A Fast IRIS Recognition Technique Based on Dimensionality Optimization and Multidomain Feature Normalization

Biometric Solutions are the need of hour for future security based technique. IRIS is considered as one of the better biometric models for representing a human Identity due to invariability with age. Recognition efficiency in IRIS Recognition technique depends upon number of features. Larger Features affects the space complexity and computational complexity of the Technique. Hence in this work we propose a unique scalable IRIS recognition technique to define a feature vector by using the descriptor of different IRIS identities like Shape, Color, Texture Frequency and Phase and proposed a mechanism for reducing the feature space by first normalizing the feature values and then subsequently reducing it by log likelihood dimensionality reduction technique aligned with PCA method. Further the reduced dimensions are re-optimized with Koghnen's Self Organizing Maps to represent the feature vectors in two dimensional feature space with fixed range in each dimension for ease of storage. Experiments are conducted over Noisy MMU Iris Database and Phoenix Iris Database to analyze the performance. Results show that the features extracted and optimized by the proposed technique gives an average accuracy of 98.6%.