Annually, 20 million children fail to receive the essential annual vaccinations they require, resulting in a tragic toll of 1.5 million children deaths caused by vaccine-preventable diseases [?]. Additionally, 3.1 million children lose their lives globally due to malnutrition [?]. A major factor contributing to this devastating reality is insufficient monitoring and the absence of official identification documents, significantly hampering accurate vaccination tracking and nutritional supplement distribution, especially in developing countries [1]. Biometric recognition for young children can be a viable solution, not only in facilitating the monitoring of vaccination and nutritional supplement distribution but also in ensuring border security, preventing newborn swapping, combating child trafficking, and providing access to healthcare, civil IDs, and so on. However, despite its paramount importance, it remains an ongoing unsolved challenge as the existing biometric technology has proven ineffective for children.

Among various biometric modalities, fingerprint recognition emerges as a promising choice for young children in terms of universality, acceptability, persistence over time from birth, and interoperability across acquisition method [1,4,5,6,7]. However, the fingerprint systems designed for adults fail to cater to young children, who possess finer ridge spacing and softer skin compared to adults; using a contact base scanner can cause their fingerprint to flatten against the surface and merge the adjacent ridges, leading to the loss of distinct features. Thus, capturing the fine details of young children’s fingerprints necessitates specially designed, child-friendly, non-contact high-resolution scanners [5,6]. Figure 1 shows the difference in the quality of fingerprints captured with low- and high-resolution scanners. Additionally, the rapid growth and finger size change in children complicate fingerprint matching over time (aging effect), calling for a system that accommodates these age-related changes. The scarcity of longitudinal fingerprint datasets for children (0-15 yrs.) limits the feasibility of studies on child fingerprint recognition. Consequently, critical questions remain unresolved, such as the earliest age for biometric use, the enrollment frequency needed for reliable recognition, and methods to accommodate age changes. Most of the studies in the literature have been conducted using child datasets captured with low resolution (500 ppi, which is standard for adults) with limited success. A few recent studies using high-resolution fingerprint scanners have indicated potential for young children, but these studies are limited in dataset size and diversity, necessitating further validation. Additionally, the literature lacks assessments on the effectiveness of high-resolution contactless fingerprint scanners for children over one year old using longitudinal datasets.

Our research addresses these gaps by analyzing data from 254 children (0-15 years) over a year. Our contributions include:

In recent years, several studies have focused on the viability of fingerprint recognition for children. Anil K.Jain et al. [4] conducted a year-long study with 309 Indian children aged 0-5 years, using both a standard 500 ppi scanner and a custom contact base 1270 ppi scanner. Their results indicated that reliable enrollment and recognition were possible at 6 months with 1200 ppi and at 12 months with 500 ppi. Galbally et al. [2] analyzed a dataset of approximately 265k different fingers from age groups ranging from 0 to 25 and 65 to 98 years using a 500 dpi scanner and reported that the aging effect profoundly impacts recognition in the 0-4 year age group, with the 5-12 year age group also presenting significant recognition challenges. Precioszzi’s [3] research on a dataset of 16,865 identities (0-20 years, images captured with 500 ppi) revealed notably low accuracy for infants (TAR 1.25 to 34.88 @ 0.1 FAR) and children under the age of 5 years (TAR = 61.88 to 78.27 @0.1 FAR) despite introducing a growth factor to counteract the aging effect. Engelsma et al. [1] collected fingerprints from 315 infants aged 0-3 months in India using a 1900 ppi scanner and developed an infant matching algorithm. Their results indicate that infants can be enrolled at 2 months and accurately recognized 3 months later with a TAR 95.2% @ FAR 0.1%. Kalisky et al. [5] introduced a 3400 ppi contactless fingerprint scanner, suggesting that newborns can be enrolled with a single finger as early as 4 days old and recognized within 15 to 30 days with a TAR 96% @ FAR 0.1%. Their research also shows that recognition can achieve a TAR 100% @ FAR 0.1% with a 6-month age gap when enrolled with ten fingers. Table 1 summarizes these prior works.

Among the studies mentioned above, Engelsma et al. [1] and Kalisky et al. [5] demonstrated the most promising recognition possibility for infants using high-resolution fingerprint scanners. However, the scope of their trials was limited in terms of dataset diversity, dataset size, and the duration of the study. Additionally, Engelsma et al. [1] reported that the contact base high-resolution scanner performed better than the contactless scanner for infants, whereas Kalisky et al. [5] showed that the contactless high-resolution scanner achieved superior performance than Engelsma et al. [1]. Therefore, further evaluation of their experimental finding is necessary. Furthermore, a literature gap exists in evaluating high-resolution contactless fingerprint scanners for children over one year using a longitudinal dataset. To bridge these research gaps, we collected data from 254 children (0 to 15 years old) over one one-year in the United States using a 3000 ppi contactless scanner (introduced by Synolo Biometrics [?]), commercial contact base scanners (500 and 1000 ppi) and performed comparative analysis to assess the effectiveness of high-resolution contactless fingerprint scanner in this age group. Our work can be seen as a contribution to research work conducted by Kalisky et al. [5], aimed at validating their proposed high-resolution scanner and experimental findings.

In this section, we discuss the preparation of our experimental datasets and the protocol employed for conducting the experiments. Additionally, we provide a detailed explanation of the outcomes derived from our assessment.

Fingerprint recognition for young children can aid in vital areas like vaccination tracking, border security, and combating child trafficking. However, current fingerprint technology falls short for children, primarily due to the variation in finger sizes and the lack of technology tailored to handle these differences. High-resolution scanners show promise for infants, yet their effectiveness across a broader age range in children remains unexplored. Our study investigates high-resolution contactless fingerprint scanners for children aged 0-15 years, focusing on the effects of fingerprint aging. We collected longitudinal data from 254 children using 500 ppi, 1000 ppi, and 3000 ppi scanners over one year. The results show that infants enrolled at 5 days old were accurately recognized after 2 months with a 100% TAR @ 0.1% FAR using 3000 ppi. Conversely, 1000 ppi scanners failed to capture fine details in young children’s fingerprints, indicating the necessity for more specialized technology. We also found that 500 ppi was adequate for children older than 4 years when the time gap between enrollment and authentication was one year. The 3000 ppi contactless scanner demonstrated superior performance in young children and outperformed 1000 ppi contact-based scanners on a large scale for both verification and identification. Yet, for older children, its performance was comparable to 500 ppi. One limitation of our study is the relatively small dataset of young children, suggesting the need for further research to validate our findings for this demographic. Future work will involve collecting more longitudinal data from additional subjects and examining factors like age at enrollment, demography, the time gap between enrollment and authentication, and image quality, using the Linear Mixed Effect model to understand their impact on match scores. We also plan to analyze the data for various time gaps and break down the 4-15 years age group into individual age groups to better understand the time gap’s impact within each specific age group.