An Extended Gravitational Force-Based Texture Descriptor for Robust Palmprint and Fingerprint Recognition

Accurate extraction of texture patterns from fingerprints and palmprints is fundamental to reliable biometric identification. However, many existing texture descriptors exhibit limited robustness when subjected to noise, illumination variations, and complex ridge structures. This study proposes the Gravitational Extended Gravitational Force Technique (G-EGFT), a physics-inspired feature extraction approach that models long-range pixel interactions to enhance ridge representation. Following image preprocessing, the features generated by G-EGFT are integrated with Gray Level Co-occurrence Matrix (GLCM) statistical descriptors to construct a compact and discriminative feature vector. The proposed framework is evaluated on two benchmark biometric datasets, namely BMPD for palmprints and FVC2002 DB2 for fingerprints, using Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Decision Tree (DT) classifiers. Experimental results demonstrate that G-EGFT consistently outperforms the conventional Gravitational Force Technique (GFT), particularly in regions containing subtle ridge variations. Among the evaluated classifiers, the Decision Tree classifier with G-EGFT achieved an accuracy of 94% on the BMPD dataset and 93% on the FVC2002 DB2 dataset. the effectiveness of the extended five-by-five interaction field. The framework is primarily designed for clear and properly acquired palmprint and fingerprint images with sufficiently preserved ridge and texture structures, while substantial blur, rotation or scale variations, and severe erosion or line wear may affect descriptor reliability and recognition performance. Although the proposed method requires additional computational effort for larger images, it provides an interpretable and efficient alternative to deep learning-based approaches while maintaining robust classification performance, computational simplicity, and reliable biometric recognition under challenging imaging conditions. For the Decision Tree classifier, G-EGFT achieved 94% accuracy on BMPD, with MAE, MSE, and RMSE values of 0.45, 0.90, and 0.95, respectively, compared with 88% accuracy and 0.60, 1.10, and 1.05 for GFT. On FVC2002 DB2, G-EGFT achieved 93% accuracy, with MAE, MSE, and RMSE values of 0.40, 0.85, and 0.90, respectively, compared with 89% accuracy and 0.55, 1.05, and 1.02 for GFT. Precision and recall values for both datasets should also be reported using the experimentally obtained values to provide a complete quantitative evaluation.