This thesis introduces a method for visualizing the spatial composition of a movie. It transforms time into space by classifying frames from a movie sequence into clusters of spatial similarity.
The approach used is derived from stochastic process theory. A training set of the data is taken from the movie. Principal component analysis is used to define the space of the frame clusters. Every frame from the movie is then clustered with respect to its distance from the designated frames in the designated space.
Movies are made up of many frames. The huge size of the data set motivates us to approach the problem in this manner. It also includes the added benefit of compression. In using this approach, we will actually be defining a new element of a movie. Elements of this nature are used to reassemble the movie. The result is a Salient Movie.