• Classification and Clustering of Printed Mathematical Symbols with Improved Backpropagation and
  Self-Organizing Map
Siti Mariyam Hj. Shamsuddin, Md. Nasir Sulaiman and Maslina Darus
Abstract. This paper proposes a derivation of an improved error signal for hidden layer in the backpropagation model, and its experimentation evaluation of utilizing various moments order as pattern features in recognition of printed mathematical symbols in the classification phase. The moments that have been used are geometric moment invariants in which they have been used as feature extraction for images with various orientations and scaling. In this study, we find that the recognition and the convergence rates are better using an improved backpropagation compared to standard backpropagation. In addition, we cluster these invariants on a visual map using Self-Organizing Map (SOM) whereby mathematical symbols with similar shape belong to the same cluster.