Quality Assessment of Volume Compression Approaches Using Isovalue Clustering

We provide an interactive tool for extracting exemplar isosurfaces from a 3D scalareld using a novel isovalue classication process.We propose a structural VQA metric that uses representative isosurfaces as benchmark structures to assess the visual quality of compressed 3D scalarelds.Our experiments on a number of benchmark data sets suggest that, compared to existing methods, the proposed isovalue classication approach provides a more distinct set of isosurfaces that are more representative of the complexity of the datasets.We examine a number of widely-used compression techniques (i.e.,discrete wavelet transform, discrete cosine transform, and tensor approximation) to establish the utility of our VQA approach. Display Omitted Visualization of volumetric data has been widely used for exploration of data from scientific simulations and biomedical imaging. Despite advances of GPU-assisted rendering, which has become the state-of-art in direct volume rendering, still many volumetric data sets are too large to be visualized interactively. Therefore, compression-domain rendering approaches are used in visualization processes in order to reduce the amount of data sent to the GPU and thus speed up interactive visualization. Hence, reliable tools to assess the quality of the reconstructed 3D data are of great importance, influencing the effectiveness of the visualization. However, numerical error analysis approaches such as mean-squared-based metrics are often inconsistent with perceived visual quality. We propose a structural volume quality assessment approach for 3D scalar volume based on the human visual system (HVS). Our approach consists of two stages: First, we provide an interactive tool for extracting significant volume features via isosurfaces from a 3D scalar field using an isovalue classification process. Second, we propose a structural volume quality assessment (VQA) metric that employs representative isosurfaces as benchmark structures. For this purpose, we use a recently developed perceptual-based mesh quality metric [1] to assess the visual quality of compressed 3D scalar fields. Our experiments on a number of benchmark data sets suggest that, compared to existing methods, the proposed isovalue classification approach provides a more distinct set of isosurfaces that are more representative of the complexity of the data sets. We examine a number of widely used compression approaches, namely, discrete wavelet transform, discrete cosine transform, and tensor approximation, to establish the utility of our volume quality assessment approach.