Designing Efficient Feature Space Reduction Schemes for Multi-Algorithmic Iris Recognition System based on Feature Level Fusion of Texture and Phase Features

Abstract

Iris recognition system has gained prominent focus because of its uniqueness, stability over time. But the recognition level of single biometric based recognition systems is greatly affected by environmental conditions, physiological deficiency. Multi-biometric systems diminish this problem with the fusion of features collected from various traits or samples of the same trait, a single trait by employing multiple algorithms or multiple instances. To gain the advantages of multi-biometric systems in iris recognition, a Multi-algorithmic iris recognition system has been proposed where Texture features from iris are extracted by using 2D-Log Gabor filter and Phase features are extracted by Haar Wavelet; And these features can be integrated at various levels like Decision, Rank, Score, feature, and pixel. Even though the feature level fusion contains rich information about biometric samples when compared to remaining fusion levels; it involves mapping complexity, high dimensional feature space. To gain advantage of feature level fusion in iris recognition and to overcome the problem of resulted high dimensional feature space, Genetic Algorithm (GA) based reduction scheme, Principal Component Analysis (PCA) reduction strategy and a hybrid reduction scheme which is a combination of PCA and GA have been applied to reduce the resulted feature space. The performance of these reduction strategies have evaluated on CASIA iris database, IIT Delhi iris database using Machine Learning approaches. The results have shown that the feature space has dramatically reduced while keeping recognition accuracy and also revealed that space and time requirements have significantly decreased after employing feature reduction schemes.