Pattern Recognition by Dynamic
The Dynamic Pyramid Approach
Yung-Lung Ma and Jung-Shin Jour
The Institute of Electrical Engineering,
National Taiwan University,
Taipei, Taiwan 107, Republic of China
This paper presents a rapid recognition method. Every input pattern can be represented by one state vector that corresponds to some class. So when a pattern is to be recognized we can calculate from the pattern data the corresponding state vector and find its class.
To reduce the calculation time, the state vector can be calculated from part of the pattern. By the optimal stopping rule, the recognition process constitutes Dynamic Programming. That is to say, the specified part calculated in the latest stage is a subset of that in the current stage.
We present another new idea on coding of the state, we call it circular layer coding. Two types of circular layer codings are deduced; one is independent circular layer coding, and the other is Markov circular layer coding.
Experiments on five classes of hand tools verify that the proposed method is both efficient and versatile.