The automatic verification of kinship is a challenging problem that has recently attracted much interest from the research community. It consists in telling whether two individuals are related or not, based on the analysis of their facial images. This is a challenging task since it has to deal with differences in race, gender and age between subjects. In addition, the unpredictable amount of genetic information shared by relatives reflects into individuals showing different degrees of facial similarity. Kinship recognition in the wild introduces more difficulties, since the images to be analyzed can have low resolutions, different illuminations, resolutions, face orientations, expressions and occlusions. Due to the characteristics of the image in analysis, which highly reduces the discriminative power of local features, we address kinship recognition in the wild with a multi-perspective holistic approach. The image pairs to be labeled as kin or non-kin are first characterized by selecting the most relevant variables from the combination of different global textural features. The resulting feature vectors are then used to feed an SVM classifier, which has been assessed on the Kinship Face in the Wild dataset over different sub-classes of parent-child relationships. Results of our experiments show that our method provides optimal accuracies with respect to other approaches on the same data and outperforms the recognition abilities of human beings.

A Multi-perspective Holistic Approach to Kinship Verification in the Wild / Bottino, ANDREA GIUSEPPE; UL-ISLAM, Ihtesham; FIGUEIREDO VIEIRA, Tiago. - (2015). (Intervento presentato al convegno bWild 2015 tenutosi a Ljubljana, Slovenia nel May 4-8, 2015).

A Multi-perspective Holistic Approach to Kinship Verification in the Wild

BOTTINO, ANDREA GIUSEPPE;UL-ISLAM, IHTESHAM;FIGUEIREDO VIEIRA, TIAGO
2015

Abstract

The automatic verification of kinship is a challenging problem that has recently attracted much interest from the research community. It consists in telling whether two individuals are related or not, based on the analysis of their facial images. This is a challenging task since it has to deal with differences in race, gender and age between subjects. In addition, the unpredictable amount of genetic information shared by relatives reflects into individuals showing different degrees of facial similarity. Kinship recognition in the wild introduces more difficulties, since the images to be analyzed can have low resolutions, different illuminations, resolutions, face orientations, expressions and occlusions. Due to the characteristics of the image in analysis, which highly reduces the discriminative power of local features, we address kinship recognition in the wild with a multi-perspective holistic approach. The image pairs to be labeled as kin or non-kin are first characterized by selecting the most relevant variables from the combination of different global textural features. The resulting feature vectors are then used to feed an SVM classifier, which has been assessed on the Kinship Face in the Wild dataset over different sub-classes of parent-child relationships. Results of our experiments show that our method provides optimal accuracies with respect to other approaches on the same data and outperforms the recognition abilities of human beings.
File in questo prodotto:
Non ci sono file associati a questo prodotto.
Pubblicazioni consigliate

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/11583/2592175
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo