A novel factor analysis-based metric learning method for kinship verification

• Published in: Multimedia tools and applications Vol. 81; no. 8; pp. 11049 - 11070
• Main Authors: Qin, Xiaoqian, Liu, Dakun, Wang, Dong
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.03.2022; Springer Nature B.V
• Subjects: Algorithms; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Discriminant analysis; Factor analysis; Machine learning; Multimedia Information Systems; Optimization; Similarity; Special Purpose and Application-Based Systems; Teaching methods; Verification; Kinship verification; Metric learning; Factor analysis
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-022-12032-w

This paper presents a novel factor analysis-based metric learning (FAML) method for kinship verification. While metric learning has achieved reasonably good performance in kinship verification, most existing metric learning methods ignore to discover semantically meaningful similarity-patterns for kinship pairwise data. To address this, we propose a FAML method to seek hidden local similarity metrics, under which kin pairs would be relatively correlated and similar in certain facial regions. Particularly, to learn such local similarity metrics, we apply a series of transformations such as orthogonal rotation and thresholding to the factor loading matrix obtained through factor analysis. Moreover, instead of only seeking metrics in a local sense, we aim to simultaneously learn a set of distance metrics to integrate the locality with the globality, thus being more robust for diversified similarity-patterns that kin pairs contain. To jointly perform genetic characteristics exploiting and metric learning, we present an efficient algorithm that employs alternating optimization to integrate prior knowledge about genetic characteristics into metric learning, such that more discriminative information can be exploited in more fine-grained details for verification. Experiments are carried out on three face kinship datasets, and the results achieved clearly demonstrate the effectiveness of the proposed method.