Kinship Verification From Facial Images Based On Deep Learning

Facial images convey many important human characteristics.Resemblance often appears between two people who are biologically related with overlapping genes.Based on the research on psychology and biology,this paper propose verify the kinship from facial images with deep learning.Kinship verification from facial images is an emerging and challenging problem.In this paper,we propose to learn a relational feature for kinship verification from facial images,which represents a particular relationship between a pair of individuals.To establish the relational feature,we design a new deep convolutional neural network-Auto Encoder(CNN-AE)model.The relational features are the activations taken from the hidden layer of the deep neural network.The CNN-AE neural network is a supervised deep learning algorithm that applies back propagation.Along with the feature extraction hierarchy,CNN-AE gradually forms compact features and fuses them to a relational feature.The proposed relational feature using deep learning is proved effective in kinship verification.The main works of the thesis are as follows:1.A new relational feature that genuinely represents a relationship between individuals is defined.The experiment shows that any state-of-the-art classifiers can be learned based on the relational feature for kinship verification.2.A new deep neural network is designed for feature extraction.The CNN-AE net keeps reducing the neuron number while the hierarchy goes higher.A effective and compact relational feature is the activations of the higher hidden layers of the CNN-AE.3.Good verification accuracy is achieved with the algorithm proposed in this paper for the father-son,mother-son,father-daughter,mother-daughter kinship verification.4.The proposed method is also applied in the self-kinship verification.Since the aging problem is caused by age,another deep neural network is used for age classification and age prediction.