Gabor-based gradient orientation pyramid for kinship verification under uncontrolled environments

Abstract

This paper presents a Gabor-based Gradient Orientation Pyramid (GGOP) feature representation method for kinship verification from facial images. First, we perform Gabor wavelet on each face image to obtain a set of Gabor magnitude (GM) feature images from different scales and orientations. Then, we extract the Gradient Orientation Pyramid (GOP) feature of each GM feature image and perform multiple feature fusion for kinship verification. When combined with the discriminative support vector machine (SVM) classifier, GGOP demonstrates the best performance in our experiments, in comparison with several state-of-the-art face feature descriptors. Experimental results are presented to show the efficacy of our proposed approach. Moreover, the performance of our proposed method is also comparable to that of human observers.