Many mathematical methods have been proposed for solving image analysis problems. They can be grouped into two major approaches, decision-theoretic or statistical approach and structural or syntactic approach. From the point of view of pattern representation and description, we can discuss image analysis in terms of a unified semantic-syntactic approach. This paper will describe the basic formulation of this approach and discuss a major result in syntax-semantics tradeoff. Recognition methods based on this proposed approach and their implementations will also be discussed.