Explore the Hybrid Approach of Texture Analysis with Deep Supervised Learning Classifiers on Palmprint Recognition System

Abstract

Emerging cyber-environments emphasize the criticality of remaining aligned with authentic security systems to safeguard the exchange and storage of unique private and public confidential digital information. The Biometric Authentication and Identification System (BAIS) is a dependable application designed to mitigate the risks of unauthorized access to valuable electronic information in the digital realm. Among various BAIS implementations, the palmprint recognition system (PRS) stands out as one of the most widely adopted systems, attributed to its ease of access, high durability, distinctiveness, and accuracy in identifying legitimate individuals while maintaining a cost-effective approach. This proposed study explores a hybrid methodology in texture feature extraction techniques alongside supervised deep learning classifiers. This can be achieved by independently and collectively assessing the fractal dimension and lacunarity texture values utilizing the Box-Counting and Gliding-Box algorithms, then integrating these with two Supervised Deep Learning algorithms known as the Deep Backpropagation Learning classifier (DBL_Net) and the Convolutional Neural Network (CNN_Net) classifier, to achieve an advanced BAIS with enhanced accuracy. The 2DPROI image datasets are sourced from a multispectral 2D-palmprint image database at Hong Kong Polytechnic University in Hong Kong. The entire task is programmed using MATLAB R2020a Update 4 (9. 8. 0. 1417392) 64-bit (win64). Ultimately, the benchmarks of this proposed system are tabulated and analyzed to substantiate the assertion of improved recognition and identification accuracy through the hybrid feature extraction process, yielding an Authentication Recognition Rate (ARR) of 99.50% and 99.75% when employing the DBL_Net and CNN_Net classifiers, respectively.