Visibility-Driven Processing of Streaming Volume Data

Abstract

In real-time volume data acquisition, such as 4D ultrasound, the raw data is challenging to visualize directly without additional processing. Noise removal and feature detection are common operations, but many methods are too costly to compute over the whole volume when dealing with live streamed data. In this paper, we propose a visibility-driven processing scheme for handling costly on-the-fly processing of volumetric data in real-time. In contrast to the traditional visualization pipeline, our scheme utilizes a fast computation of the potentially visible subset of voxels which significantly reduces the amount of data required to process. As filtering operations modify the data values which may affect their visibility, our method for visibility-mask generation ensures that the set of elements deemed visible does not change after processing. Our approach also exploits the visibility information for the storage of intermediate values when multiple operations are performed in sequence, and can therefore significantly reduce the memory overhead of longer filter pipelines. We provide a thorough technical evaluation of the approach and demonstrate it on several typical scenarios where on-the-fly processing is required.