Resumen:
This paper introduces a feature extraction scheme for offline
handwritten math symbol recognition. It is a hybrid model that involves the basic ideas of the wavelet and zoning techniques so as to define the feature vectors with both statistical and geometrical properties of the symbols, with the aim of overcoming some limitations of the individual algorithms used. Experiments over a medium-sized database of isolated math symbols investigate the performance of the new hybrid technique in comparison to other algorithms. The results show that the new model performs significantly better than the rest of algorithms tested, independently of the symbol category.