Robust and accurate vectorization of line drawings

Area of Science:

• Computer Vision
• Image Processing
• Computational Geometry

Background:

• Digitizing historical or hand-drawn documents often requires converting raster images to vector formats.
• Existing vectorization methods may struggle with noise, varying line thicknesses, and complex structures in line drawings.

Purpose of the Study:

• To present a robust and accurate method for vectorizing graphical elements in paper-based line drawings.
• To improve the precision of vector parameter estimation and analyze the computational complexity of the proposed technique.

Main Methods:

• The method involves separating binary images into layers of homogeneous thickness.
• Skeletonization of each layer is performed, followed by segmentation using a random sampling-based approach.
• The resulting skeletons are simplified, and feasibility domains are computed for accurate parameter estimation.

Main Results:

• The segmentation method demonstrates robustness, handling up to 50% noise for long primitives.
• Accurate estimation of vector parameters is achieved through explicit computation of feasibility domains.
• Theoretical performance analysis and complexity expressions for the segmentation method are derived.

Conclusions:

• The proposed method offers an effective solution for the vectorization of line drawings, outperforming existing systems in certain aspects.
• The technique provides a robust framework for handling noisy and complex graphical data from scanned documents.
• Further experimental validation confirms the efficacy and accuracy of the developed vectorization approach.