Fast Compressed Segmentation Volumes for Scientific Visualization.

Area of Science:

• Medical Imaging
• Computer Graphics
• Data Compression

Background:

• Voxel-based segmentation volumes generate large datasets, hindering storage, transfer, and interactive visualization.
• Efficiently managing these large datasets is crucial for scientific research and applications.

Purpose of the Study:

• To introduce a novel lossless compression technique for voxel-based segmentation volumes.
• To address the challenges of data size in storage, transfer, and interactive visualization.

Main Methods:

• The technique processes segmentation volumes in small bricks, encoding labels and boundaries using an iterative refinement scheme.
• Bricks are further compressed using rANS-entropy coding with global frequency tables for efficient parallel decompression.
• The method is designed for applicability with GPU-based rendering.

Main Results:

• Achieves high compression ratios, reducing data size to 1%–3% of the original.
• Demonstrates high throughput for both compression and decompression (up to gigabytes per second).
• Enables GPU-based volume visualization with on-the-fly decompression and level-of-detail rendering.

Conclusions:

• The proposed lossless compression technique significantly reduces data size for voxel-based segmentation volumes.
• It enhances performance for storage, transfer, and interactive visualization, particularly in GPU-accelerated environments.
• The method supports advanced visualization features like LOD rendering and caching for improved efficiency.