Historical Text Image Enhancement Using Image Scaling and Generative Adversarial Networks

Area of Science:

• Digital Image Processing
• Document Analysis
• Artificial Intelligence

Background:

• Historical documents often suffer from degraded text quality due to aging, watermarks, and stains.
• Poor text quality hinders document recognition, analysis, and preservation efforts.
• Effective text image enhancement is crucial for accessing and utilizing historical records.

Purpose of the Study:

• To develop an advanced method for enhancing degraded historical text images.
• To improve image resolution, de-noise, and de-blur historical documents.
• To enhance spectral and spatial features for clearer text representation.

Main Methods:

• A hybrid approach combining bi-cubic interpolation of Lifting Wavelet Transform (LWT) and Stationary Wavelet Transform (SWT) for image resolution enhancement.
• Utilizing a Generative Adversarial Network (GAN) to extract and fuse spectral and spatial features.
• A two-part process involving wavelet transformation for de-noising/de-blurring and GAN for feature fusion.

Main Results:

• The proposed method significantly enhances image resolution, effectively de-noising and de-blurring historical text images.
• The GAN component successfully fuses original and processed images, improving spectral and spatial features.
• Experimental results demonstrate superior performance compared to existing deep learning methods.

Conclusions:

• The integrated wavelet transform and GAN approach offers a robust solution for historical text image enhancement.
• This technique improves the legibility and analytical potential of degraded historical documents.
• The proposed model represents a significant advancement in digital document restoration and analysis.