Digital Image Interpretation (Visual Languages Implementation)
(A.Järg)

Digital image recognition and interpretation are the last steps of the bottom-up image processing approach. Usually when information about an object or region class is available, some recognition method is used. It is very complicated to determine which method is the best in specific case.

Syntactic pattern recognition is one domain of object recognition methods. Syntactic object description should be used whenever feature description is not able to represent complexity of the described object or when the object can be represented as a hierarchical structure consisting of simple parts. The elementary properties of the syntactically described objects are called primitives. After each primitive has been assigned a symbol, relations between primitives in the objects are described, and a relational structure results. The design of description primitives and their relation is not algorithmic. Here are some useful criteria to follow for developing primitive types:

• amount of primitive types should be as small as possible,
• primitives should be easily segmentable from the image,
• the chosen primitives must be able to form an appropriate object representation.

Each primitive have two properties called attributes. The first attribute shows the size of the primitive and the second attribute shows the orientation of the primitive. For describing relations between primitives we are using grammar rules. The grammar G is a mathematical model of a generator of syntactically correct words.

where is set of non-terminal symbols denoting connections between terminals, is set of terminal symbols denoting primitives, S is the grammar start symbol and P is a nonempty finite subset of the set where .

To get more precise description of relations between primitives we have to add to each rule two attributes. The first attribute shows connection orientational condition and the second attribute shows connection quantity condition. This way we will get so called attributed grammar rules. Based on attributed grammar rules we will build attributed binary tree where each node denotes connections between primitives and leaves den otes primitives. Attributed binary search tree is useful recognition method if we can divide input data easily to primitives according to conditions described before.