We model the video-scene as a contiguous segment of visual data that is chromatically coherent and also possesses similar lighting conditions. A video-scene is said to occur when there is a change in the long-term chromaticity and lighting properties in the video. This stems from the film-making constraints discussed earlier.

We use ideas of recall and coherence and introduce the idea of shot-lets.

Recall: The recall between two shots depends upon: (a) the distance between the two shot key-frames (b) the duration of each shot and (c) the difference in time between the two shots.

Shot-let: This is a fraction of a shot, and the key-frame associated with the shot-let is the same as that of the original shot.

Coherence: This is the recall between pairs of shots in the attention span and the rest of the memory.

We determine the coherence of the shot-lets in the attention span with the rest of the memory at every shot-let boundary. Video scene changes are deemed to occur at local coherence minima locations.