Different from traditional information retrieval, which deals with text documents, multimedia retrieval covers a wide range of media types, such as audio, video, text and other corporated complex media. It can be categorized into audio retrieval, video retrieval, image retrieval and recent flash retrieval by searching object. Sometime we regard video retrieval as multimedia retrieval because video is the most complex media stream that human being has created, which contains audio stream, pictoral stream, text stream and object tag stream, for example, human tag sequence. The media understanding problem or semantic defintion puzzlement distingurishs multimedia retrieval from developed information retrieval. From the view of psychology, it is difficult or even a mission impossible to define a multimedia stream with clear semantic conceptions, such as key words, which can fulfil any query requirements. So the means of pattern recogintion technologies are employed to measure the similiarity among media documents and assume the possibility of topic correlationship.
Multimedia retrieval is not a simple research subject, but combined with many other research fields, such as computer vision, image processing, pattern recogition, artifical intelligence, and conception expression. Many experiment systems have been put into application since 1990s and proven to be helpful, though they might be very naive. To remeber, any improvement on multimedia retrieval will bring a new idea on many other research fields and vis-a-vis.