基于iText7的PDF文本提取性能优化及代码合理性咨询

Hey there! Let's tackle your iText PDF text extraction optimization questions—both for multi-region processing and speeding up large document handling.

Extracting Text from Multiple Rectangles in a Single Page Pass #

You're right that iText supports extracting text from multiple regions in a single page parse—this is a huge performance win compared to calling GetTextFromPage once per region. Here's how to implement it, along with options to track text per region:

Basic Multi-Region Extraction (Combined Text) #

If you just need all text from your regions in one string, use FilteredTextRenderListener with multiple RegionTextRenderFilter instances:

// Example for iText 7 in C# (adjust syntax for Java if needed)
using (PdfDocument pdfDoc = new PdfDocument(new PdfReader("your-large-document.pdf")))
{
    for (int pageNum = 1; pageNum <= pdfDoc.GetNumberOfPages(); pageNum++)
    {
        PdfPage page = pdfDoc.GetPage(pageNum);
        
        // Define your target regions (x, y, width, height)
        Rectangle region1 = new Rectangle(100, 200, 300, 100);
        Rectangle region2 = new Rectangle(400, 200, 250, 150);
        Rectangle region3 = new Rectangle(100, 50, 550, 100);
        
        // Set up the filtered listener to target all regions at once
        ITextExtractionStrategy baseStrategy = new LocationTextExtractionStrategy();
        FilteredTextRenderListener filteredListener = new FilteredTextRenderListener(
            baseStrategy,
            new RegionTextRenderFilter(region1),
            new RegionTextRenderFilter(region2),
            new RegionTextRenderFilter(region3)
        );
        
        // Extract text in a single page pass
        string combinedText = PdfTextExtractor.GetTextFromPage(page, filteredListener);
        
        // Process your text here
        Console.WriteLine($"Page {pageNum} combined region text: {combinedText}");
    }
}

Track Text Per Region #

If you need to separate text from each region, create a custom extraction strategy to map text blocks to their respective regions:

public class MultiRegionTextStrategy : LocationTextExtractionStrategy
{
    private readonly Dictionary<Rectangle, StringBuilder> _regionTextMap;
    private readonly List<Rectangle> _targetRegions;

    public MultiRegionTextStrategy(List<Rectangle> targetRegions)
    {
        _targetRegions = targetRegions;
        _regionTextMap = targetRegions.ToDictionary(rect => rect, rect => new StringBuilder());
    }

    public override void RenderText(TextRenderInfo renderInfo)
    {
        base.RenderText(renderInfo);
        Rectangle textBounds = renderInfo.GetAscentLine().GetBoundingRectangle();
        
        // Match the text block to the first intersecting region
        foreach (var region in _targetRegions)
        {
            if (region.Intersects(textBounds))
            {
                _regionTextMap[region].Append(renderInfo.GetText());
                break; // Adjust if text might overlap multiple regions
            }
        }
    }

    // Get a dictionary of region-to-text mappings
    public Dictionary<Rectangle, string> GetRegionTexts()
    {
        return _regionTextMap.ToDictionary(kv => kv.Key, kv => kv.Value.ToString().Trim());
    }
}

Use this strategy like so:

using (PdfDocument pdfDoc = new PdfDocument(new PdfReader("your-large-document.pdf")))
{
    for (int pageNum = 1; pageNum <= pdfDoc.GetNumberOfPages(); pageNum++)
    {
        PdfPage page = pdfDoc.GetPage(pageNum);
        List<Rectangle> regions = new List<Rectangle> { region1, region2, region3 };
        
        MultiRegionTextStrategy regionStrategy = new MultiRegionTextStrategy(regions);
        FilteredTextRenderListener filteredListener = new FilteredTextRenderListener(regionStrategy);
        
        PdfTextExtractor.GetTextFromPage(page, filteredListener);
        
        // Access text per region
        var regionTexts = regionStrategy.GetRegionTexts();
        foreach (var pair in regionTexts)
        {
            Console.WriteLine($"Page {pageNum} - Region ({pair.Key}): {pair.Value}");
        }
    }
}

Performance Optimization Tips #

1. Single-Page Pass for Multiple Regions #

This is the biggest win: instead of parsing a page once per region, you parse it once. For pages with 3+ regions, this can cut per-page processing time by 50% or more.

2. Reuse Document/Reader Instances #

Never open and close the PdfDocument or PdfReader for each page. Keep the document open for the entire processing session to avoid redundant file I/O and initialization overhead.

3. Optimize Memory Usage #

For extremely large documents, use low-memory settings to reduce GC pressure:

ReaderProperties readerProps = new ReaderProperties()
    .SetMemoryUsageSetting(MemoryUsageSetting.MemoryUsageMode.LOW_MEMORY);
using (PdfReader reader = new PdfReader("your-large-document.pdf", readerProps))
using (PdfDocument pdfDoc = new PdfDocument(reader))
{
    // Process pages here
}

4. Parallel Processing (With Caution) #

If your system has available cores, process pages in parallel. Note that PdfDocument is not thread-safe, so create a separate reader/document instance per thread:

int totalPages = pdfDoc.GetNumberOfPages();
Parallel.For(1, totalPages + 1, pageNum =>
{
    using (var threadReader = new PdfReader("your-large-document.pdf"))
    using (var threadPdf = new PdfDocument(threadReader))
    {
        PdfPage page = threadPdf.GetPage(pageNum);
        // Perform extraction as shown earlier
    }
});

Test this first—parallel processing increases memory usage, so it's best suited for documents with many simple pages.

5. Use the Latest iText Version #

Newer iText releases include performance tweaks for text extraction and PDF parsing, so always upgrade to the latest stable version.

Expected Performance Improvements #

• For documents with multiple regions per page: Single-pass extraction can reduce total processing time by 30-70% compared to multi-pass extraction.
• Combining this with memory optimizations and parallel processing can easily bring your 10+ minute runtime down to 2-5 minutes for a 100+ page document (results vary based on PDF complexity).

内容的提问来源于stack exchange，提问作者SuperJMN