TEXTURE MODELING AND SIMULATION FOR SYNTHETIC PALM VEIN IMAGE GENERATION SYSTEM

Abstract

Unavailability of large-scale palm vein databases due to their intrusiveness have posed challenges in exploring this technology for large-scale applications. Hence, this research modelled and generated synthetic palm vein images from only a couple of initial samples using statistical features. Variations were introduced to the three optimized statistical features (S5; the original images were employed as training images and the best variation in the first experiment  as training images, S4; the best variation in the first experiment  as training images while the original images were used as testing images, S3; mean vectors, covariance matrices and correlation coefficient, S2; mean vectors and covariance matrices, S1; mean vectors, Non-Synthetic; acquired image) which were used to generate synthetic palm vein images employing statistical and Genetic Algorithm (GA) approaches and were evaluated based on Equal Error Rate (EER), Average Recognition Accuracy (ARA) and Average Recognition Time (ART). The results obtained from the experiment showed that EERs were 0.22, 0.51, 0.58 and 4.36 for S3, S2, S1 and NS respectively. S3 had superior ARA (99.83%) compared with S2 (99.77 %), S1 (99.70 %) and NS (98.33 %). The ARTs obtained were 84.97s, 75.55s, 84.04s and 681.74s for S1, S2, S3 and NS respectively with S2 (75.55s) having significantly least value. Furthermore, EER, ARA and ART for S4 were 0.43, 99.00%, and 12.13s, respectively while the corresponding values for S5 were 1.43, 97.50%, and 680.13s, respectively. The research outcome justifies the extraction of mean vectors, covariance matrices and correlation coefficient.