Connectivity-enforcing Hough transform for the robust extraction of line segments

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Global Hough Transform (HT) methods offer robustness but fail in cluttered images due to lack of connectivity consideration.
• Local methods enforce connectivity but lack robustness against line crossings or corrupted segments in realistic scenarios.

Purpose of the Study:

• To address limitations of existing line detection methods by integrating connectivity into the Hough Transform voting process.
• To develop a robust line segment extraction technique that overcomes challenges in cluttered and complex images.

Main Methods:

• Introduced Segment Extraction by Connectivity-Enforcing HT (STRAIGHT), a novel approach incorporating connectivity into the HT voting mechanism.
• Edge points' contributions are considered based on their proximity and agreement in position and direction to potential line segments.
• A hierarchical implementation utilizing Hough space mapping ensures computational feasibility.

Main Results:

• STRAIGHT successfully extracts the longest connected line segments by integrating segment extraction into the HT voting process.
• Demonstrated superior performance over existing methods in extracting complete segments from synthetic and real-world cluttered images.
• The method effectively handles challenging scenarios like crossing lines and corrupted segments.

Conclusions:

• STRAIGHT offers a significant advancement in line segment extraction, combining the robustness of global methods with the connectivity of local approaches.
• The proposed technique provides a computationally feasible and effective solution for robust line detection in complex image data.
• This method paves the way for improved performance in various computer vision applications requiring accurate line segment identification.