Toward Large-Scale Palmprint Image Analysis by a Rich Orientation Code
Dandan Fan,Xu Liang,Wei Jia,David Zhang
DOI: https://doi.org/10.1109/tsmc.2024.3377535
2024-01-01
IEEE Transactions on Systems Man and Cybernetics Systems
Abstract:Palmprint recognition has gained considerable attention in recent years, accompanied by significant progress. However, large-scale palmprint recognition, which holds great potential for extensive civilian applications like university access control, remains underexplored. In this work, we propose the CUHK-T dataset, the largest palmprint dataset to date, containing over 10k individuals ’ palmprints. As the data volume expands, the heightened complexity, such as similar principal lines from different palms, necessitates a recognition method capable of extracting more discriminative features. Motivated by the rich palm lines distributed in palmprint, including not only nonintersecting, but also intersecting line segments, we model intersecting lines, investigate their properties, and propose a novel and explainable palmprint recognition method. The model treats the nonintersecting line segment as a special case, allowing for the extraction of orientation information from both types of line segments. In addition to the rich orientation information of the intersecting lines, the extracted feature accounts for the relative width of these lines. These advancements enable the extracted rich orientation code to be more discriminative and representative for palmprint. We then present a bitwise similarity measurement for efficiently and effectively comparing two rich orientation codes. Our extensive experiments and evaluations with popular palmprint recognition algorithms demonstrate the effectiveness and superior performance of our method on the large-scale dataset. These results also serve as a foundational baseline, facilitating the advancement of further research in the domain of large-scale palmprint recognition.