A Model for Representing Significant Segments of a Video Image Based on Locally Positional Coding on a Structural Basis