Dorsal Finger crease classification system using local binary pattern and its variants-A new finger biometric modality

Interest in hand-based biometrics has achieved strong attention in the last decades due to the characteristics of uniqueness, reliability, user-friendliness, higher acceptability, and low data acquisition cost, which makes it more attractive for secure commercial applications. The dorsal finger crease has recently been studied as a biometric identifier and shows promising results for individual identification due to its stable structure and accurate data. This research provides a local binary pattern (LBP) based biometric classification method using dorsal finger crease features. LBP use the local texture patterns to characterizes the image texture. It has high discriminative power, invariance to monotonic illumination, and computation simplicity which makes it more suitable for the dorsal finger crease recognition system. To this end, the performance of local binary pattern and its various descriptors including uniform LBP, average LBP, median LBP, local directional pattern (LDP), and locally oriented optimal pattern (LOOP) are investigated to find out how these descriptors can be efficiently employed for dorsal finger crease classification system. A self-collected, real-time database of dorsal finger creases is used for the experiments, and Support Vector Machine (SVM) has been used in the classification stage to assess how well various LBPs are performing and to verify the necessity of our dataset. Experimental results demonstrate that the average and median LBPs consistently perform better and obtain a better classification performance over other LBP variants.