Deep Volumetric Ambient Occlusion.

Area of Science:

• Computer Graphics
• Medical Imaging
• Artificial Intelligence

Background:

• Direct volume rendering requires efficient computation of volumetric ambient occlusion for realistic visualizations.
• Existing methods often struggle with real-time performance and interactivity, especially with complex volumetric datasets.

Purpose of the Study:

• To introduce a novel deep learning technique, Deep Volumetric Ambient Occlusion (DVAO), for predicting per-voxel ambient occlusion.
• To enable real-time, interactive direct volume rendering by leveraging a neural network that considers global transfer function information.

Main Methods:

• A deep neural network was designed to predict volumetric ambient occlusion.
• The network integrates global information from the transfer function for accurate occlusion prediction.
• Various transfer function representations and neural network injection strategies were analyzed.

Main Results:

• The proposed DVAO approach successfully predicts volumetric ambient occlusion.
• The method supports real-time volume interaction, as the neural network only needs re-execution upon changes in global information.
• DVAO demonstrates generalization capabilities across different imaging modalities, including computed tomography data.

Conclusions:

• DVAO offers an efficient and effective solution for real-time volumetric ambient occlusion in direct volume rendering.
• The study provides insights and recommendations for applying deep learning in similar volume rendering scenarios.
• The technique's ability to generalize suggests broad applicability in scientific visualization and medical imaging.