Optimization techniques of iris-signature for better performance
Muthana H. Hamd · International Journal of Advances in Intelligent Informatics · 2019
A biometric system-based iris trait is developed to identify 30 and 10 subjects, taken from CASIA-v1 and real-iris datasets respectively. The proposed unimodal system uses Fourier Descriptors (FDs) to extract the iris features and represent them as an iris-signature graph after applying Dougman’s rule for segmentation. The four classifier results: Back Propagation (BP), Radial Basis Function (RBF), Probabilistic, and Euclidian Distance (ED) are compared for quality purposes. The input machine vector of 150 values can be optimized to include only high-frequency coefficients of the iris-signature, so two optimization techniques are applied and compared. The first one, selects sequentially new feature values with different lengths from the enrichment graph region that has rapid frequency changes, while the second technique is based on the standard deviation formula for choosing the high variance coefficients as new feature vectors. The accurate recognition results with contrast to the vector-lengths of the second technique has achieved better performance, and for some classifiers, the accuracy rate is maintained with lowest connection cost and run-time. After optimization, the Probabilistic and BP have satisfied better accuracy rate with lowest vector length Among all other classifiers.