[Back]

Tri-subject Kinship Face Database (TSKinFace)


TSKinFace Database [1]

Overview:

The aim of kinship verification through computer vision is predicting whether given images have kin relation. Kinship verification is an emerging problem which is full of challenging due to the fact the appearance gap encountered in a kinship problem is much larger than that in a conventional face recognition setting(e.g, given two face images with different sex and different ages, verify whether those subjects are father and daughter).

Most of current research,however,mainly focus on the one-versus-one[2] or one-versus-multi[3] kinship realtion.Instead we focus on the child-parents(one-versus-two) relationship. We call this tri-subject kinship verification. To investigate the performance of tri-subject kinship verification for this scene, we collected a database in the Wild. All images in the database are harvested from the internet based on knowledge of public figures family and photo-sharing social network such as flickr.com. Family facial images are collected based on the face region and eye position automatically and respectively estimated by the face detector[4] and eye localization[5]. We resize the cropped faces to the size 6464 (pixels).

In particular, we are interested in three kinds of child-parents families in real life, i.e., Father-Mother-Daughter (FM-D), Father-Mother-Son (FM-S) and Father-Mother-Son- Daughter (FM-SD). For each type, we collected 274, 285 and 228 family photos respectively, with one photo per family. Using these, we constructed two kinds of family-based kinship relations in the TSKinFace database: Father-Mother-Son(FM-S) and Father-Mother-Daughter(FM-D). The FM-S and the FM-D contain 513 and 502 groups of tri-subject kinship relations, respectively. Hence we have 1015 tri-subject groups in our database totally. The families included in our database are diverse in terms of races as well. For FM-S relation, there are 343 and 170 groups of tri-subject kinship relations for Asian and non-Asian, respectively. And for FM-D relation, the numbers for Asian and non-Asian groups are respectively 331 and 171.

Figure1 shows some image groups of child-parents pair from our TSKinFace database.

Figure 1: Some family image groups of our TSKinFace database, where each group consists of a family triple of a father, a mother and a child. The first row shows three Father-Mother-Daughter (FM-D) relation families, respectively and the second row are three Father-Mother-Son (FM-S) relation families, accordingly.

How to Get It?

We provide two versions of the database for you to download: 1) raw family images with background, 2) group images warped,as follows, [download].  SIFT features   download.

(1): family images in original resolution.

(2): group images in size of 6464.

Other Materials Related to [1] 

KinFaceW database one-versus-one kinship images used in [1] from this dataset.  

Family101: one-versus-multi kinship images used in [1] from this dataset.  

Codes in Matlab

  download.

References:

[1] Xiaoqian Qin, Xiaoyang Tan,Songcan Chen, Tri-Subject Kinship Verification: Understanding the Core of A Family, IEEE Transactions on Multimedia,  Accepted, 2015.

[2] J. Lu, X. Zhou, Y.-P. Tan, Y. Shang, J. Zhou, Neighborhood repulsed metric learning for kinship verification, IEEE Trans. Pattern Anal. Mach. Intell. 36 (2) (2014) 331C345.

[3] R. Fang, A. C. Gallagher, T. Chen, A. Loui, Kinship classification by modeling facial feature heredity, in: ICIP13, 2013.

[4] P.Viola and M.J.Jones. Robust real-time face detection. International Journal of Computer Vision (IJCV), 57(2):137-154, 2004.

[5] X.Tan, F.Song, Z-H.Zhou and S.Chen. Enhanced Pictorial Structures for Precise Eye Localization under Uncontrolled Conditions, In: Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'09),pp. 1621C1628, Miami, Florida, USA, June 2009.

Copyright and disclaimer:

Copyright 2015, Xiaoyang Tan

The dataset is provided for research purposes to a researcher only and not for any commercial use. Please do not release the data or redistribute this link to anyone else without our permission. Contact {x.tan}@nuaa.edu.cn if any question.

If you use this dataset, please cite it as,

Xiaoqian Qin, Xiaoyang Tan,Songcan Chen, Tri-Subject Kinship Verification: Understanding the Core of A Family.  IEEE Transactions on Multimedia, 2015