Experimental Evaluation of Semantic Depth of Field, a Preattentive Method for Focus+Context Visualization Verena Giller , Manfred Tscheligi , Johann Schrammel , Peter Fr¨ ohlich , Birgit Rabl , Robert Kosara , Silvia Miksch , and Helwig Hauser CURE: Center for Usability Engineering, Austria, http://www.cure.at/ Vienna University of Technology, Austria, http://www.asgaard.tuwien.ac.at/, rkosara, silvia @ asgaard.tuwien.ac.at VRVis Research Center, Austria, http://www.VRVis.at/vis/, mailto:Hauser@VRVis.at Abstract We introduce the Semantic Depth of Field (SDOF) tech- nique, which is an alternative focus+context method for in- formation visualisation. SDOF blurs objects which are cur- rently out of focus, i.e., not interesting for the user. In the experimental study described in this paper, we found that SDOF supports the preattentive perception of sharp targets and the numerical estimation of the amount of data being in focus. We investigated the influence of distracting encod- ings (color and orientation), as well as threshold values for human perception. We also tested applications of SDOF (a textviewer and a scatterplot) with regard to their effective- ness in guiding the user’s attention and with highlighting of data items. Keywords: visualization, information visualization, focus+ context visualization; blur, depth of field; preattentivity; ex- perimental user study 1 Introduction Information visualization (InfoViz) is the use of computer- supported, interactive, and visual representations of abstract data to facilitate cognition. The goal of InfoViz is to ease un- derstanding, to promote a deeper comprehension of the data under investigation, and to foster new insights into the under- lying processes. Because most data lack an inherent structure that could be directly understood as spatial, it is important to find a mapping of data dimensions to space, color, etc., so that the user can understand the data. It is also important not to change this mapping too often, because that requires the user to learn a new mapping (because his or her mental map [9] is destroyed). When a lot of data is shown, it is necessary to be able to zoom in on certain parts to get more detail. At the same time however, the user must be able to retain an idea of where in the data the zoomed data is, so as to understand it in relation to the rest of the data. This is called a focus+context (F+C) approach, and has been the subject of a considerable amount of work in visualization (see the section 2 for an overview). Semantic Depth of Field (SDOF) is an F+C technique that is based on an effect known from photography and cine- matography, called depth of field (DOF). A lens or lens sys- tem does not depict all objects equally sharp, but only those points which are in a plane parallel to the film plane at a cer- tain distance from the lens – all other objects are blurred [7]. We use the fact that the human eye is used to ignoring blurred objects (because the eye also has a limited depth of field) by blurring currently irrelevant objects. This way, the user’s attention is guided to the most relevant objects without losing the (blurred and thus less prominent) context. In this paper, we present the results of a user study that examined the following hypotheses: exploring if SDOF sup- ported preattentive perception of sharp target items, the nu- merical estimation of presented data (values, amounts, etc.), work in combination with another coding as an orthogonal (additional) dimension, the guidance of the user’s attention and the highlighting of data in a concrete application context. The study consisted of two parts in which the test partici- pants were confronted with visual non-meaningful images (experimental study) and with meaningful application con- text tasks (evaluation of applications). In the following section, we discuss some existing F+C methods and uses of blur in visualization. In the subsequent section, we present the idea of SDOF, describe the design of our study, discuss its results, and finally provide some con- clusions and future plans. 2 Related Work Most focus+context methods are distortion-oriented [3] (or, as we call them, spatial) ones, e.g. fisheye views or transfor- mations into hyperbolic space [3]. These methods magnify important areas, while at the same time shrinking less rele- vant ones. This way, it is possible to put more objects into the same amount of screen space without losing the ability to recognize details for the more important ones. The methods we call dimensional methods do not change the layout or space allocation of a visualization, but show dif- ferent data where needed. The user moves a focus window