Hidden Biometric Identification/Authentication based on Phalanx Selection from Hand X-Ray Images with Safety considerations
Résumé
The paper presents a new approach called "Phalanx-Code" based on hand X-Ray image. The main motivation is to prove the effectiveness of hidden biometrics, based on the mentioned modality applied with few selected phalanges, in the context of human identification/authentication processes. Indeed, the proposed approach aims at preventing forgeries when used in highly secured systems such as Vital Records, Airport Control, Forensic Medicine, etc. During the acquisition process, we use an X-Ray machine with controlled voltage and current exposure. Each image will follow an automatic low level processing in order to extract some reference points. These points will guide the extraction of Hand phalanges during the segmentation phase. Hence, the hand will be modeled by features extracted from the phalange structures using Fourier Descriptors and Eigen Vectors. To evaluate the performance and the robustness of our algorithms, we have constructed a database consisting of 88 Right hand X-Ray images from 42 non-pathological adult individuals. Surprisingly, our experiments show that humans are fully distinguishable within the study of few selected phalanges per individual. With two phalanges only, a hand-code can be modeled with a vector in R10 with an Equal Error Rate (EER) does not exceeding 0.3984%. To achieve 0% of EER, four phalanges are enough to model a hand with a vector in R16. Despite the use of X-Ray devices, our technique could be considered safe since the exposure rate of radiations remains below the standard thresholds.