An improved rotation-invariant thinning algorithm

Area of Science:

• Computer Vision
• Image Processing
• Computational Geometry

Background:

• Thinning algorithms are crucial for reducing binary images to single-pixel wide skeletons.
• Existing algorithms, like the one by Ahmed and Ward, aim for rotation-invariant thinning but have limitations.
• Failures in skeletonization can occur on specific image features, such as two-pixel wide lines.

Purpose of the Study:

• To identify and address the shortcomings of a recently proposed rotation-invariant thinning algorithm.
• To propose a modified thinning method that corrects identified failures.
• To ensure accurate skeleton generation from binary images, particularly for challenging line widths.

Main Methods:

• Analysis of the Ahmed and Ward thinning algorithm to pinpoint failure cases.
• Development of a modified thinning approach based on graph connectivity principles.
• Testing the modified algorithm on binary images with two-pixel wide lines.

Main Results:

• Demonstration of specific examples where the Ahmed and Ward algorithm fails.
• Successful correction of these failures using the proposed modified method.
• Validation of the modified algorithm's ability to produce accurate single-pixel wide skeletons.

Conclusions:

• The proposed modification effectively resolves the identified shortcomings in the original thinning algorithm.
• Graph connectivity provides a robust basis for improving skeletonization accuracy.
• The enhanced algorithm offers a more reliable solution for generating single-pixel wide skeletons from binary images.