Multiresolution Volume Filtering in the Tensor Compressed Domain.

Area of Science:

• Computer Graphics
• Scientific Visualization
• Image Processing

Background:

• Signal processing and filter operations are crucial for visual data analysis.
• Interactive filtering of large-scale volume data is computationally intensive due to GPU memory and bandwidth constraints.

Purpose of the Study:

• To develop a novel and efficient approach for multiscale volume filtering within an interactive visualization framework.
• To enable interactive filtering of large volume datasets that exceed GPU memory limitations.

Main Methods:

• A hierarchical multiresolution tensor approximation model is used to decompose raw volume data offline.
• Convolution filter operators are applied directly in the compressed tensor approximation domain.
• A technique is employed to prevent aliasing during multiresolution filtering, ensuring accurate full-resolution filtering at low cost.

Main Results:

• The proposed method allows accurate filtering at full spatial volume resolution within the compressed domain.
• Filtered results are reconstructed and displayed at a variable level-of-detail.
• The system demonstrates scalability for handling large volume datasets interactively.

Conclusions:

• The developed compression-domain volume filtering approach significantly enhances interactive visualization of large datasets.
• This method overcomes GPU memory limitations, offering a scalable and efficient solution for complex filter operations.