Gaussian Proximal Hough Transformative Regularized Incremental Extreme Learning Machines for Palmprint Detection

Abstract

Palmprint detection is employed for identifying individuals based on the unique patterns present on the surface of their palms. Palmprint detection aims to be used in biometric authentication systems for security applications such as access control, forensic identification, and identity verification. Several research works have been developed for palmprint detection but have faced relative difficulty achieving higher accuracy. In this paper, Gaussian Proximal Hough Transformative Regularized Incremental Extreme Learning (GPHTRIEL) is developed with higher accuracy. First, palm images are collected from the dataset. Preprocessed images are provided as input to Outlier Regularized Incremental Extreme Learning Machines, consisting of three types of layers. The input layer receives preprocessed palm images. The first hidden layer performs image segmentation. Next, a set of geometric features is extracted in the second hidden layer and sent to the third hidden layer. Finally, feature matching is performed using the Sokal–Sneath similarity index function. With this, the outlier robust function correctly detects palmprints with a minimum error. An experiment is carried out with different factors. The analyzed research results indicate that the GPHTRIEL technique achieves improved performance in 6% accuracy, 11% sensitivity, and 10% specificity and minimizes 14% computation time compared to conventional methods.