The paper aims to design and develop a text matching picture system to realize on-line automatic picture matching facing text data. Based on the description relationship between pictures and text, a “picture-tag” bipartite network was built. And then based on the “picture-tag” bipartite network, the relevance scores between pictures and tags was computed by random walk with restart. The keywords of the input text were extracted by TextRank model and the set composed of key words was queried with the key words as tags. Based on the off-line calculation result, the relevance scores between text and pictures were obtained by integrating the relevance scores between pictures and tags. And the top k most relevant pictures were returned according to the relevance scores. The experimental results show that the MAP score of the top 5 five returned pictures can reach 0.839, which can accurately return the expected pictures. The system can accurately match pictures for input text.