Impact of viewpoint
To address the final question, a 2 × 3 mixed Analysis of Variance (ANOVA) was conducted to explore accuracy of performance when matching hands and ears across good, medium and poor viewpoints. For this analysis, accuracy levels were standardised by expressing them as a proportion of the performance level attained in the optimal (good) condition (see Table
1). This ensured a focus on the relative impact of a
change in viewpoint, and prevented the findings being affected by variation in absolute levels of performance across the stimuli.
The ANOVA revealed a main effect of stimulus type (F
(1, 48) = 41.59, p < 0.001, partial η2 = .464), with better overall performance for hands than for ears. In addition, a main effect of viewpoint emerged (F
(2, 96) = 409.52, p < 0.001, partial η
2 = 0.895), with better performance when presented with more optimal viewpoints. These effects were qualified by the expected interaction between stimulus type and viewpoint (F
(2,96) = 24.79, p < 0.001, partial η
2 = 0.34).
Analysis of the simple main effects confirmed a significant effect of viewpoint for both hands (F
(2,48) = 47.27, p < 0.001, partial η
2 = 0.66) and ears (F
(2,48) = 233.48, p < 0.001, partial η
2 = 0.907) suggesting that the performance for both stimulus types suffered as the view became less optimal. However, a series of Bonferroni-corrected comparisons confirmed that performance with hands was not affected by a change from good to medium images (t
(24) = 2.04, p > 0.05) but was only affected by a change from medium to poor images (t
(24) = 6.72, p < 0.001). In contrast, performance with ears was affected as soon as the image moved away from optimal, with significant differences in performance levels between good and medium images (t
(24) = 14.92, p < 0.001) as well as between the medium and poor images (t
(24) = 4.16, p < 0.001).
In accounting for these results, it was possible that ear processing was more affected by a change in viewpoint than hand processing because ear processing was an inherently difficult task. Important in this regard was the demonstration of equivalent absolute levels of performance in the best image case (t
(48) = 2.48, ns) despite the differences between hands and ears as stimuli. Consequently, the substantial impact of viewpoint for ears could not easily be attributed to an inherent difficulty when matching ears. However, the possibility remained that the difficulty when matching ears was revealed not in baseline performance levels, but in a greater vulnerability as the image quality was changed. Such an explanation was compatible with the predictions for this study in which the flexibility of the hand was expected to minimise the impact of a sub-optimal viewpoint. Indeed, these two accounts would be difficult to separate out.
Taking all analyses together, the results of Experiment 1 provided support for the predictions. Specifically, the change in viewpoint had a significant effect when matching hands, but had a greater effect, from an equivalent starting point, when matching ears. These results supported the prediction that the inherent flexibility of the hand-enabled exposure to a variety of viewpoints with the consequence that canonicality was less strong for hands than ears.
In terms of implications for the hand as a biometric, the data here led to the conclusion that when matching hands, performance could survive moderate changes in viewpoint whereas when matching other more rigid biometrics (such as ears), a change in viewpoint compromised performance quite substantially. As such, these data confirmed a greater reliability of the hand as a biometric cue across optimal and moderately optimal viewing conditions.
Several aspects of the current results were interesting and unanticipated, and as such warrant some consideration. In particular, it was interesting to note impairment in the performance of the two experts as viewpoint changed. Whilst it was not possible to assess the extent of the impact of viewpoint statistically for each of the experts (there being only one expert for each stimulus type), it was possible to determine whether the experts were affected to the same degree as the novice participants.
To this end, a series of one-sample t tests was conducted, comparing the decline in performance shown by the expert, to the decline in performance shown by the group of novices. This confirmed that novice performance declined more than expert performance as the viewpoint became less optimal. This was evident when matching ears as the image changed from good to medium (ears: t
(24) = 6.25, p < 0.001; hands: t
(24) = 1.08, ns), and when matching both ears and hands as the image changed from medium to poor (ears: t
(24) = 8.64, p < 0.001; hands: t
(24) = 11.23, p < 0.001). Consequently, these results suggested that whilst the experts were affected by a change in viewpoint, they were affected less than novices.
This latter analysis did not sit within the main purpose of this Experiment but nevertheless raised questions: For example, could the provision of training be sufficient to improve performance levels from that of the novice towards that of the expert. Relatedly, could the provision of training ameliorate the negative impact of the sub-optimal viewpoint so that trained participants come to show greater resilience than novices when presented with sub-optimal viewpoints?
Whilst representing an important applied issue, such questions relate well to the theoretical consideration of Blanz et al., (
1999) regarding the criteria underpinning a canonical view. Indeed, it may be argued that expertise brings with it a capacity to use a range of cues so that the matching task can still be completed even when a subset of the cues is unavailable through occlusion in a sub-optimal image. Similarly, it may be argued that expertise brings the capacity to show better understanding of function, and greater levels of exposure to non-standard viewpoints through expert study. All factors may lead to the prediction that canonicality is less strong (or the negative impact of a non-canonical image can more easily be overcome) when the viewer brings expertise to their viewing task.
Experiment 2 was conducted to present an examination of these emergent questions. Through the provision of video instruction, the performance of a group of ‘trained’ participants was compared to that of the novices and experts studied in Experiment 1. It was anticipated that training would improve overall levels of performance, and would reduce the impact of a change in viewpoint compared to the novices such that the performance of the trained group would more closely resemble that of the experts.