Real-Time Thinning Algorithms for 2D and 3D Images using GPU processors.

Area of Science:

• Computer Vision
• Image Processing
• Scientific Computing

Background:

• Skeletonization is crucial for image analysis and machine learning.
• Current methods can be computationally intensive, limiting real-time applications.

Purpose of the Study:

• To propose novel, efficient GPU-based algorithms for 2D and 3D binary image skeletonization.
• To enhance the speed of image processing for machine learning tasks.

Main Methods:

• Developed algorithms utilizing bit-encoded binary images for simultaneous multi-point processing.
• Employed Boolean algebra and bitwise logical operators for point simplification, avoiding complex encoding/decoding.
• Leveraged GPU parallelization for enhanced computational efficiency.

Main Results:

• The 2D algorithm achieved speeds of 3.53 ns/32x32 pixels and 0.25 ms/1024x1024 pixels, outperforming a multi-threaded algorithm by 52-18,380x.
• The 3D algorithm processed 128x128x128 voxels in 0.27 ms and 512x512x512 voxels in 20.32 ms, 32-46x faster than a border-sequential GPU algorithm.
• Demonstrated significant speedups for both 2D and 3D skeletonization.

Conclusions:

• The proposed GPU techniques enable highly efficient, real-time 2D and 3D binary image skeletonization.
• These advancements can substantially improve the performance of various machine learning applications requiring rapid image analysis.