| Kinship is one of the most important social relationships in human society.In recent years,image based character kinship analysis and recognition is becoming an important research direction,attracting the attention of many researchers.However,due to the task itself,the data set and other factors,there are still many shortcomings in the existing work.There are still many difficulties and challenges in the identification of the kinship relationship of the person in the image.This paper mainly focuses on group photographs covering kinship,and divide them into pseudo-family photos with partial kinship and family photos with all kinship.The relatives of them are excavated and identified respectively,and the corresponding solutions are put forward in a targeted way.The main contents are as follows:Firstly,the related works of kinship verification and kinship identification tasks are reviewed and combed,and the representative methods are introduced in detail,including design ideas and key algorithms.Secondly,an algorithm for discriminating family members based on graph structure is proposed.For group photos with partial kinship,the pseudo-family members are eliminated using face and photo context information.After face detection,a full connection graph is created based on the number of people.The pair-wise kinship is verified,using as the adjacency attribute of the edge.The pruning operation is performed on the graph structure and the connection state between the vertices is traversed to identify the family members.In addition,the photo context is added to secondary constraint of the adjacency property.The accuracy of the proposed scheme is verified which has been tested on the image datasets collected in this chapter.Finally,an automatic family tree construction algorithm based on constraint rules is introducted in detail.The character roles and kinship types are identified in a single family photo.The typical frameworks of core family and extended family are designed,calculating the person's age,gender and genetic similarity.Under the kinship constraint rules,the family standard frame position layer,the family member role range and the specific family role of person in the photo are sequentially confirmed,and all members are described in the family tree.The experimental results on the family dataset in this chapter demonstrate the effectiveness of the method. |
