Considering the picture has the futures that a strong interpretation of events and convenient disseminating, this paper studies extraction of data from a large number of news web pages, and organizational structure chart presented to the users. It achieves dynamic pages based on page template extraction and analysis, processing converted to the corresponding sets of datastructure. Based on the news cosine correlation graph data sets from different sites are differentiated, and in accordance with the appropriate standards for data collection to score sorted. This system is based on hadoop distributed platform, considering the large number of users and imgsets. This paper will describe the design and implementation of our system in detail, and report the results of running the system on Baidu news image column.