INDEXING AND RETRIEVAL METHODS OF MOVING IMAGE DATABASE FOR TV-NEWS

The indexing and retrieval methods for TV news has recently become an important matter. The video image contains all sorts of objects and it is difficult to represent them by indexing terms used for literature searching. Any television networks in Japan are developing retrieval s ystems for video tapes of TV news, but these systems are lacking for t he theoretical consideration for indexing and retrieval methods. This study focuses on the indexing methods for moving graphical image and s uggests the new indexing language MPDL (the Moving Picture Description Language). MPDL has seven frames (subjects, time, locations, objects, characteristics, actions, sensibility) and allows indexers to assign one or more indexing terms for each frames. Indexing terms for 'action s' frame are consist of some verbs. The retrieval experiment using 110 video image, obtained from TV news, indexed by MPDL and other indexin g methods and ten search requests result that as concerns recall ratio and precision, MPDL is superior other methods. MPDL suggests many opp ortunities for improving the indexing and retrieval for the moving gra phical image.