Cálculo da distância entre cristas em digitais neonatais para o aprimoramento de fatores de redimensionamento

Abstract

The biometric identification of newborns and infants is a technical challenge due to the size difference between infant fingerprints and traditional recognition systems, which are calibrated for adults. To overcome this problem, this work investigates and compares different ridge distance calculation methods to formulate image scaling factors. The methodology evaluates five experimental scenarios. The first two represent approaches from the literature, using a fixed scaling factor and factors separated by age groups. The other three scenarios propose continuous mathematical equations based on the exact age of the child, using spatial domain calculation, frequency domain calculation, and a reference method applied to a high-resolution database. The neonatal images were resized using bicubic interpolation and tested in their original and segmented versions. The performance of the methods was evaluated based on image quality using the NFIQ 2 metric and biometric recognition accuracy through TAR and FAR rates. The results show that segmented images present superior performance in all cases. Furthermore, the application of continuous equations, especially in the frequency and reference methods, increased the visual quality of the fingerprints compared to the static methods from the literature. However, for babies up to six months old, this increase in quality did not generate proportional gains during recognition, and the TAR reached very similar values across all tests. It is concluded that, even with similar recognition rates in the initial phases, continuous resizing individualized by age generates images with better overall quality and maintains the stability of the identification system throughout the child’s growth.