Performance Comparison between Zernike Moment Invariant and Fractal Codes features in the Application of Zip Code Recognition using RBF Neural Network

This paper presents a system for off-line recognition of segmented (isolated) handwritten Farsi/Arabic characters and numerals. We have used Zernike Moment Invariant and Fractal Codes as two different kinds of features in this system. Also Radial Basis Function (RBF) neural network that is used for many engineering problems and pattern recognition tasks has been employed in this work. Simulation results on our database, which were gathered from various people with different ages and different educational backgrounds, indicate that the ZMI and fractal codes are suitable features for segmented handwritten Farsi/Arabic characters and numerals recognition and the best performances of this system are 91.5% and 92.8% for characters and numerals recognition respectively