Cascading adaptive binary image feature maps with vision transformer for iris spoof detection
Deepika Sharma, Arvind Selwal · Applied Soft Computing · 2025
Iris presentation attack detection (PAD) module is an essential component of iris recognition system used to address the susceptibility against various spoof attacks. Although convolutional neural network (CNN)-based iris PAD methods have exhibited remarkable performance but suffers from limited generalization capabilities to unseen attacks. Furthermore, recent deep learning-assisted PAD approaches uphold parameter dependency and inductive bias problems for extracting texture information from images. Thereby pertaining to these issues, the current expansion of the Vision Transformer (ViT) architecture has surrogated the CNN model for image classification tasks. Hence, in this work we are using ViT for the first time in iris spoof detection problem along with newly explored handcrafted local image features . We propose a novel hybrid approach for iris spoof detection by coalescing unique local image features with efficient vision transformer. Three novel and discriminatory image feature maps (Central Local Adaptive Binary Patterns (CLABP), Left Local Adaptive Binary Patterns (LLABP) and Right Local Adaptive Binary Patterns (RLABP) are extracted from iris images and given as input to ViT model. Afterward, ViT extract multiple global context information from iris feature maps that is aimed at various head scales. Finally, classification task to discriminate live and fake iris images is performed by ViT. The experimental results demonstrate superior performance of our approach on benchmark datasets such as Notre Dame, IIITD-WVU and NDCLD15 with an average classification error rate (ACER) of 1.94 %, 1.52 %, and 1.97 % respectively.