The ability to recognize faces and to detect differences and similarities between faces has proved to be fundamental in the evolution of humans and in the conditioning of their social behaviors. In this paper, we investigate basic mechanisms underlying this ability, focusing in particular on the relevance of local and global features and on some interesting differences characterizing judgments of similarity with respect to judgments of dissimilarity.In a first experiment, a set of participants is involved in order to evaluate the human response with respect to a simple judgment protocol based on two-alternative forced choice. Triplets of face stim uli are evaluated first with the aim of identifying (between two candidate faces) the face more similar to a reference face. The protocol is then repeated for the same triplets but involving a different set of participants and asking to identify the face less similar to a reference face. These visual judgments of similarity and dissimilarity are finally analyzed and compared with the results of a closely related computational experiment based on the same set of triplets: in this case, however, the similarity-dissimilarity measure is derived by automatically extracting facial points and matching with regression techniques (LASSO and Elastic Net) two configurations of image descriptors: the first capturing holistic information, the second capturing local information, that is few localized facial features.Our results suggest that computational models based on holistic cues (emphasizing the concept of the whole as a composed set of interdependent parts) better fit judgments of humans participating to the first experiment (similarity judgments). On the other hand, models based on spatially localized cues do not offer significant accuracy. Vice versa, computational models based on local cues better fit dissimilarity judgments and are less adequate to express similarity information. Notably, our results provide some empirical evidence that local and global cues are both important in face perception, but with different roles. This finding supports the hypothesis that similarity and dissimilarity should not merely be considered as opposing concepts, as they could derive from different processing paths. (C) 2019 Elsevier B.V. All rights reserved.

On the importance of local and global analysis in the judgment of similarity and dissimilarity of faces

Bicego, Manuele;
2019-01-01

Abstract

The ability to recognize faces and to detect differences and similarities between faces has proved to be fundamental in the evolution of humans and in the conditioning of their social behaviors. In this paper, we investigate basic mechanisms underlying this ability, focusing in particular on the relevance of local and global features and on some interesting differences characterizing judgments of similarity with respect to judgments of dissimilarity.In a first experiment, a set of participants is involved in order to evaluate the human response with respect to a simple judgment protocol based on two-alternative forced choice. Triplets of face stim uli are evaluated first with the aim of identifying (between two candidate faces) the face more similar to a reference face. The protocol is then repeated for the same triplets but involving a different set of participants and asking to identify the face less similar to a reference face. These visual judgments of similarity and dissimilarity are finally analyzed and compared with the results of a closely related computational experiment based on the same set of triplets: in this case, however, the similarity-dissimilarity measure is derived by automatically extracting facial points and matching with regression techniques (LASSO and Elastic Net) two configurations of image descriptors: the first capturing holistic information, the second capturing local information, that is few localized facial features.Our results suggest that computational models based on holistic cues (emphasizing the concept of the whole as a composed set of interdependent parts) better fit judgments of humans participating to the first experiment (similarity judgments). On the other hand, models based on spatially localized cues do not offer significant accuracy. Vice versa, computational models based on local cues better fit dissimilarity judgments and are less adequate to express similarity information. Notably, our results provide some empirical evidence that local and global cues are both important in face perception, but with different roles. This finding supports the hypothesis that similarity and dissimilarity should not merely be considered as opposing concepts, as they could derive from different processing paths. (C) 2019 Elsevier B.V. All rights reserved.
2019
Face analysis; Biometrics; Image matching; Similarity-dissimilarity of visual content
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11562/1017192
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? 1
social impact