combine(_modelDir, "det.onnx"),
                    ClsModelPath = Path.Combine(_modelDir, "cls.onnx"),
                    RecModelPath = Path.Combine(_modelDir, "rec_en.onnx"),
                    KeysPath     = Path.Combine(_modelDir, "en_keys.txt")
                });

                var result = engine.Run(imagePath);
                results[imagePath] = string.Join("\n",
                    result.TextBlocks
                          .OrderBy(b => b.BoundingBox.Top)
                          .Select(b => b.Text));
            });

        return new Dictionary<string, string>(results);
    }
}

using IronOcr;

public class BatchOcrProcessor
{
    // Thread-safe: create IronTesseract per thread, no shared state required
    public Dictionary<string, string> ProcessBatch(IReadOnlyList<string> imagePaths)
    {
        var results = new System.Collections.Concurrent.ConcurrentDictionary<string, string>();

        Parallel.ForEach(imagePaths, imagePath =>
        {
            // Lightweight construction — no model loading overhead per thread
            var ocr = new IronTesseract();
            var result = ocr.Read(imagePath);
            results[imagePath] = result.Text;
        });

        return new Dictionary<string, string>(results);
    }
}

using IronOcr;

public class BatchOcrProcessor
{
    // Thread-safe: create IronTesseract per thread, no shared state required
    public Dictionary<string, string> ProcessBatch(IReadOnlyList<string> imagePaths)
    {
        var results = new System.Collections.Concurrent.ConcurrentDictionary<string, string>();

        Parallel.ForEach(imagePaths, imagePath =>
        {
            // Lightweight construction — no model loading overhead per thread
            var ocr = new IronTesseract();
            var result = ocr.Read(imagePath);
            results[imagePath] = result.Text;
        });

        return new Dictionary<string, string>(results);
    }
}

using RapidOcrNet;
// Also requires: SixLabors.ImageSharp or System.Drawing for TIFF frame extraction

public class TiffOcrProcessor
{
    private readonly RapidOcrEngine _engine;

    public TiffOcrProcessor(string modelDir)
    {
        _engine = new RapidOcrEngine(new RapidOcrOptions
        {
            DetModelPath = Path.Combine(modelDir, "det.onnx"),
            ClsModelPath = Path.Combine(modelDir, "cls.onnx"),
            RecModelPath = Path.Combine(modelDir, "rec_en.onnx"),
            KeysPath     = Path.Combine(modelDir, "en_keys.txt")
        });
    }

    public string ProcessMultiPageTiff(string tiffPath)
    {
        var pageTexts = new List<string>();
        var tempDir = Path.Combine(Path.GetTempPath(), Guid.NewGuid().ToString());
        Directory.CreateDirectory(tempDir);

        try
        {
            // External library required to split TIFF frames
            var framePaths = SplitTiffIntoFrames(tiffPath, tempDir); // not in RapidOcrNet

            foreach (var framePath in framePaths)
            {
                var result = _engine.Run(framePath);
                pageTexts.Add(string.Join("\n",
                    result.TextBlocks
                          .OrderBy(b => b.BoundingBox.Top)
                          .Select(b => b.Text)));
            }
        }
        finally
        {
            // Clean up temp frame files
            Directory.Delete(tempDir, recursive: true);
        }

        return string.Join("\n\n", pageTexts);
    }

    private IEnumerable<string> SplitTiffIntoFrames(string tiffPath, string outputDir)
    {
        // Requires external library — implementation depends on what is installed
        throw new NotImplementedException("Add SixLabors.ImageSharp or similar");
    }
}

using RapidOcrNet;
// Also requires: SixLabors.ImageSharp or System.Drawing for TIFF frame extraction

public class TiffOcrProcessor
{
    private readonly RapidOcrEngine _engine;

    public TiffOcrProcessor(string modelDir)
    {
        _engine = new RapidOcrEngine(new RapidOcrOptions
        {
            DetModelPath = Path.Combine(modelDir, "det.onnx"),
            ClsModelPath = Path.Combine(modelDir, "cls.onnx"),
            RecModelPath = Path.Combine(modelDir, "rec_en.onnx"),
            KeysPath     = Path.Combine(modelDir, "en_keys.txt")
        });
    }

    public string ProcessMultiPageTiff(string tiffPath)
    {
        var pageTexts = new List<string>();
        var tempDir = Path.Combine(Path.GetTempPath(), Guid.NewGuid().ToString());
        Directory.CreateDirectory(tempDir);

        try
        {
            // External library required to split TIFF frames
            var framePaths = SplitTiffIntoFrames(tiffPath, tempDir); // not in RapidOcrNet

            foreach (var framePath in framePaths)
            {
                var result = _engine.Run(framePath);
                pageTexts.Add(string.Join("\n",
                    result.TextBlocks
                          .OrderBy(b => b.BoundingBox.Top)
                          .Select(b => b.Text)));
            }
        }
        finally
        {
            // Clean up temp frame files
            Directory.Delete(tempDir, recursive: true);
        }

        return string.Join("\n\n", pageTexts);
    }

    private IEnumerable<string> SplitTiffIntoFrames(string tiffPath, string outputDir)
    {
        // Requires external library — implementation depends on what is installed
        throw new NotImplementedException("Add SixLabors.ImageSharp or similar");
    }
}

using IronOcr;

public class TiffOcrProcessor
{
    private readonly IronTesseract _ocr = new IronTesseract();

    public string ProcessMultiPageTiff(string tiffPath)
    {
        using var input = new OcrInput();
        input.LoadImageFrames(tiffPath); // All frames loaded — no external library needed

        var result = _ocr.Read(input);
        return result.Text; // Pages assembled in order automatically
    }

    public IEnumerable<(int PageNumber, string Text, double Confidence)> ProcessTiffWithPageData(string tiffPath)
    {
        using var input = new OcrInput();
        input.LoadImageFrames(tiffPath);

        var result = _ocr.Read(input);

        foreach (var page in result.Pages)
            yield return (page.PageNumber, page.Text, page.Confidence);
    }
}

using IronOcr;

public class TiffOcrProcessor
{
    private readonly IronTesseract _ocr = new IronTesseract();

    public string ProcessMultiPageTiff(string tiffPath)
    {
        using var input = new OcrInput();
        input.LoadImageFrames(tiffPath); // All frames loaded — no external library needed

        var result = _ocr.Read(input);
        return result.Text; // Pages assembled in order automatically
    }

    public IEnumerable<(int PageNumber, string Text, double Confidence)> ProcessTiffWithPageData(string tiffPath)
    {
        using var input = new OcrInput();
        input.LoadImageFrames(tiffPath);

        var result = _ocr.Read(input);

        foreach (var page in result.Pages)
            yield return (page.PageNumber, page.Text, page.Confidence);
    }
}

using RapidOcrNet;

public class FormFieldExtractor
{
    private readonly RapidOcrEngine _engine;

    public FormFieldExtractor(string modelDir)
    {
        _engine = new RapidOcrEngine(new RapidOcrOptions
        {
            DetModelPath = Path.Combine(modelDir, "det.onnx"),
            ClsModelPath = Path.Combine(modelDir, "cls.onnx"),
            RecModelPath = Path.Combine(modelDir, "rec_en.onnx"),
            KeysPath     = Path.Combine(modelDir, "en_keys.txt")
        });
    }

    // Extract text within a defined region by filtering block coordinates manually
    public string ExtractFieldByRegion(string imagePath, float regionLeft, float regionTop,
                                        float regionRight, float regionBottom)
    {
        var result = _engine.Run(imagePath);

        // Filter blocks whose bounding box intersects the target region
        var blocksInRegion = result.TextBlocks
            .Where(b =>
                b.BoundingBox.Left   < regionRight  &&
                b.BoundingBox.Right  > regionLeft   &&
                b.BoundingBox.Top    < regionBottom &&
                b.BoundingBox.Bottom > regionTop)
            .OrderBy(b => b.BoundingBox.Top)
            .ThenBy(b => b.BoundingBox.Left);

        return string.Join(" ", blocksInRegion.Select(b => b.Text));
    }
}

using RapidOcrNet;

public class FormFieldExtractor
{
    private readonly RapidOcrEngine _engine;

    public FormFieldExtractor(string modelDir)
    {
        _engine = new RapidOcrEngine(new RapidOcrOptions
        {
            DetModelPath = Path.Combine(modelDir, "det.onnx"),
            ClsModelPath = Path.Combine(modelDir, "cls.onnx"),
            RecModelPath = Path.Combine(modelDir, "rec_en.onnx"),
            KeysPath     = Path.Combine(modelDir, "en_keys.txt")
        });
    }

    // Extract text within a defined region by filtering block coordinates manually
    public string ExtractFieldByRegion(string imagePath, float regionLeft, float regionTop,
                                        float regionRight, float regionBottom)
    {
        var result = _engine.Run(imagePath);

        // Filter blocks whose bounding box intersects the target region
        var blocksInRegion = result.TextBlocks
            .Where(b =>
                b.BoundingBox.Left   < regionRight  &&
                b.BoundingBox.Right  > regionLeft   &&
                b.BoundingBox.Top    < regionBottom &&
                b.BoundingBox.Bottom > regionTop)
            .OrderBy(b => b.BoundingBox.Top)
            .ThenBy(b => b.BoundingBox.Left);

        return string.Join(" ", blocksInRegion.Select(b => b.Text));
    }
}

using IronOcr;

public class FormFieldExtractor
{
    private readonly IronTesseract _ocr = new IronTesseract();

    // Use CropRectangle to OCR only the target region — no post-filter needed
    public string ExtractFieldByRegion(string imagePath, int x, int y, int width, int height)
    {
        var region = new CropRectangle(x, y, width, height);

        using var input = new OcrInput();
        input.LoadImage(imagePath, region);

        return _ocr.Read(input).Text;
    }

    // Extract all fields with their coordinates from a full-page scan
    public IEnumerable<(string Text, int X, int Y, double Confidence)> ExtractAllWords(string imagePath)
    {
        var result = _ocr.Read(imagePath);

        foreach (var page in result.Pages)
            foreach (var word in page.Words)
                yield return (word.Text, word.X, word.Y, word.Confidence);
    }
}

using IronOcr;

public class FormFieldExtractor
{
    private readonly IronTesseract _ocr = new IronTesseract();

    // Use CropRectangle to OCR only the target region — no post-filter needed
    public string ExtractFieldByRegion(string imagePath, int x, int y, int width, int height)
    {
        var region = new CropRectangle(x, y, width, height);

        using var input = new OcrInput();
        input.LoadImage(imagePath, region);

        return _ocr.Read(input).Text;
    }

    // Extract all fields with their coordinates from a full-page scan
    public IEnumerable<(string Text, int X, int Y, double Confidence)> ExtractAllWords(string imagePath)
    {
        var result = _ocr.Read(imagePath);

        foreach (var page in result.Pages)
            foreach (var word in page.Words)
                yield return (word.Text, word.X, word.Y, word.Confidence);
    }
}

<!-- Remove this entire block from .csproj -->
<ItemGroup>
  <Content Include="models\**\*.*">
    <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
  </Content>
</ItemGroup>

<!-- Remove this entire block from .csproj -->
<ItemGroup>
  <Content Include="models\**\*.*">
    <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
  </Content>
</ItemGroup>

var ocr = new IronTesseract();
ocr.Language = OcrLanguage.Spanish;
var result = ocr.Read("spanish-document.jpg");

var ocr = new IronTesseract();
ocr.Language = OcrLanguage.Spanish;
var result = ocr.Read("spanish-document.jpg");

// Before: manual sort + coordinate extraction
var sorted = result.TextBlocks
    .OrderBy(b => b.BoundingBox.Top)
    .ThenBy(b => b.BoundingBox.Left);

foreach (var block in sorted)
    Console.WriteLine($"{block.Text} at ({block.BoundingBox.Left}, {block.BoundingBox.Top})");

// After: structured access, already in order
foreach (var page in result.Pages)
    foreach (var word in page.Words)
        Console.WriteLine($"{word.Text} at ({word.X}, {word.Y})");

// Before: manual sort + coordinate extraction
var sorted = result.TextBlocks
    .OrderBy(b => b.BoundingBox.Top)
    .ThenBy(b => b.BoundingBox.Left);

foreach (var block in sorted)
    Console.WriteLine($"{block.Text} at ({block.BoundingBox.Left}, {block.BoundingBox.Top})");

// After: structured access, already in order
foreach (var page in result.Pages)
    foreach (var word in page.Words)
        Console.WriteLine($"{word.Text} at ({word.X}, {word.Y})");

# Find all files that reference RapidOcrNet
grep -r "RapidOcrNet\|RapidOcrEngine\|RapidOcrOptions" --include="*.cs" .

# Find model path configuration
grep -r "DetModelPath\|ClsModelPath\|RecModelPath\|KeysPath" --include="*.cs" .

# Find MSBuild model copy entries
grep -r "det\.onnx\|cls\.onnx\|rec.*\.onnx\|keys\.txt" --include="*.csproj" .

# Find model validation logic
grep -r "ValidateModel\|models/" --include="*.cs" .

# Find ONNX Runtime references
grep -r "OnnxRuntime\|Microsoft\.ML" --include="*.csproj" .

# Find language-switching patterns (multiple engine instances per language)
grep -r "CreateEnglishEngine\|CreateChineseEngine\|rec_en\|ch_rec\|en_keys\|ch_keys" --include="*.cs" .

# Find all files that reference RapidOcrNet
grep -r "RapidOcrNet\|RapidOcrEngine\|RapidOcrOptions" --include="*.cs" .

# Find model path configuration
grep -r "DetModelPath\|ClsModelPath\|RecModelPath\|KeysPath" --include="*.cs" .

# Find MSBuild model copy entries
grep -r "det\.onnx\|cls\.onnx\|rec.*\.onnx\|keys\.txt" --include="*.csproj" .

# Find model validation logic
grep -r "ValidateModel\|models/" --include="*.cs" .

# Find ONNX Runtime references
grep -r "OnnxRuntime\|Microsoft\.ML" --include="*.csproj" .

# Find language-switching patterns (multiple engine instances per language)
grep -r "CreateEnglishEngine\|CreateChineseEngine\|rec_en\|ch_rec\|en_keys\|ch_keys" --include="*.cs" .