A pattern recognition algorithm based on the rapid transform

The author studies an application of the rapid transform for classification of entire two-dimensional shapes. The algorithm is based on the transform invariance property under cyclic shifts. A boundary will be represented by a sequence of segments defining the feature vector components of the shape. It is assumed that a sequence of segments corresponding to a rotated pattern, can be deduced from an original sequence of the same pattern by a number of cyclic shifts. Then, the transform of the vector of segments is independent of translation and rotation of the pattern in two dimensions. The classification algorithm will be tested on data derived from a library of two-dimensional aircraft patterns. The results show the algorithm efficiency in terms of computation time and classification