Spaghetti Labeling: Directed Acyclic Graphs for Block-Based Connected Components Labeling.

Area of Science:

• Computer Vision
• Image Processing
• Algorithm Optimization

Background:

• Connected Components Labeling (CCL) is fundamental to image analysis.
• Existing CCL methods face computational challenges, especially with large datasets.
• Recent strategies like decision forests and state prediction show promise but have limitations.

Purpose of the Study:

• To develop a more efficient and scalable Connected Components Labeling algorithm.
• To overcome the limitations of manual state construction and large code size in prior methods.
• To integrate block-based processing with state prediction for enhanced performance.

Main Methods:

• A novel algorithm combining block-based masks, state prediction, and code compression.
• Modeling the algorithm as a Directed Rooted Acyclic Graph (DRAG) with multiple entry points.
• Automatic generation of the DRAG without manual intervention.

Main Results:

• The proposed approach demonstrates superior performance compared to state-of-the-art algorithms.
• Outperforms existing methods on both synthetic and real-world datasets.
• Achieves significant improvements across all tested settings.

Conclusions:

• The integrated block-based and state prediction method offers a significant advancement in CCL.
• The automatically generated DRAG model provides an efficient and scalable solution.
• This work paves the way for more effective image processing and computer vision applications.