Binary cross coupled discriminant analysis for visual kinship verification

作者：

Highlights：

• In this article, a novel binary cross-coupled discriminant analysis (BC2DA) based feature descriptor is proposed. It is a pixel-wise encoding method that converts the original kin pairs into decimal encoded kin image pairs. The encoding is done to increase the correlation between input pairs. This results in reducing the discrimination between kin pairs at the feature extraction stage itself.

• To understand the core of a family, a new feature descriptor multi cross-coupled discriminant analysis method (MC2DA) is proposed. Tri-subject encoding leads to an increase in the correlation between child with both parents without affecting relations between child and non-family members.

• In both cases various facial sections are consider for analysis of verification performance such as full-face (FF), right face (FR), left face (FL), lower face (FLo), upper face (FU), facial key points (FK), and facial smooth region (FS).

摘要

•In this article, a novel binary cross-coupled discriminant analysis (BC2DA) based feature descriptor is proposed. It is a pixel-wise encoding method that converts the original kin pairs into decimal encoded kin image pairs. The encoding is done to increase the correlation between input pairs. This results in reducing the discrimination between kin pairs at the feature extraction stage itself.•To understand the core of a family, a new feature descriptor multi cross-coupled discriminant analysis method (MC2DA) is proposed. Tri-subject encoding leads to an increase in the correlation between child with both parents without affecting relations between child and non-family members.•In both cases various facial sections are consider for analysis of verification performance such as full-face (FF), right face (FR), left face (FL), lower face (FLo), upper face (FU), facial key points (FK), and facial smooth region (FS).

论文关键词：Kinship verification,Feature descriptor,Facial section,Bi-subject relationship,Tri-subject relationship

论文评审过程：Received 18 December 2021, Revised 28 March 2022, Accepted 12 July 2022, Available online 22 July 2022, Version of Record 2 August 2022.

论文官网地址：https://doi.org/10.1016/j.image.2022.116829