Matched Spatial Filtering by Primitive Reference Patterns and Correlation Diagrams for Shape Invariant Pattern Recognition

In this paper, twelve linear primitive reference patterns are proposed for synthesizing a matched spatial filter, known as a shape invariant filter, which can recognize all patterns with various shapes. Furthermore, reprocessed data, given by correlation between the input object of the system and the reference patterns “correlation diagrams” are introduced for processing peak signals in a digital system. By correlating the diagram from the input object with the standard diagrams, the 26 of the alphabet letters were recognized in the experiment. Moreover, feasibility of the rotational invariance of the filter was suggested by the computer simulation.