Abstract:
Living in the modern age of technology where there are high slogans of face analytics, data mining, social media analysis, Kinship verification finds its way from the automatic tagging of pictures,videos to the surveillance,security,human trafficking control and many more applicable areas. Automated Kinship Verification answers the question that the two individuals are blood relatives or not, by just analyzing the facial features.Kinship Verification using still images and videos is relatively challenging and an open research area in computer vision.A novel methodology based upon the generalization abilities of deep learning algorithms particularly Convolutional Neural Networks(CNN) is presented.Proposed methodology makes use of four different deep descriptors extracted from four distinct pretrained networks and fuse them in unique way with a modified approach of Sequential Forward Selection(SFS) with the goal of dimensionality reduction along with increase of performance.To prove the efficacy of proposed framework, experiments are performed on KinFaceW-I,KinFaceW-II and UvA-Nemo Smile databases demonstrating promising results in both case of images and videos.
