Automatic Visual Kinship Veri cation

Abstract

Veri cation of kinship from facial images is attracting more and more attention from the research community, is an emerging research topic in computer vision. Checking whether two people are from the same family or not can be automatically checked by facial images. Many potential applications: such as creating family trees, organizing family albums, an- notating images; the search for missing children and forensic medicine, are targeted by the veri cation of kinship. This paper presents a successful kinship veri cation system, which utilizes two consec- utive methods (MSR+NDM) in the image preprocessing stage to enhance image quality and overcome issues relating to contrast, lighting, and noise. Additionally, we propose a new descriptor based on the histograms of a Two dimensional Discrete Wavelet Trans- form (Hist-2D DWT). We further investigate the complementarity of handcrafted (LPQ, Hist-2D DWT) and deep features (VGG16, ResNet50) by fusing them at the score level using the Logistic Regression method. Extensive experiments conducted on two kinship datasets, veri cation accuracies of 95.18% and 91.81% have been reached under Cornell KinFace and TS KinFace datasets.