Information Technology Reference
In-Depth Information
uses a syntactic approach which employs functional blocks for the semantic analy-
sis and interpretation of the image [92]. In this case, the input image undergoes
pre-processing, which consists of:
•
filtering and pre-processing the input image;
•
approximating the shapes or locations of the analysed objects; and also
•
coding the image with terminal components of the introduced language.
The completion of these stages represents the data anew in the form of hierar-
chical structures of a semantic tree and produces subsequent steps of deriving this
representation from the initial symbol of the grammar [114], [117].
While pre-processing image data, a cognitive recognition system must in most
cases execute the segmentation, identify picture primitives and also determine the
relations between them. The classification proper consists in recognising whether
the given representation of input data belongs to the class of data generated by the
formal language defined by one of the grammars that can be introduced. Such
grammars can be classified as sequential, tree and graph grammars and recogni-
tion with their use takes place during a syntactic analysis conducted by the system
[36], [92], [97], [114].
Traditional data analysis processes based on cognitive resonance have been ex-
tended to include stages at which the system learns using the knowledge collected
in its knowledge bases and by analysing situations it does not understand. In this
case, if the system encounters a situation it does not understand, i.e. one unde-
fined in its knowledge base so far, it cannot correctly classify it and match the pat-
tern. Then the system enters a state of 'surprise' and incomprehension of the ana-
lysed data, which state it can, however, use to supplement the knowledge base
with new cases of pattern classification and data understanding, as a result of
which it becomes necessary to add new, undefined examples to the expert knowl-
edge base. The process by which the system learns is presented in Figure 2.11.
Fig. 2.11.
System training process
