• No results found

Factors affecting the enhanced distinctiveness of familiar faces

2.5.1 Degree of familiarity and distinctiveness boost

To test whether the degree or level of familiarity with a face has an influence on its relatively enhanced perceived distinctiveness, we grouped faces by type of familiarity relationship and plotted

acquaintance (2−6) good friend (7−8) sister/self (9−10) −0.1 −0.05 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 p<0.00032 p<0.011 Familiar − Unfamiliar* (distinguishability difference)

*familiar with exactly one face

Boost in distinguishability depends on level of familiarity

Figure 2.19: The degree of familiarity enhances distinguishability of faces (analysis is based on pairs of half-familiar pairs). p-values are based on 2-sided t-tests.

the boost in distinguishability when comparing with an unfamiliar face. We find that, on average, faces of oneself or one’s sister, or that of a good friend (familiarity 7 - 8), receive a larger boost in distinctiveness relative to less familiar faces. The result is shown in Figure 2.19, which is essentially the average displacement above the diagonal in Figure 2.11 by familiarity type, except leaving unique second viewers as unique data points for more statistical power. We note that we left out face pairs which had confusability score of less than 0.2 (33rd percentile), rated by the other 8 viewers, because, as we shall discuss shortly, face pairs which are already easily distinguished do not gain much in distinguishability from familiarity with one of them.

We find the same effect of greater familiarity level on face pair dissimilarity judgments. This is shown in Figure 2.20. As with distinguishability, we find that the more a face is familiar, the more its dissimilarity to other unfamiliar faces is enhanced. This boost is of greater magnitude for whole faces, due likely to more complete recognition of the face, but the secondary trend of increasing with familiarity level is weaker. One reason for this might be that, with outer features revealed, even weakly familiar faces are easily recognized and thus perceived differently. The analysis is done exactly analogously to the distinguishability analysis, including leaving out face pairs which are already too dissimilar, as they are not affected by the familiarity of one of the faces (we leave out faces below the 60th percentile of similarity judgment by the other 8 viewers).

acquaintance (2−6) good friend (7−8) sister/self (9−10) −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 p<0.00063 p<0.0069 p<0.0068

*familiar with exactly one face

Familiar

Unfamiliar*

(dissimilarity difference)

Boost in dissimilarity perception depends on level of familiarity

Whole Face (pairs others rated > 0.17) Inner Only (all qualifying pairs)

Figure 2.20: The degree of familiarity enhances dissimilarity judgments (analysis is based on pairs of half-familiar faces). The dissimilarity boost is generally larger for judgments between whole faces (including outer features) than for pairs showing only inner features, possibly because faces are recognized more readily and completely with jaw and hairline cues. p-values are based on 2-sided t-tests.

2.5.2 Baseline similarity/confusability and distinctiveness boost

0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 p<0.021 p<0.059 p<0.068

Confusability by Other Viewers (Non−Overlapping Bins) *familiar with exactly one face

Familiar

Unfamiliar*

(distinguishability difference)

Magnitude of familiarity effect depends on confusability

Figure 2.21: The extent to which a face pair becomes more distinguishable to a familiar viewer depends on the underlying confusability of the pair.

As mentioned above, face pairs which are already very dissimilar or distinguishable, even without the benefit of familiarity, do not gain much perceived dissimilarity or distinguishability when one of the faces in the pair is familiar. This is reflected in the left part of Figure 2.21 and Figure 2.22, showing a relatively smaller distinctiveness boost for the least similar face pairs. For these analyses, the baseline similarity and confusability were established by taking the average among the 8 remaining subjects not used to compute the difference (between the half-familiar’s and fully unfamiliar viewer’s similarity ratings or confusability level). However, whereas the most confusable face pairs show a relativelysmaller distinctiveness boost (compared to ones in the middle of the baseline confusability range), the most similar face pairs are the ones which show the greatest average distinctiveness boost. We thought perhaps the difference in some of the experimental parameters (outer vs. inner faces, different numbers of face pairs across experiments) might be responsible for this discrepancy, but the same general trends hold, even when these differences are eliminated. One explanation is that the MorphDiscrim task was simply too difficult for subjects near the high end of baseline

face pair confusability, and so performance there floored, whereas face pairs near the middle of the confusability levels just admitted a broader performance range within which the familiarity effect could be measured: this would suggest that perhaps the most similar looking face pairsactually are

most affected by familiarity, both in terms of dissimilarity perception and distinguishability, but that the nature of our experimental setup prevented the accurate measuring of the latter. In either case, it is clear that the least confusable or similar face pairs are perceptually affected least from familiarity. −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 p<0.088 p<0.039

Similarity Rated by Other Viewers (Non−Overlapping Bins) *familiar with exactly one face

Familiar

Unfamiliar*

(dissimilarity difference)

Magnitude of familiarity effect depends on similarity

Figure 2.22: The extent to which a face pair becomes more dissimilar to a familiar viewer depends on the underlying similarity of the pair.