IronOCR on Windows: DeNoise() Amplifies Background Noise into Speckle (Colored-Paper Scans)

Overview

DeNoise() (aliased internally as Despeckle()) is built for one job: removing small, isolated dark specks on a clean white background. It doesn't handle continuous texture.

Colored-paper documents scanned as black and white lose their color contrast and turn into a uniform gray grain across the whole page. DeNoise()'s fixed threshold misreads that grain as widespread speckle — so it makes the image noisier, not cleaner.

Fix: use AdaptiveThreshold() instead of DeNoise().

Environment

• OS: Windows 11
• Language/Runtime: C# (.NET 8 WinForms)

Version Metadata

• Version found: 2026.6.1
• Version resolved: Unknown

Cause

DeNoise() is built only for isolated dark specks on an otherwise clean background — it does not adapt to continuous texture. Colored-paper pages scanned as black and white lose their color contrast and become a uniform gray-toned grain across the whole page, which the fixed global threshold misreads as widespread speckle rather than isolated noise. Internal notes treat this as a design limitation of DeNoise() for this input type, not an environment-specific bug.

Solution

1. Recommended: Replace DeNoise() with AdaptiveThreshold() for this document type:

   using (var input = new OcrInput())
   {
   
       foreach (var imagePath in pageImagePaths)
       {
           input.LoadImage(imagePath);
       }
   
       // Replaces input.DeNoise() for this document type
       input.AdaptiveThreshold(0.15f);
   
       var ocrResult = ocr.Read(input);
   
       ocrResult.SaveAsSearchablePdf(outputPath);
   }

   Suggested tuning range for thresholdLimit: 0.10 – 0.20.
2. If AdaptiveThreshold() alone doesn't fully clean up other documents in your batch, apply Contrast() first to widen the separation between text and background before binarizing:

   input.Contrast(1.3f); // try a range of 1.3f–1.4f
   
   input.AdaptiveThreshold(0.15f);

    

Workarounds That Don't Work

• Applying DeNoise() alone, with no Deskew or other filter in the pipeline — confirmed ineffective; it worsens the image by amplifying background noise.