Pollination based optimization for feature reduction at feature level fusion of speech & signature biometrics

A scheme for the feature level fusion of two behavioral biometrics speech and signature using fusion method weighted sum is proposed. Feature reduction is performed using modified feature selection algorithm based on Pollination based optimization which has never been applied to the problem earlier. The modified algorithm is applied to the fusion method to search the feature space for optimal and best feature subset. This optimization is first time used to extract features of speech and signature biometric modalities after fusing. Multimodal offline database of text independent speech and signature has been collected from 40 users. Experimental results have proven that the systems works well sum fusion rule along with modified PBO algorithm has been used. The accuracy of the system is 100%. The FAR and FRR of the system is zero. Also, system´s robustness for noisy samples has been analyzed which shows accuracy of 70.0%. The overall time taken by the system is from 4.70 to 5.72 seconds.