Enhanced Contactless Fingerprint Segmentation Using a Transformer‐Based Residual <scp>UNet</scp>

ABSTRACT Biometric recognition from contactless finger images or fingerphotos has attracted attention due to improved data safety and hygiene as compared to contact‐based counterparts. However, accurate recognition depends on precise segmentation of the fingertip region, as poor segmentation can introduce spurious features and degrade matching performance. This study proposes TransResUNet , a deep learning model that integrates Transformer blocks and residual connections into the UNet architecture to enhance segmentation accuracy under challenging conditions such as background clutter, illumination changes, scale variations, and resolution changes. Transformer blocks capture global feature relationships and gradually enhance feature interactions, while residual connections strengthen local feature learning and preserve detail. Experimental results on ISPFDv1 and ISPFDv2 datasets demonstrate superior performance of the method with intersection‐over‐union (IoU) scores of 0.975 and 0.987, and accuracies of 0.998 and 0.997, respectively. Cross‐dataset evaluation further proves the ability of the model to generalize under different acquisition conditions. The proposed method also shows robustness to noise and blur, offering a promising approach for accurate contactless fingerprint segmentation under diverse conditions, thereby contributing to the development of more secure and reliable biometric recognition systems.