Conectividade recursiva de classificação de classes (R3C) aplicada à segmentação de imagem binária para melhoria da impressão digital infantil

Abstract

Infant fingerprint recognition still can’t match the recognition rates achieved for adult biometrics, particularly due to problems inherent to young children fingerprints such as size and skin mal-leability, leading to poor image quality and inconsistent ridge and valley structures. To address these issues, this study introduces Recursive Class Connectivity Classification (R3C), an image enhancement framework designed to iteratively refine binary segmentation that relies on previ-ous predictions to guide an expansion of segmentation coverage. The framework was evaluated on three infant fingerprint datasets using three different enhancement approaches as fingerprint ridge and valley classifiers. Experimental results show that R3C can significantly improve matching rates, particularly for newborns, in which identification performance was heightened by more than 40%. For older children, the proposed method achieved a 86% true identification rate (TAR), the highest of all cases tested, while also reducing the equal error rate (EER) by about 10%, a more modest gain that suggests further refinements are needed for broader compatibility.