Pattern Recognition (PR) is a general term which may mean any of the following with subtle conceptual differences.
Given a pattern (e.g., image objects such as a human face, printed or written text, a natural scene, etc.), a name (perceptual category) is generated as the output.
The input patterns of different classified/categorized into a set of classes/categories.
The associator learns to establish the connection between two patterns
(concepts).
For example, ``fire'' and ``hot'', ``banana'' and ``yellow'', ``peace''
and ``war'', ``
'' and ``Einstein''. Minor error can be tolerated.
Similar to the content addressable nature of the brain. An incomplete pattern can be recognized as a complete pattern most similar to the input one.
Examples:
Different methodologies can be used for PR, such as statistical methods and neural network methods. In general all PR algorithms can be categorized into two types: supervised learning (learning with teacher) and unsupervised learning (learning without teacher).
Supervised methods assume the availability of some a priori knowledge, and they are carried out in two stages:
Certain parameters of the learning process are obtained based on some training data, a set of pairs composed of input and the corresponding desired output, typically provided by experts in the field.
The input data of unknown patterns are classified or recognized.
Unsupervised methods do not assume any a priori knowledge. The classification or recognition is based totally and directly on the input data.