A novel sub-windowing local binary pattern approach for dorsal finger creases based biometric classification system

Biometric authentication systems have been widely deployed in various applications, including security systems, bank transactions and authentication on smart electronic devices. Obtaining the salient and distinctive features is very important for achieving high accuracy in biometric authentication systems. Local binary pattern (LBP) variants are the best-performing local descriptors and are popular due to computational simplicity and flexibility. However, most of the existing LBP variants consider a 3 × 3 window with one specific central pixel for all neighborhoods, which affects the sensitivity to non-monotonic intensity changes and reduces the robustness of the feature description. Thus, a new variant of LBP called TD-LBP is introduced, which is based on the four T-shape sub-windows and two diagonal (D) regions. Inspired by the sub-windowing approach to capture the microstructure information of the image, TD-LBP first divides the 3 × 3-pixel window into four sub-regions of T-shape structure and then takes two diagonal regions to extract more texture information. Three different classifiers, artificial neural network (ANN), support vector machine (SVM), and k-nearest neighbor (KNN) are employed to evaluate the effectiveness of the proposed approach for dorsal finger crease biometric system. Experiments conducted on the self-collected dorsal finger crease dataset demonstrate the prominent performance and suitability of the proposed TD-LBP for a newly explored finger crease biometric identifier. The proposed approach was able to achieve 96.67 %, 89.26 %, and 82.22 % classification accuracies for ANN, SVM, and KNN classifiers, respectively. Moreover, we clearly validate the viability of the proposed TD-LBP descriptor for the dorsal finger crease biometric trait by comparing the results with state-of-the-art biometric system based LBP descriptors. The significance of the TD-LBP method is demonstrated with improved verification and identification results through receiver operating characteristic (ROC) and cumulative match characteristic (CMC) curves respectively.