Joint EUROGRAPHICS - IEEE TCVG Symposium on Visualization (2003) G.-P. Bonneau, S. Hahmann, C. D. Hansen (Editors) Interative Feature Speifiation for Fous+Context Visualization of Complex Simulation Data Helmut Doleish, Martin Gasser, Helwig Hauser VRVis Researh Center, Vienna, Austria http://www.vrvis.at/vis/ Doleish@VRVis.at, Martin.Gasser@VRVis.at, Hauser@VRVis.at Abstrat Visualization of high-dimensional, large data sets, resulting from omputational simulation, is one of the most hallenging fields in sientifi viualization. When visualization aims at supporting the analysis of suh data sets, feature-based approhes are very useful to redue the amount of data whih is shown at eah instane of time and guide the user to the most interesting areas of the data. When using feature-based visualization, one of the most diffiult questions is how to extrat or speify the features. This is mostly done (semi-)automati up to now. Espeially when interative analysis of the data is the main goal of the visualization, tools supporting interative speifiation of features are needed. In this paper we present a framework for flexible and interative speifiation of high-dimensional and/or omplex features in simulation data. The framework makes use of multiple, linked views from information as well as sientifi visualization and is based on a simple and ompat feature definition language (FDL). It allows the definition of one or several features, whih an be omplex and/or hierarhially desribed by brushing multiple dimensions (using non-binary and omposite brushes). The result of the speifiation is linked to all views, thereby a fous+ontext style of visualization in 3D is realized. To demonstrate the usage of the speifiation, as well as the linked tools, appliations from flow simulation in the automotive industry are presented. 1. Introdution Visualizing high-dimensional data resulting from omputational simulation is a demanding proedure, posing several omplex problems whih inlude, for example very large size of data sets and inreased dimensionality of the results. In this paper, we present a formal framework that supports interative and flexible analysis of omplex data using a desriptive and intuitive language for defining features and multiple linked views with information visualization (InfoViz) and sientifi visualization (SiViz). In the following, we shortly disuss a few key aspets, whih are important for the new approah presented in this paper. Feature-based visualization visualization whih fouses on essential parts of the data instead of showing all the data in the same detail at the same time, is alled feature-based visualization. This kind of visualization gains inreasing importane due to bigger and bigger data sets whih result from omputational simulation, so that not all of the data an be shown simultaneously. For feature-based visualization, proper feature extration methods are essential. Up to now, feature extration mostly is done (semi-)automatially 14 with little interative user intervention, often as a preproessing step to the visualization. But for interative analysis, in many ases, the question of what atually is (or is not) onsidered to be a feature refers bak to the user: depending on what parts of the data the user (at an instane of time) is most interested in, features are speified aordingly. Therefore, flexible feature extration requires effiient means of user interation to atually speify the features. Separating fous and ontext in InfoViz when dealing with large and high-dimensional data sets in InfoViz, simultaneous display of all the data items usually is impossible. Therefore, fous-plus-ontext (F+C) tehniques are often employed to show some of the data in detail, and (at the same time) the rest of the data, at a lower resolution, as a ontext for orientation 3 6. Thereby the user s attention is di- 239
Doleish, Gasser, Hauser / Interative Feature Speifiation for F+C Visualization of Complex Simulation Data Figure 1: Flexible Feature Speifiation: simulation data of a atalyti onverter is shown, two features have been speified based on our feature definition language, using the different views for interation and visualization. (see also olorplate) reted towards the data in fous (e.g., through visual enlargement), whereas the rest of the data is provided as ontext in redued style (transluently, for example). This is espeially useful when interating with the data, or when navigating through the visualization. To disriminate data in fous from ontext information, a so-alled degree of interest (DOI) funtion an be used 6, assigning a 1D DOI-value out of the unit interval to eah of the n-dimensional data items (1 represents in fous, 0 is used for ontext information). Defining the DOI funtion in literature, impliit tehniques for DOI-speifiation are desribed (e.g., fous speifiation through dynami querying 15 ) as well as expliit tehniques, suh as interative objet seletion 9 or brushing 1 17. When brushing, the user atively marks a subset of the data set in a view as being of speial interest, i.e., in fous, using a brush-like interfae element. In addition to standard brushing, several useful extensions to brushing have been proposed. Multiple brushes and omposite brushes 12, and the use of non-binary DOI funtions for smooth brushing 4 extend the available toolset for interative DOI speifiation. Also, more omplex brushes like those designed for hierarhial data 5, or suh using wavelets 18 or relative information between different data hannels 8 have been proposed reently. 240 Complex and high-dimensional feature definition when analyzing simulation data, one very often enountered problem is the limited flexibility of urrent brushing and interation tehniques. Brushing is usually restrited to simple ombinations of individual brushes, as well as missing support of high-dimensional brushes due to the tight oupling of GUI interations and the representation of the brush data itself. For fast and flexible analysis of the usually large and high-dimensional simulation data, omplex and also highdimensional brushes are neessary. In this paper, we present a formal framework, that is very losely oupled to the data, allowing to define and handle suh brushes interatively. Linking multiple views the ombination of InfoViz and SiViz methods 7 4, espeially for the interative visualization and analysis of simulation data, improves the understanding of the data in terms of their high-dimensional harater as well as the data relation to the spatial layout. Linking several views 2 to interatively update all hanges of the data analysis proess in all views simultaneously is a ruial property for making optimal use of multiple (different) visualization views. In previous work 4 we showed how a satterplot (or a histogram) an be used to smoothly speify features in multi-dimensional data from simulation, and how this fous+ontext disrimination an be used for seletive visualization in 3D. In this paper, we now present a formal framework (our feature definition language, see setion 2)
Doleish, Gasser, Hauser / Interative Feature Speifiation for F+C Visualization of Complex Simulation Data for speifying features in simulation data together with advaned interation tehniques (see setion 3), allowing for fast and flexible exploration and analysis of omplex and high-dimensional data (appliation examples in setion 4). Finally, a short overview about implementation details is given, as well as onlusions and some future work topis are presented. 2. Using a Feature Definition Language When dealing with results from omputational simulation, usually very large and high-dimensional data sets are investigated. Previous work already showed, that interative speifiation of features with tight referene to the atual data attributes is very valuable for visualization of suh data sets 7 4. For a fast and flexible analysis of these results, powerful and intuitive tools are needed the here desribed approah provides flexibility in terms (a) of multiple options to differently view the data, and (b) a wide range of user interations to onstrut and adapt feature speifiations. Whereas previous work mainly foussed on viewing (a) so far, we mostly improve on interation (b) in this paper. To generalize the speifiation of features (enabling feature desriptions whih are portable between data sets, for example) and to also formally represent the state of an analysis session, e.g., to allow for loading/saving of interative visualization sessions, we present a ompat language for feature speifiation, i.e., a feature definition language, here alled FDL for short. Figure 2: Feature definition language: sketh of its struture. A sketh of the FDL-struture is presented in Fig. 2. Here the different key omponents of this language are shown, namely the feature speifiation itself (root), feature sets (level 1), features (level 2) and feature harateristis (level 3). In the following subsetions, these four different hierarhial layers of the FDL are disussed in more detail. 2.1. Feature Speifiation A desription of a feature speifiation usually is losely oupled to a data set (the one that is to be analyzed). Alternatively, it ould also be portable to similar data sets, when data semantis oinide. In the regular ase, a feature speifiation therefore has a referene to the soure data set, as well as to one or multiple feature sets (see below). Our FDL is realized as an XML 13 language appliation, whih makes it easy to handle and the resulting FDL-files readable. This also allows to save feature speifiations as files, and load them again at any later point in time to resume an analysis session. Additionally, XML-files an be edited using a text-editor, whih allows to re-adjust feature speifiations also on a file level. The expliit representation of feature speifiations in the form of FDL-files makes using feature speifiations on other data sets possible. Of ourse, are has to be taken that only data hannels are referred to, whih are available in all these data sets. With portable feature speifiations it is possible to generate general feature definition masks, whih an be applied very easily (and interatively adapted, if neessary). 2.2. Feature Sets A feature set subsumes an arbitrary number of features whih all are to be shown simultaneously (like an impliit logial OR-ombination). Within eah single view, always only one feature-set is used for F+C disrimination, all the other feature-sets are inative at that time. Multiple feature sets an be used to interatively swith foi during an analysis session or to intermediately ollet features in a "repository" feature set, not used at a ertain point in time. Multiple views an be used for simultaneously showing different feature sets (one per view). 2.3. Features Features are speified by one or multiple feature harateristis. The DOI funtion related to eah feature is built up by an (impliit) AND-ombination of all DOI funtions of all assoiated feature harateristis. Multiple features are used to support named feature identifiation and intuitive handling of interesting parts of the data by the user. Eah feature an be moved or opied from one feature set to any other. In Fig. 1 two distint features have been speified, one denoting areas of bakflow, and another one, showing vorties. The latter one onsists only of one simple feature harateristi (see below), brushing high values of turbulent kineti energy, whereas the first feature onsists of a logial ombination of two separate feature harateristis. 2.4. Feature Charateristis Feature hararateristis an be either simple or omplex. Whereas simple feature harateristis store diret brushing information with respet to one data attribute (or hannel) to derive a DOI funtion, omplex feature harateristis imply a reursion. Simple feature harateristis store a referene to the data hannel whih it is based on, as well as infomation about 241
Doleish, Gasser, Hauser / Interative Feature Speifiation for F+C Visualization of Complex Simulation Data explained in more detail in setion 4), pressure (x-axis) vs. veloity (y-axis) values are plotted. Interative operations "NOT-AND" and "SUB" are mapped to "NOT"-"AND" and "AND"-"NOT" ombinations in FDL, respetively. Figure 3: four examples of 2D brush types whih users found useful during interative analysis (atalyti onverter example, pressure [x] vs. veloity [y]): (a) high veloity and high pressure (logial AND), (b) low veloity or low pressure (log. OR), () all but high vel. and high pressure (NOT-AND), and (d) high pressure but not low veloity (SUB = AND-NOT). (see olorplate for shades of red) how the data of this hannel is mapped to a DOI funtion (being the output of this harateristi). Espeially the possibility for the user to diretly interat with the data attributes by speifying feature harateristis and modifying them interatively is very intuitive and straight-forward. In Fig. 1 a simple feature harateristi named "negative veloity in X- diretion" is shown in the seletion bounds editor. Simple feature harateristis support disrete and smooth brushing (via speifying perentages of the total brushing range, where the DOI-values derease gradually). Complex feature desriptions on the other hand provide logial operations (AND, OR, NOT) for the user to ombine subsequent feature harateristis in an arbitrary, hierarhial layout. For ombining smooth brushes, whih an be interpreted as fuzzy sets, fuzzy logial ombinations are used, usually implemented in form of T-norms and T- onorms 11. We integrated several different norms for the above mentioned operations. By default, we use the minimum norm (T M ) in our implementation: this means, when doing an AND-operation of several values, the minimum value is taken, and for the OR-operation the maximum respetively. In Fig. 3, four examples of 2D brush types, whih users found useful during interative analysis sessions, are shown. The data displayed in the satterplot views omes from the atalyti onverter appliation shown in Fig. 1 (whih is also 242 3. Interation One main aspet of analyzing results from simulation is that investigation is often done interatively, driven by the expert working with the visualization system. Therefore, interation is one of the key aspets that has to be onsidered when designing a system whih should support fast and flexible usage (as desribed previously). Espeially the task of searhing for unknown, interesting features in a data set, and extrating them, implies a very flexible and intuitive interfae, allowing new interation methods. In the following subsetions, we ategorize the main types of interation whih our system supports. Note that these interations are designed to meet users most often requested requirements for suh an analysis tool. 3.1. Interative Feature Speifiation through Brushing The first type of interation that has to be onsidered when designing an interative analysis tool for exploring simulation data, is brushing. In our system, interative brushing of data visualization is possible in all views exept for the 3D SiViz view, whih is used for 3D F+C visualization of the feature speifiation results (see setion 4). Brushing is used to define feature harateristis in the FDL interatively. As many types of appliations also request non-binary brushing, we allow smooth-brushing 4 in all the interation views. One example of using a 2D smooth brush, employing a logial AND operation of two simple feature harateristis is shown in Fig. 3 (a). Here, a region of relatively high veloity and high pressure values is brushed in a satterplot view, defining (a part of) a feature. As an be seen from this figure, a smooth brush defines two regions. A ore part of the brush is defined, where data of maximal interest is seleted (mapped to DOI values of 1). It is padded by a border, where DOI values derease gradually with inreasing distane from the ore part. 3.2. Interative Feature Loalization Another very often used type of interation is the so-alled feature loalization. It is usually provided in the ontext of simulation data, that has some spatial ontext. When analyzing this kind of data, the first interest is often, where features of speifi harateristis are loated in the spatial ontext of the data. Interatively defining and modifying features in different views, oupled with linking, the speifiation immediately results in a 3D rendering whih provides fast loalization of the features in the spatial ontext of the whole data set. For an example see Fig. 4 (a)-(), where the bakflow regions are interatively loalized to be in the entrane of the atalyti onverter hamber.
Doleish, Gasser, Hauser / Interative Feature Speifiation for F+C Visualization of Complex Simulation Data Figure 4: Interative feature speifiation and refinement: (a)-(): first step: defining bakflow region in a atalyti onverter (see also Fig. 1) in a satterplot view (a) by seleting negative x-flow values, diret linking to a seond satterplot view (b) and the 3D view (). (d)-(f): seond step: AND-refinement with a new seletion in the seond satterplot view (e), bak linking of the interation via feedbak visualization (olor of points aording to newly alulated DOI values) to the first satterplot view (d). Now only the bakflow region is seleted, that exhibits general veloity above a speified threshold (f). 3.3. Interative FDL Refinements After having defined multiple features via brushing and loalized them, often interative refinement of these features is the next step. Refining the feature speifiation an be either done by interative data probing (see below) or by imposing further restritions on the feature speifiations, e.g., by adding additional feature harateristis to the atual state of a feature. One example of suh an interative FDL refinement is shown in Fig. 4 (d)-(f). As a first step (first row), all parts of the data, that exhibit bakflow, have been seleted, defining a feature that spans over two distint regions in the spatial domain. In the refinement step (seond row) a logial AND-ombination of the first feature speifiation (a) with a new seletion in a seond satterplot view of the same data (but showing two other data attributes) is performed (e). Thereby only those bak-flow regions of the data are put into fous, whih exhibit a general veloity above a speified threshold (f). 3.4. Interation with Tree Viewer Interation with a tree viewer (see Fig. 1, left upper window) as a GUI for FDL is another very useful way to adapt or extend feature sets and features, as well as their harateristis. The tree viewer provides standard GUI elements, suh as textfields for manual input of numbers or range sliders, for example. Naming of the different nodes of the FDL, as well as editing all the feature harateristis, and also the management of the tree struture (through opy, delete, or move of the different nodes and subtrees) are the most often used interation methods in this viewer. It strongly depends on the nature of users of whether mouse-interations or keyboard-input are preferred when speifying features. Sometimes, in the ase of well-known thresholds, for example, the keyboard-input to the tree viewer is faster and more aurate then mouse-interation to an InfoViz view. 3.5. Interative Data Probing Another form of interatively exploring features is using a data-probing approah. Thereby, after having speified a feature (via brushing, for example) the one or other feature harateristi an be hanged interatively (e.g., by using a range slider). In all linked views (showing the same data and showing different data attributes) immediate feedbak of DOI hanges an give new insights into different data as- 243
Doleish, Gasser, Hauser / Interative Feature Speifiation for F+C Visualization of Complex Simulation Data pets. Espeially for exploratively investigating value ranges and better understanding of assoiated patterns in the data sets, this interation metaphor is very useful. 3.6. Interative Management of Views One key aspet of a system whih provides multiple, different views of one data set, is the interative management and linking of these views. Our system supports an arbitrary number of InfoViz views (urrently satterplots and histograms), as well as SiViz views. Views an be opened and losed at any point in time without distrating the feature speifiation. In the InfoViz views, the mapping whih assigns data hannels to the axes an be hanged interatively. In the 3D SiViz view the mapping of a data attribute to rendering properties (olor and/or opaity) via transfer funtions an be interatively modified, too. Additionally, the different axes of all available views an be linked (and unlinked) interatively, allowing rapid updates in multiple views. 4. Visualization and Results from Appliations After having disussed our feature speifiation framework as well as the important role of interation for analysis of simulation data, now the visualization part and typial appliations are presented. Below general aspets of visualization during analysis are presented. Then, two different appliation examples are desribed in detail. For high quality versions of the images presented here, as well as for additional examples and movies whih illustrate the interative behaviour of working sessions with our framework, please refer to http://www.vrvis.at/vis/researh/fdl-vis/. 4.1. Visualization for Analysis When visualization is used to support analysis of large, highdimensional data sets, the use of multiple views, as well as of flexible views (with respet to data dimensionality) is very important. Our system supports an arbitrary number of eah type of InfoViz views, as well as SiViz views. When interatively working with data, two types of views in a multiple views setup an be distinguished: Atively linked views are the views, whih are primarily used for interation purposes, i.e., for speifying the features, whereas passively linked views are primarily used for F+C visualization of the data, providing interative updates. 3D SiViz views The 3D SiViz views of our system are used as passively linked views for providing a F+C visualization and interative feature loalization. The F+C disrimination is mainly aomplished by using different transfer funtions for fous and ontext parts (and interpolating inbetween, for smooth F+C disrimination). The transfer funtions in use do not only speify olor and opaity, but 244 also the size of the glyphs, that are used to represent single data items (see Fig. 1, lower left window for a 3D SiViz view, showing a smooth F+C visualization). Two main tasks of this F+C visualization an be identified. The support for feature loalization and the visualization of data values through olor mapping. Feature loalization, as already desribed in setion 3, plays a major role in interative analysis based on features. By using a F+C visualization, the user attention is automatially drawn to the more prominently represented foi, i.e., the features. Value visualization is another very useful task of visualization in this view, and it is aomplished by oloring glyphs aording to the assoiated data hannel. Of ourse, interative user manipulation of rendering parameters (opaity, size of glyphs, or zoom and rotate) are neessary, very useful, and support the analysis task, too. InfoViz views Apart from supporting interation, the InfoViz views (satterplots & histograms in our system) are very valuable for visualization purposes, too. They visualize the data distribution (1D or 2D) and also give visual feedbak of F+C disrimination. Points in the satterplot views, for example, are olored aording to the DOI value of the assoiated data item. Fully saturated red points are shown for data in fous, whereas the saturation and lightness of points dereases with dereasing DOI values, respetively (see Fig. 3 for examples). In the InfoViz views it is espeially useful that the mapped data attributes an be hanged interatively. Mapping spatial axis information to one of the satterplot axes, for example, is very intuitive in our appliations (see below). Additionally, using several satterplots, omparable to a (redued) satterplot matrix, often adds information about the data and internal relations of different data attributes. 4.2. Results from Air-Flow Analysis We now want to give a step-by-step demonstration of how a typial analysis session takes plae, espeially to show the importane of interation when analyzing simulation data. (1) In a first step, a data set is loaded: in our example, results from air-flow simulation around a ar (just on one entral slie, from front to bak of the ar) are shown. To also ope with 2D-slies of 3D-data, we adapted our 3Drendering view aordingly. It should be noted, that the general flow diretion in this appliation is in X-diretion, past the ar from front to bak. Before a tree viewer is opened automatially, an empty feature set is generated for preparation of an analysis session. A SiViz view is then opened interatively, to show the general spatial layout of the data (see Fig. 5 for the initial view setup). In this figure the unstrutured grid of the data set is shown, overall veloity information is mapped to olor (green denotes low, red relatively high veloity values).
Doleish, Gasser, Hauser / Interative Feature Speifiation for F+C Visualization of Complex Simulation Data Figure 5: Air-Flow around a moving ar: After loading the data set, an empty feature set is reated, and the spatial layout of the data is shown, overall veloity information is mapped to olor (green denotes low, red high veloity). (2) As a first start into feature speifiation (foussing on non-horizontal, slow flow at this step of the analysis) a satterplot view is opened, showing V-veloity (vertial omponent of overall veloity values), mapped to both axes. In this satter plot an OR-brush is used to selet relatively large positive V-flow, as well as relatively large negative one, too. Then the x-axis of the satterplot view is hanged to show overall veloity and an AND-refinement is done to limit the feature speifiation to slow flow (see Fig. 6, upper right view). To furthermore visualize the feature speifiation up to this step, a seond satterplot view is opened, showing feature and ontext distribution with respet to the spatial X- oordinates and visosity (mapped to y-axis of the view, see Fig. 6, lower right). In an interation panel of the tree viewer, the restrition of V-veloity omponents is further adapted, to meet the user s needs (see Fig. 6 for a sreen apture after this step). (3) A further AND-refinement, restriting the feature speifiation to "high visosity" values is added by using the seond satterplot view. As a result of this step, only features behind the ar are part of the new fous (see Fig. 7). (4) Yet another AND-refinement, further restriting the feature speifiation to high values of turbulent kineti energy (a value also omputed by the simulation), is performed in the tree viewer (see Fig. 8). This lips away parts of the previously seleted features, leaving only the parts that exhibit stronger rotational behavior. (5) To get a better idea of the vortial strutures indued, interative probing on one part of the feature speifiation (positive V-veloity) is performed. When limiting the fous to negative V-flow only, the downfaing parts of the upper as well as of the ounterrotating, lower vortex beome visible (see Fig. 9). 4.3. Results from Catalyti Converter Analysis A seond example presented here is an appliation, where the data omes from a simulation of a atalyti onverter from automotive industry. The results of another analysis session are shown. The data is given on an unstrutered grid in 3 spatial dimensions, and has 15 different data attributes for eah of the approximately 12000 ells of the grid. The data set and a orresponding feature speifiation is shown in several views in Fig. 1. The data set onsists of basially three spatially distint parts, the flow inlet on the left hand side, the hamber of the atalyti onverter (middle), and the flow outlet on the right-hand side (see Fig. 1, left lower window for a 3D SiViz view). The other views shown in Fig. 1 inlude: the tree view for handling the FDL (inluding a pop-up window for hanging the brush properties on the x-omponent of the veloity), a satterplot view (right upper window) plotting x-veloity vs. x-oordinates for eah data point, and a histogram, showing the distribution of x-veloity values over the data range. Two distint features have been speified using the InfoViz views and the FDL tree viewer. The first feature defines all bakflow regions in the data set (with negative x- omponent of the veloity, as general flow is in x-diretion). Two suh regions are identified at the entrane of the hamber, a weaker one at the bottom of the atalyti onverter, 245
Doleish, Gasser, Hauser / Interative Feature Speifiation for F+C Visualization of Complex Simulation Data Figure 6: First step of analysis (non-horizontal slow flow): a tree viewer showing the urrent feature speifiation in the upper left (interation panel for adjusting a simple feature harateristi shown), a satterplot view used for feature speifiation in the upper right (veloity vs. V-Veloity omponent), the SiViz view for f+ visualization in the lower left, a seond satter plot for visualization of f+ distribution (X-oordinates vs. visosity). and a stronger one at the top. The seond feature desription defines all regions, with high turbulent kineti energy, these are the regions, where vorties are appearing usually in the flow. As an be seen, two vortex ores are easily separated from the rest of the data at the inlet and outlet of the atalyti onverter in this ase. Both, the vorties and the bakflow regions have been brushed smoothly, to show some information about the gradient of the values in the 3D rendering view. Note, that the oloring in the 3D view is mapped from another data hannel, namely data values of absolute pressure. This allows to visualize an additional data dimension for all the data, that was assigned to be in fous beforehand. In the here applied olor mapping, green denotes relative low values of absolute pressure, and red orresponds to relative high values. 5. Implementation The presented prototype system inludes the desribed simulation data analysis tools and runs interatively on a standard PC (P3, 733MHz, 756MB of memory, GeFore2) for the data sets shown (in the range of 20.000 to 60.000 ells, 246 Figure 7: Step 2 of analysis: AND-refinement, restriting feat. spe. to high visosity values in the seond satterplot view. Only features behind the ar are part of fous now. 15 to 50 data attributes assoiated to eah ell). The ells of the data are organized in unstrutured grids. For the rendering of these grids a visibility algorithm was implemented,
Doleish, Gasser, Hauser / Interative Feature Speifiation for F+C Visualization of Complex Simulation Data Figure 8: Step 3 of analysis: another AND refinement, further restriting to high values of turb. kineti energy, performed in the tree viewer. Only parts with strong rotational omponent are in fous. (see also olorplate) based on the XMPVO algorithm 16 presented by Silva et al. With newer, more powerful PC-setups we already managed to visualize data sets onsisting of over a million data ells, but sorting for 3D rendering an not be performed interatively anymore. For the implementation of our prototype, a hybrid approah was taken; UI Interation and handling of the FDL is realized in Java, whereas mesh aess and the rendering of the visualization views is implemented in native ode (we used MS Visual C++). Native methods are alled via the JNI API, and the gl4java pakage was used to make the GL rendering ontexts available to the Java GUI toolkit. The mesh aess has been realized by using our own data mesh format. Data oming from different data soures an be easily onverted to this format via linked readers. When designing the presented FDL, several onsiderations were taken, inluding for example: ease of implementation (lose to the visualization sytstem and the data), allow for manual input by the user (preferably ASCII-based, with semantis), verifiation should be possible (to hek for invalid definitions), and many more. To meet all these design onsiderations, is was deided to use the XML language 13 for storage of the FDL and as interfae to other appliations. For writing and reading feature speifiations to and from FDL-files, the Apahe Crimson parser (delivered with the SUN Java SDK) is used, but any other validating XML Parser ould also be used. We use a DTD (Doument Type Definition) for the verifiation of the FDL trees. The purpose of a DTD is to define the legal building bloks of an XML doument. It defines the doument struture with a list of legal elements. 6. Conlusions and Future Work We presented a framework for flexible and interative, high-dimensional feature speifiation for data oming from omputational simulation. For analyzing simulation data, a feature-based F+C visualization is a good approah, to ope with the data sets large and high-dimensional nature and to guide the user and support interative analysis. For F+C visualization interative fous speifiation is very useful, if real-time updates of multiple linked views are available. Atual features in simulation data often only are aptured with a omplex type of speifiation (hierarhial speifiation, multiple data hannels involved). This is why we believe, that using a simple language to define features hierarhially, namely our feature definition language, helps to extrat and manage features during an interative analysis session. In ombination with using multiple InfoViz views (for data examination and feature speifiation) and SiViz views (for F+C visualization of the interatively extrated features) it is a very useful approah. Future work will inlude extensions of the here presented FDL as well as of the analyzing tools. A parallel oordinates view 10 whih has been developed earlier 8 an already be used passively to visualize the high-dimensional data, and will be integrated fully in the very near future, as well as new (hardware- and software-based) volume rendering tehniques will be inluded, too. FDL extensions will mainly deal with inluding the views setting and ouple it more tightly with the feature speifiation tree, as well as timedependent issues. Currently only steady simulation data an be visualized and the logial next step will be, to enhane the FDL as well as all the orresponding visualization and inter- 247
Doleish, Gasser, Hauser / Interative Feature Speifiation for F+C Visualization of Complex Simulation Data Figure 9: Step 5 of analysis: interative probing of V- veloity reveals different behavior of vortial strutures, only downfaing parts are shown here. ation views to ope with time-dependent data sets. Feature speifiation for time-dependent data sets will be one of the key-issues of future researh. Aknowledgements This work has been arried out as part of the basi researh on visualization at the VRVis Researh Center in Vienna, Austria (http://www.vrvis.at/vis/), whih partly is funded by an Austrian researh program alled Kplus. All data presented in this paper are ourtesy of AVL List GmbH, Graz, Austria. The authors would like to thank Robert Kosara, for his help with preparing this paper. Speial gratitude goes also to our ollegue Markus Hadwiger, who helped with parts of the implementation of the underlying mesh-library system, and the ollegues from the Software Competene Center in Hagenberg, Austria, who helped with their knowledge about fuzzy sets and fuzzy ombinations. Referenes 1. R. Beker and W. Cleveland. Brushing satterplots. Tehnometris, 29(2):127 142, 1987. 2. Andreas Buja, John A. MDonald, John Mihalak, and Werner Stuetzle. Interative data visualization using fousing and linking. In Pro. of IEEE Visualization 91, pages 156 163. 3. S. Card, J. MaKinlay, and B. Shneiderman. Readings in Information Visualization: Using Vision to Think. Morgan Kaufmann Publishers, 1998. 4. Helmut Doleish and Helwig Hauser. Smooth brushing for fous+ontext visualization of simulation data in 3D. In Pro. of WSCG 2002, Plzen, Czeh Republi. 5. Ying-Huey Fua, M. O. Ward, and E. A. Rundensteiner. Struture-based brushes: A mehanism for navigating hierarhially organized data and information spaes. 248 IEEE Trans. on Visualization and Computer Graphis, 6(2):150 159, 2000. 6. George W. Furnas. Generalized fisheye views. In Pro. of the ACM CHI 86 Conf. on Human Fators in Computing Systems, pages 16 23, 1986. 7. D. L. Gresh, B. E. Rogowitz, R. L. Winslow, D. F. Sollan, and C. K. Yung. WEAVE: A system for visually linking 3-D and statistial visualizations, applied to ardia simulation and measurement data. In Pro. of IEEE Visualization 2000, pages 489 492, 2000. 8. H. Hauser, F. Ledermann, and H. Doleish. Angular brushing of extended parallel oordinates. In Pro. of IEEE Symp. on Information Visualization, pages 127 130, 2002. 9. H. Hauser, L. Mroz, G. I. Bishi, and E. Gröller. Twolevel volume rendering. In IEEE Transations on Visualization and Computer Graphis, volume 7(3), pages 242 252. IEEE Computer Soiety, 2001. 10. A. Inselberg and B. Dimsdale. Parallel oordinates: a tool for visualizing multidimensional geometry. In Pro. of IEEE Visualization 90, pages 361 378. 11. E. P. Klement, R. Mesiar, and E. Pap. Triangular Norms, volume 8 of Trends in Logi. Kluwer Aademi Publishers, Dordreht, 2000. 12. A. Martin and M. O. Ward. High dimensional brushing for interative exploration of multivariate data. In Pro. of IEEE Visualization 95, pages 271 278. 13. Webpage of the World Wide Web Consortium on XML. See URL http://www.w3.org/xml/. 14. F.H. Post, H. Hauser, B. Vrolijk, R.S. Laramee, and H. Doleish. Feature extration and visualization of flow fields. In Eurographis State of the Art Reports, pages 69 100, 2002. 15. Ben Shneiderman. Dynami queries for visual information seeking. Tehnial Report UMCP-CSD CS-TR- 3022, Department of Computer Siene, University of Maryland, College Park, Maryland 20742, U.S.A., January 1994. 16. C. Silva, J. Mithell, and P. Williams. An exat interative time visibility ordering algorithm for polyhedral ell omplexes. In Pro. of IEEE Symp. on VolVis 98, pages 87 94, 1998. 17. M. O. Ward. XmdvTool: Integrating multiple methods for visualizing multivariate data. In Pro. of IEEE Visualization 94, pages 326 336. 18. Pak Chung Wong and R. Daniel Bergeron. Multiresolution multidimensional wavelet brushing. In Roni Yagel and Gregory M. Nielson, editors, Pro. of the Conf. on Visualization, pages 141 148, Los Alamitos, Otober 27 November 1 1996. IEEE.
Doleish, Gasser, Hauser / Interative Feature Speifiation for F+C Visualization of Complex Simulation Data Figure 10: Two examples for feature-based flow visualization using our framework for interative feature speifiation and four illustrations of different ombination modes for smooth brushes (middle row) (for verbose aptions see figures 1, 3, and 8). 302