Метод порівняння відбитків пальців, заснований на моделі попереднього вирівнювання

The subject of this article is the development and analysis of a new method for comparing fingerprints that uses the geometric Euclidean characteristics of minutiae for biometric identification. The research focuses on minutiae—special points of interruption or bifurcation of papillary lines—as key biometric features, contrasting them with traditional global and local descriptors such as SIFT, HOG, or LPQ. The purpose of the article is to develop a robust and efficient fingerprint comparison method that uses Euclidean geometric characteristics of minutiae and pre-alignment to improve the accuracy of biometric identification without relying on machine learning. The research task was performed in three stages: first, studying the theoretical provisions of minutia-based descriptors and their invariance to affine transformations (shift, rotation, scaling); second, development of a model using a shift vector and distance functions for matching minutiae; third, experimental verification of the model, determination of optimal parameters, and evaluation on a standard data set. Methods include theoretical analysis and experimental evaluation. The theoretical basis establishes the stability of alignment to shifts. The descriptor is formed through minutiae coordinates, distance functions, and alignment optimization. An image processing algorithm with filtering and minutiae analysis is used to extract features. The results are achieved through experimental verification on the FVC2000 (DB1_B) dataset and demonstrate high performance, as evaluated by classification metrics and execution time. Conclusions indicate the theoretical and practical achievements of the research. The model demonstrates theoretical and practical resistance to Euclidean shifts, with advantages for processing prints from different scanners. Experiments confirm the effectiveness of shift detection, achieving a high Van Rijseberg score (0.735), although dependence on positive matches requires additional filtering of false positives. The method is workable and can be applied in practice.