An image-based approach to interactive crease extraction and rendering

Ridge and valley manifolds are receiving a growing attention in visualization research due to their ability to reveal the shapes of salient structures in numerical datasets across scientific, engineering, and medical applications. However, the methods proposed to date for their extraction in the visualization and image analysis literature are computationally expensive and typically applied in an offline setting. This setup does not properly support a userdriven exploration, which often requires control over various parameters tuned to filter false positives and spurious artifacts and highlight the most significant structures. This paper presents a GPU-based adaptive technique for crease extraction and visualization across scales. Our method combines a scale-space analysis of the data in pre-processing with a ray casting approach supporting a robust and efficient one-dimensional numerical search, and an image-based rendering strategy. This general framework achieves high-quality crease surface representations at interactive frame rates. Results are proposed for analytical, medical, and computational datasets.