A Novel Method to Recognize Shapes based on ANN and Allotting Numbers to Shapes

This paper presents a new method to correctly identify objects based on their 2D shapes. Hitherto, the best method to correctly determine the identity of a particular 2D shape has been using artificial neural networks (ANNs). But ANNs suffer from the drawback that in order to correctly identify a particular 2D shape as a particular object, the ANN has to be extensively trained with images of the object in all possible shapes and sizes and also at various orientations with respect to the observer. It would be so much better if a method can be devised which allots a number to a particular 2D shape of the object as viewed by an observer so that, this way, the shapes of the object as seen from different angles and having various sizes can be represented by a range of consecutive numbers. Now, if the number allotted to a random shape by the method is found to be falling within that range, the shape can be correctly identified as the object. The present paper illustrates such a method to correctly identify shapes. It is shown that the training requirement for the method is minimal. Also, the efficacy of the method is shown by way of various examples.