• No results found

5.5 Experiment 3

5.5.8 User Study 3.3: Global shape task

Subjects: Five graduate students performed the local shape task. All participants had normal or corrected-to-normal vision and were screened for normal color sensitivity using the standard Ishihara test for color blindness. Participants were compensated for their time.

Stimuli: The participants viewed images of the fruit data under the following visualization tech- niques: color mapping, point-correspondence glyphs, principal curvature texture, principal curvature texture with shadows, principal curvature texture with point-correspondence glyphs. The rocking animation present in the previous experiment groups is not included in this user study.

Design: This user study compares the five visualization techniques for enabling the global shape task. Each participant viewed 25 surface pairs per visualization technique, for a total of 125 trials.

Participants estimated the global shape of the intersecting objects and determined if either was a specified target object. Trials were randomly ordered for each participant.

Participants were asked to respond to the following question: Is thefruitpresent in this pair of objects?

The textfruitis replaced with the name of one of the 6 types of fruit in the data set. Training was similar to the other user studies in this experiment group.

Hypothesis

The hypothesis is that the percentage of correct participant responses depends on the visualization technique.

Independent variables:The design directly manipulates one independent variable – the visualiza- tion technique. Also recorded are the following predictor variables: a random unique identifier for each participant, the participant’s gender, and the participant’s response time.

Dependent variable:

For the color mapping technique, I expected participants to perform the task well in the case where the target fruit was the visible surface (8 of 25 trials). I further expected participants to perform at chance if the target fruit was not visible, either as the hidden surface (5 of 25 trials) or not present (12 of 25 trials). I expected an 66 percent correct responses for the color mapping technique. For the remaining techniques, I expected participants to perform well as the techniques would show the fruit silhouettes, which is sufficient to recognize the fruit in most cases.

Results

Unless otherwise noted, independent variables have no statistical significance (p> .05).

Analysis: ANOVA analysis finds significant main effects for the visualization technique (p< .001). Figure 5.18 shows the overall percentages of correct responses and 95% confidence intervals by visualization technique; the figure also shows that the performance of participants is better than chance. A Tukey’s HSD test finds that the color mapping technique produced the worst performance, the point-correspondence technique produced better performance, and the principal curvature texture techniques produced the best performance. No statistical differences can be found within the group of texture techniques.

Responses to the questionnaire agreed with the performance analysis. Participants indicated a marginal overall preference for the principal curvature texture with point-correspondence glyphs tech- nique.

User Study 3.3: Global Shape Task

0 0.2 0.4 0.6 0.8 1

Color Mapping Curvature Curvature + Shadows

Correspondence Curvature + Correspondence

Display Technique

Correct (%)

Figure 5.18: This figure shows the overall percentages of correct responses and their 95% confi- dence intervals for the global shape task. Tukey’s HSD test finds that the techniques fall into three statistically different groups. Principal curvature texture techniques make up the highest performing group. The color mapping technique is the lowest performing technique. The point-correspondence technique falls between the other two groups.

Discussion: The principal texture based techniques provided the best performance. The color mapping technique provided the worst performance. The point-correspondence technique provided performance between these two extremes. Again, principal curvature texture with point-correspondence produced the most correct responses but did not show a statistically significant difference from the next best performer.

The color mapping technique did perform better than expected (66 percent correct responses ex- pected). Further analysis shows that the participants responded as expected in the cases where the specified fruit was present whether it was hidden or not but outperformed expectation when the spec- ified fruit was not present in the visualization. One possible explanation is that certain fruits (such as

the banana) produced easily recognizable color patterns, so it would be relatively easy to determine that it was not present.

5.5.9 Experiment discussion

The red-yellow-green color scale does appear to perform better, at least for directly reading a distance metric from the color scale, than the red-grey-blue color scale. The red-yellow-green scale made no significant difference for the local shape task. The color map visualization also enabled the poorest performance for the global shape task. This is not surprising because the technique shows only a single surface, and the second surface must be inferred from the pattern of color. Those color patterns are often ambiguous, making it difficult to determine the shape of the second surface.

Principal curvature texture alone appears to be a relatively poor technique overall. Although it was in the best performing group for the global shape task, it was in the worst performing group for the other two tasks. Its performance in the global shape task may be entirely attributable to showing the silhouettes of both objects.

Point-correspondence alone performed as expected overall. It was surprising that it was not in the best performing group for the global shape task, but this is more likely attributed to the density of the correspondence glyphs along the interior surface interfering with silhouette completion. The intersection-refactored variant of point-correspondence in this study performed better overall than would an non-refactored version because the line glyphs would not be a useful shape cue in regions where they lie inside the non-refactored surface.

Principal curvature texture with shadows continued to perform at least as well if not better than curvature texture alone. The shadowed version of the technique was also preferred over the unshad- owed version by participants.

Principal curvature texture with point-correspondence glyphs proved to be both the best overall performer and the participant favorite. Participant performance with the technique was in the best performing group for all tasks, and it received the highest scores and ranking in the questionnaire for all tasks.