Automatisiertes Verfahren zur zeitlichen Segmentierung eines Videos in Szenen unter Beruecksichtigung verschiedener Typen von Uebergaengen zwischen Bildfolgen

WO 2008155094 A2 UPAB: 20090116 NOVELTY - The method involves recognizing image sequence boundaries as scene boundaries. Similarity between two image sequences e.g. shots, is determined by selection and comparison of representative images from the sequences, where the images lies outside of transitions. The sequences are selected such that the selected images are spaced from each other. A film grammar and probability based allowance of all types of transitions are provided with respect to separation and/or connectivity action on the sequences surrounding the transitions. USE - Method for temporal segmentation of videos e.g. fictional, narrative videos e.g. motion picture, sitcom and animated film, and non-fictional videos such as news, documentary film, educational film, sports program and music program, into scenes. ADVANTAGE - The image sequence boundaries are recognized as the scene boundaries, thus facilitating proper combination of the image sequences. The film grammar and the probability based allowance of all types of transitions are provided with respect to separation and/or connectivity action on the sequences surrounding the transitions, thus optimally reducing failure rate in recognition of the image sequence boundaries, and hence improving accuracy and robustness of automatic segmentation of the videos into scenes. The method considers differently weighted applications of effects of the transitions in collation of the image sequences into the scenes