Content-based representative frame extraction for digital video

We consider a content-based adaptive clustering algorithm (CBAC) to extract a fixed pre-determined number of representative frames to summarize a given digital video. In our algorithm, shot boundary detection is not needed. The inter-frame changes are compared globally for the extraction of representative frames. Small units of the video are dynamically clustered into two clusters based on the amount of change in content within the unit. One cluster, which has units of low content change, is designated for deletion and the other cluster, which has units of high content change, is for retention. The algorithm adaptively converges to the desired number of frames by deleting some redundant frames from each unit of the deletion cluster during every iteration.