Research plan

An outline of the steps of the proposed research is presented in the following sections. The L-systems-based visualization system developed by Pinkney will be used as the test-bed for topics being researched.

1. Framework to integrate the iconographic technique with other visualization tools
   1. Multiple alternate views
      One complete iconographic display visualized using another visualization technique. In the initial stage parallel coordinates will be used to represent the same number of dimensions mapped to the icons.
   2. Brushing and linking
      Selected regions in the iconographic display representation visualized using the selected technique (parallel coordinates). Both representations (iconographic and parallel coordinates) will be linked so any change in one representation will effect the other representation.
   3. Linked and dynamic query-like interactions
      • Interactions involving both representations.
      • Investigate the use of these interactions as a mechanism to dynamically explore different mappings. For example, swapping two axes in the parallel coordinates representation will be equivalent to swap the corresponding mapping in the iconographic representation (selected region). All of these interactions will be based on a direct manipulation paradigm.
      • Investigate the use these interactions for determining relationships between data attributes and patterns that pop up in the iconographic representation. For example, sliders could be associated with each axis in the parallel coordinates representation. These sliders would be combined conjunctively or disjunctively in order to provide a more flexible data selection mechanism.
   4. Extension of the integration The integration model used with the first technique (parallel coordinates) will be extended to other visualization techniques. Other techniques such as RadViz, dimension stacking, scatterplot matrix will be considered.
2. Icon Interactions
   1. Dynamic mapping interaction
      Development of an interaction that changes the mapping in the region defined by the mouse. The mapping could be based on for example global statistics in the data (variance, correlation, standard deviation, etc.). A simple criteria for changing the mapping based on the correlation could be mapping high correlated attributes to the same geometric attribute of the icon (angle, length, or intensity). A random selection could be also considered.
   2. Dynamic visualization interaction
      The interaction in this case would consist in representing the area defined by the mouse using one of the visualization techniques being integrated with the iconographic technique.
   3. Dynamic icon interaction
      The development of an interaction that changes the type of icon in the region defined by the mouse will be also considered. The selection of the icon would be user definable or might be based on some global criteria or even randomly chosen. The icon selection will be static (a fixed icon used to represent the area in the region of interest).
3. Framework to develop icon operations
   Spatial filter operations will be used to manipulate the iconographic display in order to accentuate the presence of patterns in the iconographic display in regions with specific trends.
   1. Extension of spatial filter operations
      • Icons instead of pixels. Icons instead of pixels will be used in implementing these operations. Each icon represented by a box of m n pixels.
      • Neighbor icons: Image operations are applied to a data with a strictly spatial order (the image). Each pixel in the image has a very well defined set of neighbors. For a pixel in the position (i,j) their immediate neighbors are those in the positions (i-1,j-1), (i,j-1), (i+1,j-1), (i-1,j), (i+1,j),(i-1,j+1), (i,j+1), and (i+1,j+1). The same is valid for the iconographic representation but only for coherent data sets. In the case of non-coherent data sets, the position of icons in the iconographic display is determined by the variables chosen for driving the axes. A different approach must be used to determine the neighbors of a particular icon.
      • Degree of neighboring. A degree of neighboring has been defined in order to take into account the difference distances between the icon being processed and their neighbors.
   2. Filter operations. These are some of the filters that will be developed.
      • Low pass filter
      • High pass filter
      • Prewitt spatial filter
      • Laplacian spatial filter
      • Median spatial filter
4. Framework to integrate the iconographic technique with analysis tools
   1. Dynamic analytic support in visualization
      The approach here will be similar to that taken with the integration of visualization tools. First basic statistics for the complete iconographic display will be considered and then the same statistics will be calculated, but on selected regions defined over the iconographic display. Basic statistics include mean, minimum and maximum values, number of different possible values and their distribution, variance, etc.
   2. Automatic analytic support in visualization
      The next step is to include more complex statistics providing some idea about possible relationships between attributes. Correlation and cross-correlation will be considered. The use of data mining tools will also be explored.
   3. The role of artificial data sets.
      Experiments with artificial data, with known structures, will be developed in order to test the identification of some simple patterns in the iconographic display. The idea is to identify how a known structure in the data would look in the iconographic representation and how a particular mapping or icon would effect that representation.