Using Paper Texture for Choosing a Suitable Algorithm for Scanned Document Image Binarization

Area of Science:

• Document image analysis
• Computer vision
• Digital image processing

Background:

• Intrinsic document features (paper color, texture, aging, translucency, print/handwriting) influence image processing and enhancement.
• Image binarization, converting color images to monochrome, is a critical step in document processing pipelines.
• Recent binarization competitions reveal no single algorithm excels across all document types.

Purpose of the Study:

• To investigate the impact of historical document texture on image binarization performance.
• To identify the optimal document image-binarization scheme balancing quality and processing time.
• To provide a method for selecting appropriate binarization algorithms based on document characteristics.

Main Methods:

• Utilized scanned historical documents, focusing on texture as a primary feature.
• Evaluated 63 widely used binarization algorithms.
• Tested five different input image versions for each algorithm, resulting in 315 total schemes.
• Assessed schemes based on a quality-time trade-off.

Main Results:

• Document texture significantly impacts the effectiveness of different binarization algorithms.
• A specific subset of algorithms demonstrated superior quality-time trade-offs for certain document textures.
• The selection of an algorithm from the 315 schemes depends heavily on the input document's intrinsic features.

Conclusions:

• Document texture is a key feature for selecting effective image binarization methods.
• A tailored approach to binarization, considering document characteristics, is necessary for optimal results.
• This research provides a framework for choosing the best binarization scheme for historical document image processing.