Aiming at the deficiency of identify semantic in traditional keyword tosearch characters the way the picture when you use the query statement to query the relationship between people. The paper designs and achieves an image search architecture based on character relationships. We offer an approach to extract the latent relationships between persons from news corpus. Domain ontology library creates pictures of people and in the ontology contains a total of 174 relatives, friends, colleagues and other kinds of character relationship with the hierarchy properties. An ontology-driven query interface for the sentence which provides a query-oriented relationship of one sentence, first We extract the critical component of the query statement for the relative persons which comsolidation rules based on dependency grammar tree from the dependency syntax tree, then generate the SPARQL sentence with the help of triple supplement algorithm, and use SPARQL sentence to query the image ontology library. The semantic retrieval was realized. Finally, the experiment results were given to verify the feasibility and effectiveness.