Stochastic image segmentation using spatial-temporal context

A stochastic syntactic method for the analysis of time-varying images is presented. The time-varying phenomenon is analyzed using a language translation schema. Stochastic tree grammars are used to describe the patterns, and the stochastic translation is used to characterize the evolution process of the time-varying image sequence. It is shown that the necessary conditions for the existence of a matched filter are preserved under the stochastic translation. Therefore, a spatial filter can be designed for the patterns at each stage of the image sequence, and a temporal filter can be designed for the trajectories formed by the pattern primitives. These filters can iteratively extract the contextual information in a pattern. Emphasis is on the relationships between the filters