| 摘要 |
PROBLEM TO BE SOLVED: To solve such a problem that a conventional highlight creation of moving pictures on the basis of machine learning requires to create learning data files which have learnt training moving pictures extracted from a sufficient number of sample moving pictures by experts in advance, however various patterns are present even in the same program genre in the broadcast contents of the television programs, and it is therefore difficult to prepare learning data covering whole patterns of broadcast contents of television programs in advance. SOLUTION: When a user finds out a highlight interval not detected yet during viewing a television program moving picture whose highlight detection processing is finished and the user selects the highlight interval from a UI, the selected highlight interval is added to the learning data as new training data. It is possible to detect a highlight scene with a similar characteristic vector by using the updated learning data for a succeeding detection processing. Further, in updating the learning data, a limit is placed on a rate of data which the user can add in order to avoid considerably losing the performance at the initial setting. COPYRIGHT: (C)2008,JPO&INPIT
|
