| 摘要 |
The automated recognition of scene boundaries supports the user on browsing a video. Conventional methods do not make satisfactory use of the information in a video which is available for scene segmentation. According to the inventive method, the determination of the similarity between two image sequences is carried out by selection and comparison of representative images from each image sequence, which are as far apart as possible from each other in time but which do not lie in gradual transitions (DISSOLVE). Additionally or alternatively a film grammar and probability based allowance for all types of transitions (CUT, DISSOLVE) is provided, regarding the separating and/or connectivity action on the image sequences around the transitions and differently weighted application (attrCUT, attrDIS) of the effects (sttat(k)) of the different types f transitions in the collation of image sequences k (A, B, C, D, E, F, G) into scenes. Thus, according to the invention, all types of transitions with a corresponding weighting on determining the image sequence boundaries are included as scene boundaries and the separating and connecting effect thereof are considered and this insight used on a probability basis in the decision in existing scene recognition methods. |
