IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 12, NO. 5, SEPTEMBER/OCTOBER 2006Saliency- g uided Enhancement for Volume VisualizationYoungmin Kim and Amitabh Varshney, Member, IEEEAbstract—Recent research in visual saliency has e stablished a computational measure of perceptual importance. In this paper wepresent a visual-saliency-based operator to enhance selected regions of a volume. We show how we use such an operator on auser-specified saliency field to compute an emphasis field. We further discuss how the emphasis field can be integrated in to thevisualization pipeline through its modifications of regional luminance and chrominance. Finally, we validate our work usin g an eye-tracking-based user study and show that our new saliency enhancement operator is more effective at eliciting viewer attention thanthe traditional Gaussian enhancement operator.Index Terms—Saliency, visual attention, perceptual enhancement, vol u me rendering, non-photoreali stic rendering✦1 INTRODUCTIONVisually depicting large volume datasets in a comprehensible way hasbeen a long-standing challenge. Transfer functions have been widelyused to help visualize the f eatures and details in volumes by assigningvarying optical properties such as color and opacity to different densi-ties of a volumetric scalar field. Significant advances have been madein the art and the science of devising transfer functions that success-fully show the i nherent structures w ithin a given volume dataset. De-spite these impressive advances the t ransfer functions remain a map-ping of the physical appearance to the local geometric attri butes suchas the local density of the scalar field and its first and higher-orderderivatives. Notwithstanding the pioneering work in dual-domain in-teractions by Kniss et al. [12], transfer functions by and large remainill-suited to directly afford the appearance manipulation of selected re-gions of a volume. As the volume datasets have grown in complexity,so too has the need to emphasize and draw visual attention to appro-priate regions in their visualization. This paper addresses the grow ingneed for tools and techniques that can draw visual attention to user-specified regions in a direct volume rendering environment. Towardsthis goal we seek solutions based on multi-scale methods for visualsaliency that can be used to guide visual attention based on varyingperceptual importance.In this paper, we introduce a new visualization enhancement op-erator that is inspired by the center-surround mechanism of visualsaliency. Our goal is to enhance human perception of the volume databy guiding a viewer’s attention t o specific regions of interest. Since ourmethod considers the influence of each voxel at multiple scales, it canemphasize volumetric features at an appropriate visual scale. Existingtransfer functions, based on local geometry and its derivatives, wouldfind it difficult to achieve a similar level of multi-scale emphasis. Oursaliency-guided enhancement framework provides scientists and med-ical researchers a valuable tool to enable them to easily emphasizeand de-emphasize regions of interests even in large volume datasets,successfully guiding user’s visual att ention t o desired regions withoutsacrificing their local context. Saliency-guided emphasis is likely tofind use in large-scale visual knowledge discovery applications whereknowledge discovery modules could identify the regions satisfying acertain criteria and then present them visually with subtle variations todraw a user’s attention to t hose regions in order of their importance.The main contributions of this paper are:• Youngmin Kim is with University of Maryland, College Park, E -mail:[email protected].• Amitabh Varshney is with University of Maryland, College Park, E -mail:[email protected] received 31 March 2006; accepted 1 August 2006; posted online 6Novem ber 2006.For information on obtaining reprints of this article, please send e-mail to:[email protected].(a) (b)Fig. 1. Saliency-guided Enhancement for Volume Visualization: Im-age (a) shows the traditional volume visualization and image (b) showsthe result of applying our saliency-guided enhancement operator to themouth.• We present a new saliency-based enhancement operator to guidevisual attention in volume visualization.• We discuss augmenting the existing visualization pipeline byincorporating enhancement operators to increase the visualsaliency of different regions of a volume dataset.• We present an eye-tracking-based user study that shows that oursaliency-enhancement operator is successful in eliciting viewerattention i n volume visualization.2 RELATED WORKDirect volume rendering models the attenuation of light in a volumecomposed of particles with varying densities and opacities [9, 15].Volume rendering has evolved considerably over the past two decadesand now engineers, scientists, medical researchers, and visual design-ers use a rich suite of tools and techniques to specify the visual ap-pearance of a volume based on their needs. Transfer functions haveplayed a crucial role in broad use of direct volume rendering. Thedesign of transfer functions to generate informative visualizations hasbeen a significant challenge that has been addressed by a number ofresearchers [22]. A number of heuristics are used to guide the users inselecting appropriate transfer functions. For instance, Levoy [15] sug-gested the use of the gradient magnitude to i dentify surfaces in volumedata. Kindlmann and Durkin [10] used the first and second derivativesalong the gradient direction to calculate a boundary emphasis to be in-cluded in the opacity transfer function. In addition to the design of theopacity transfer function, general multi- dimensional transfer functionswere studied to better convey the boundaries and features in volumedata [11, 12 , 13, 18].IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 12, NO. 5, SEPTEMBER/OCTOBER 2006Fig. 2. (a) The traditional visualization pipeline. (b) Saliency-enhanced visualization pip e line. The saliency field is modified by the enhancementoperator to generate an emphasis field. The emphasis field is used to enhance the perception of features in volume by modulating appearanceattributes such as luminance, chrominance, and texture detail.Stylized rendering in volume visualization has attracted extensiveresearch interest in the last few years. Treavett and Chen [28] devel-oped techniques for