Image basic features indexing techniques for video skimming

In this paper a comparison of the most widespread automatic indexing techniques, suitable in skimmed video generation, and their performances is presented. To evaluate the performances, using the low-level frame features, the signatures are computed, the shots are identified using neural network clustering techniques, in each shot the mean distance between contiguous frames is computed and the shot is resampled according to a related distance value to produce a skimmed video sequence. The most relevant feature proves to be the angular spectrum. Using this feature the mean value of the skimming factor is 2.6 in the used test set.