3D-SLARM: Practical Lossless Volumetric Image Compression via a 3D-Scanning Lightweight Autoregressive Model.

Area of Science:

• Medical Imaging
• Computer Vision
• Data Compression

Background:

• Lossless compression is crucial for preserving data integrity in volumetric images.
• Existing learned methods face challenges in balancing compression ratio, coding speed, and model complexity.

Purpose of the Study:

• To propose a novel, lightweight, and efficient model for lossless volumetric image compression.
• To address the limitations of current methods in achieving high compression ratios with fast coding speeds.

Main Methods:

• Introduced a 3D-scanning lightweight autoregressive model (3D-SLARM).
• Integrated a 3D plane scanning module for parallel voxel coding.
• Employed a lightweight feature extraction module with serial re-parameterization (SerRep) and non-centric masked convolution (NCMC).
• Utilized a lightweight distribution parameter and adaptive range predictor (DPARP) module for efficient parameter prediction and adaptive probability range generation.

Main Results:

• 3D-SLARM achieves state-of-the-art lossless compression performance on most volumetric image datasets.
• The model demonstrates fast coding speeds.
• The proposed architecture is lightweight and practical for real-world applications.
• Effective handling of both 8-bit and high bit-depth volumetric images.

Conclusions:

• 3D-SLARM offers a practical solution for lossless volumetric image compression.
• The model successfully balances high compression ratios, rapid coding speeds, and a lightweight design.
• The novel components, including 3D plane scanning and SerRep/NCMC, contribute to its effectiveness.