Composition of a dewarped and enhanced document image from two view images

Area of Science:

• Computer Vision
• Image Processing
• Document Analysis

Background:

• Conventional methods for document image enhancement often require specialized equipment or make assumptions about document content.
• Capturing high-quality document images with digital cameras at different angles presents challenges like geometric distortion and visual artifacts.

Purpose of the Study:

• To develop an algorithm for composing geometrically dewarped and visually enhanced images from two document images taken at different angles.
• To enable high-quality digitization of documents, including text and pictures, without content-specific assumptions.

Main Methods:

• Estimating the unfolded document surface using corresponding points between two images.
• Employing structure reconstruction, 3-D projection analysis, and random sample consensus (RANSAC) with a cylindrical surface model.
• Utilizing image mosaicking via graph cut-based energy minimization to select optimal image regions, reducing blur and reflections.

Main Results:

• The proposed algorithm robustly generates visually pleasing document images.
• The resulting images achieve an optical character recognition (OCR) rate comparable to that of images from a flatbed scanner.
• The method successfully handles geometric dewarping and visual enhancement, including mosaic composition.

Conclusions:

• The algorithm offers a practical solution for high-quality document digitization using standard digital cameras.
• It overcomes limitations of conventional methods by avoiding special equipment and content assumptions.
• The approach is versatile, applicable to both text and image digitization, and improves OCR accuracy.