Block-Based Connected-Component Labeling Algorithm Using Binary Decision Trees

Area of Science:

• Computer Vision
• Image Processing
• Algorithms

Background:

• Labeling algorithms are crucial for image analysis.
• Efficiency is often limited by memory access operations.
• Minimizing neighborhood operations is key to faster labeling.

Purpose of the Study:

• To propose a fast labeling algorithm.
• To reduce memory access points and neighborhood operations.
• To enhance the efficiency of image labeling.

Main Methods:

• Utilizes a block-based view and raster scan.
• Employs a block-based scan mask to select pixels.
• Integrates block-connected relationships using binary decision trees.
• Reduces unnecessary memory access and simplifies pixel locations.

Main Results:

• The algorithm significantly reduces memory access.
• Binary decision trees require fewer leaf nodes and depth levels.
• Demonstrated faster labeling performance on high-resolution and foreground images.
• Experimental results show superiority over other methods on synthetic and real datasets.

Conclusions:

• The proposed block-based labeling algorithm is highly efficient.
• It offers a significant speed improvement for image labeling tasks.
• This method is effective for processing high-resolution and complex image datasets.