Method for summarizing unknown content of video

Claim 1

A computerized method for summarizing unknown content of a video, comprising the steps of selecting one or more low-level features of the video; partitioning the video into segments, the segments being sequences of frames, according to the low-level features and grouping the segments into a plurality of disjoint clusters, each cluster containing similar segments, wherein similarity is based on a distance metric and wherein similar segments are identified by using time series data of the low-level features in a self-correlated cross-distance matrix and by finding minimum cost paths between pairs of points in the cross-distance matrix, said paths satisfying path cost criteria and direction constraints; labeling the plurality of clusters according to the low-level features by choosing labels out of sets of labels describing the low-level features; finding high-level patterns among the labels by looking for labels correlated in time, the correlation either being simultaneous or time shifted; and extracting frames from the video according to the high-level patterns to form a content-adaptive summary of the unknown content of the video.