Research On Facial Kinship Verification Methods For Service Robot

With the rapid development of image recognition technology,facial kinship verification has gradually attracted more and more researchers'attention due to its important application value.As an emerging biometric technology,it has many potential applications,including missing children search,family album management,and social media analysis.Different from the traditional face recognition problem,facialkinship verification faces greater challenges,mainly due to the difficulty of extracting similarity features and the small scale of existing data sets.From the perspective of features and data,this paper proposes multiple methods to improve the performance of facial kinship verification,and designs a visual interface to provide support for the application of actual service robots.The main work results of the thesis are summarized as follows:(1)To address the difficult problem of feature extraction,a local feature attention model is proposed,which guides the network to focus on the feature area through the pre-covered data enhancement method,which improves the discriminative power of the feature;(2)To address theproblem of the small size of the data set,a negative example sampling method based on reinforcement learning is proposed.Through the design of a verification network and a sample screening network,the negative samples with strong discriminative ability are effectively selected,which improves the discriminative ability of features;(3)To address theproblem of the imbalance of positive and negative samples,a method for mining difficult samples based on meta-learning is proposed.By automatically learning a discriminative representation model from the data,the imbalance of samples is effectively resolved from another perspective.The methods proposed in the thesis are compared with the existing mainstream methods.Our methods achieve higher verification rate than most existing mainstream facial kinship verification methods,which proves the effectiveness of the proposed methods.We also designed an interactive facial kinship verification system to integrate our proposed methods,which provides technical support for practical facial kinship verification applications.