Keywords: Label Layout, Interactive Illustrations, ... from hand-made illustrations and implemented them ... his module determines the best alternative for ploys a ...
Adaptive Labeling for Illustrations Timo Götzelmann, Kamran Ali, Knut Hartmann, and Thomas Strothotte Department of Simulation and Graphics Otto-von-Guericke University of Magdeburg Universitätsplatz 2, D-39106 Magdeburg, Germany E-mail: {timo, kamran, knut, tstr}@isg.cs.uni-magdeburg.de
Abstract Labels are an effective mean to visualize the association between textual and graphical information. Therefore, they are often employed in tutoring materials and technical documentations to support learning tasks. In this paper we present a novel system that integrates real-time layout algorithms for internal as well as external labels. This enables to display a maximum number of labels within an interactive 3D illustration. Keywords: Label Layout, Interactive Illustrations, Computer Supported Learning.
1. Introduction Effective tutoring materials employ complex illustrations which are associated with textual annotations offering verbal explanations for visual objects. Human illustrators employ a number of different techniques to integrate textual and visual information: labels, legends, and figure captions. Textual labels can either overlay their co-referential visual objects (internal label) or are arranged on the background (external label). Additional graphical objects (anchor points, connecting lines) might emphasize the co-referential relation between external labels and their associated visual objects. In practice, illustrations employ a variety of different labeling styles for internal and external labels. The choice of label layout styles reflects both space restrictions imposed by the overall layout as well as subjective preferences of illustrators. Several research prototypes incorporate interactive 3D browser to explore complex spatial configurations [8,10,11], but there are very few solutions for automated labeling. We believe that an aesthetic label layout can increase the learning efficiency of interactive 3D tutoring systems. Therefore, we extracted a set of metrics from hand-made illustrations and implemented them in real-time layout algorithms. The users can freely
explore 3D scenes while an aesthetic label layout is maintained automatically at interactive rates. Our system implements several labeling styles and employs several strategies to achieve a frame-coherent layout (i.e., to minimize visual disturbance caused by moving layout elements during user interaction).
2. Related Work Hand-made illustrations in scientific and technical textbooks commonly employ a large number of internal and external labels [7]. They provide denotations for visual objects, contain instructional information, or guide the user attention to important visual objects. Labels are thus an important tool for instructional images such as explosion diagrams [8]. Several researchers incorporate labels within interactive 3D information systems [10,11]. However, the automatic computation of an effective and aesthetic label layout is often considered to be a complex and computationally very slow optimization task [4,5] that is inappropriate for real-time applications. In AR/VR the term view management [2,3] refers to a related, but a more general task, the integration of 2D information into 3D scene. The label layout algorithms within these research prototypes neither achieve the quality nor the number of different layout styles as found in hand-made illustrations in scientific and technical documentations. Most of the existing layout algorithms either rely on user interactions [8,11] or are based on rough shape approximations [3]. Two recent technical papers discuss metrics which are employed in real-time layout algorithms to achieve various labeling styles for external [1] and internal labels [6] and their integration within an interactive 3D browser. These layout algorithms also consider the correct shape of complex 3D objects.
3. Architecture The label layout methods have to consider geometric properties (the shape the projected 3D objects), space
restrictions as well as user preferences. An internal labeling is appropriate for objects which are big enough to accommodate labels. If there is not sufficient space, the labels cannot be displayed internally without causing overlaps with other objects. The same problem occurs vice versa for external labels when the space outside of the 3D model gets too small. In order to meet different global requirements effectively, the labels should be classified as external or internal. We designed an architecture that considers these requirements during label classification (internal or external) as well as content selection (see Figure 1). The analysis module extracts information about the shape of the individual 3D objects on the view plane. The classification module uses that information to decide whether to place a label inside or outside of the object. This decision is forwarded to the particular layout modules of the layout manager. Each of these modules then handles the associated subset of labels and renders them on Figure 1: Architecture top of the rendered image.
3.1 Analysis This module analyses color-coded projections of the current 3D scene with typical image analysis algorithms. We experimented with several skeletonization algorithms as they heavily differ in their computational complexity, their robustness to small changes of the shape and the quality of the resulting skeleton. We chose a scan line algorithm due to it low computational complexity and its robustness. This module creates a skeleton graph that represents all possible labeling paths. Later on, several quality metrics [6] are applied on these labeling paths to select the best path (according to our weighted criteria) and to determine the path quality (score). The paths and their score are passed to the classification module.
3.2 Classification This module determines the best alternative for labeling an object. Due to the spatial continuity principle [9] that recommends placing related items as close as possible, internal labels are preferred over
external labels. The classifier employs a set of rules: Available Space: If there is not sufficient space to label the object internally (even within the smallest level of detail) an external label is chosen. Enforce external: This criteria aims to guarantee a readable label layout. A poor path score of an internal label (e.g., due to high curvature) would lead to the selection of an external label. Enforce internal: If there is even not enough background space to place external labels, the cut-off parameter to choose external labels is adjusted to force the selection of internal labels for as many objects as possible.
3.3 Layout Manager The layout manager determines the best label positions for a given label classification. It incorporates two independent modules for internal and external labels that work on their own label subsets. Internal Labeling: This module evaluates the label path in order to determine if a horizontal labeling is possible or the label should be aligned to the medialaxis of the object. Then the best positions [6] of the individual text letters of the label are calculated and the letters are rendered onto the scene. Abbreviated annotations could be chosen due to space restrictions. External Labeling: This module computes the anchor points on the objects, and then determines the best positions of the labels over the background space. Different external labeling styles [1] can be selected for an alternative label placement. Finally the connecting lines are added to the scene.
4. Results Our prototypical application employs the Open Inventor library. An external module (domain expert) provides the label content at different levels of detail. If this information is not available, the layout modules display internal object descriptors of the 3D model as internal and external labels. Our system computes a label layout at interactive rates. Thus, the users are able to (1) interact with 3D models to get an impression about the spatial configuration and (2) benefit from the annotations added to the 3D illustration. During user interaction, the labeling style for each label is adapted dynamically to the current view of the object. This increases potentially the number of labels which can be displayed independent of the current view. Moreover, the labeling scheme follows the spatial continuity principle. Figure 2 shows three frames from a zooming sequence which exemplifies the dynamic label adaptation.
5. Future Work Although, both internal and external layout modules implement some techniques for a frame-coherent label layout, the transition from an internal label to the external label or vice versa is not coherent. In addition, connecting lines may cross internal labels. At the moment there is no support for multi-line labels. Moreover, the readability of the internal labels is not optimal due to lack of contrast and aliasing effects. In future, we plan to integrate information retrieval methods to update the label contents dynamically. We also plan to perform a user study to find out optimal parameters for the weighted classification algorithm and the individual parameters of the layout modules.
[8] W. Li, M. Agrawala, and D. Salesin, “Interactive ImageBased Exploded View Diagrams”, In Proc. of Graphics Interface 2004 (GI’04), pages 203–212, 2004. [9] R. Mayer, Multimedia Learning, Cambridge University Press, 2003. [10] B. Preim, A. Raab, and T. Strothotte, “Coherent Zooming of Illustrations with 3D-Graphics and Text”, Proc. of Graphics Interface’97, pages 105–113, 1997. [11] F. Ritter, H. Sonnet, K. Hartmann, and T. Strothotte, “Illustrative Shadows: Integrating 3D and 2D Information Displays”, In Proc. of Int. Conf. on Intelligent User Interfaces (IUI’03), pages 166–173, 2003.
6. Conclusion This paper summarizes a new method for automated textual annotation of interactive 3D illustrations. The label layout manager considers the available screen space during label classification and uses different representations of the textual information. The label classification module to decide internal and external labels as well as specialized layout modules for both labeling types exploit a common data structure - the skeleton graph. The layout algorithms can be adapted according to layout requirements and individual preferences. Finally, our system first implements different label layout styles in real-time.
7. References [1] K. Ali, K. Hartmann, and T. Strothotte, “Label Layout for Interactive 3D Illustrations”, Journal of the WSCG, 13(1):1–8, 2005. [2] R. Azuma and C. Furmanskim, “Evaluating Label Placement for Augmented Reality View Management”, In Proc. of the IEEE and ACM Int. Symposium on Mixed and Augmented Reality (ISMAR 2003), pages 66–75, 2003. [3] B. Bell, S. Feiner, and T. Höllerer, “View Management for Virtual and Augmented Reality”, In Proc. of Symposium on User Interface Software and Technology (UIST’01), pages 101–110, 2001. [4] J. Christensen, J. Marks, and S. Shieber, “An Empirical Study of Algorithms for Point-Feature Label Placement”, ACM Transactions on Graphics, 14(3):203–232, 1995. [5] S. Edmondson, J. Christensen, J. Marks, and S. Shieber, “A General Cartographic Labeling Algorithm”, Cartographica, 33(4):13–23, 1997. [6] T. Götzelmann, K. Ali, K. Hartmann, and T. Strothotte, “Interactive Labels“, In Computational Aesthetics in Graphics, Visualization and Imaging, 2005 (submitted). [7] H. Gray, “Anatomy of the Human Body”, Lea & Febiger, Philadelphia, 20th edition, 1918.
Figure 2: Zooming into the 3D heart model.