Finger vein and iris based multi biometric authentication system using pattern net neural network

Abstract

Biometric based authentication is a most needed activity in a corporate and business world. Genuineness, accuracy and reliability are the most common characteristics of any authentication system. This requires any multimodal unique biometric traits combined with better fusion strategy. Biometric authentication refers to automated methods used to identify a person by the features such as face, iris, vein, finger print, palm print etc. The biometric features finger vein and iris are used in this research to design a reliable multi-biometric authentication system. A highly sophisticated scheme is used in this research to pre-process the finger vein and iris images. In this research a novelC2 code derived using orientation andmagnitude information extracted from finger vein and iris images to improve the authenticating system is proposed. The C2 code eliminates feature selection operator reducing the process complexity as it combines the orientation and magnitude information from finger vein and iris image inputs. This methodology is suitable to the entire environment where biometric authentication system is required due to its reduced data handling complexity. The number of feature vector is proportional to the precision of pattern recognition system however it is associated with large memory allocations as well as reduced speed due to computational burden. The extracted feature vectors are stored in cloud based data storage where the data are retrieved during the verification time. The cloud environment makes the option for large storage of data set and globally it can be accessed anywhere in the world through internet. The reduced feature vector is used for identification using Neural Net. With the actual data sample used for training, the resulting neural network R2 value is very low of 0.745, and the best performance validation was 0.12656 at 5th epoch. In order to improve the performance of the pattern net ANN, the feature data set is analyzed for noise, and the data points contributing the noise are removed an