A New Method for the Feature Selection in Pattern Recognition Algorithms

Most of the pattern recognition algorithms extract certain features of the input signal and compare them with some pre-saved patterns. Important criteria for such systems are the defining features which construct the feature vector and assign them proper weighting. Different methods for the optimization of the pattern recognition algorithms have been proposed in the literature, but most of them tend to fail to optimize the system when the number of the features increases beyond a certain value. The present paper presents a new method for the selection and weighting procedure of the features. The proposed method is applied to a voice recognition system and the simulation results are reported. The results showed about 28% improvement in the system performance in comparison with the not-optimized system.