"Face Kinship Verification Using Retinex Filter and Elliptical Mask based New Version of Gabor Filter"

Abstract

The human facial appearance is an important cue for genetic similarity measure because children and their parents are biologically related and children usually resemble their parents more than other adults. Motivated by this finding, computer vision researchers have investigated the problem of kinship verification via facial image analysis while some encouraging results have been obtained, this problem still remains unsolved especially when face images were captured in unconstrained an environment which makes this problem extremely challenging. In our thesis, we present a kinship verification system that starts with pair of facial images of the child and parent, then as a final result is determine whether two persons have a kin relation or not. The first step in our approach is the face preprocessing with two methods Retinex filter and Elliptical mask, then feature extraction stage based on hist-Gabor wavelets that is used before the powerful dimensionality reduction method called TXQDA. Finally, decide whether there is a kinship or not. We conducted several experiments to demonstrate the effectiveness of our approach with Cornell KinFace benchmark database. Our results indicate that our system is robust compared to other existing approaches.