M²BRTPC: A Novel Modified Multimodal Biometric Recognition for M²BRTPC: A Novel Modified Multimodal Biometric Recognition for Toddlers and Pre-School Children Approach Toddlers and Pre-School Children Approach

A biometric system based on the characteristics of adults recently achieved an outstanding result. Over the last few decades, many applications have been developed for adults, such as ﬁ ngerprint, face, iris, and hand-vein. Infant iden-ti ﬁ cation suffers from many problems and does not enroll all possible ages. Identifying children using one or any biometric features is the ﬁ rst step in ensuring that they are recognized before the law and that their rights are guaranteed. From another perspective, this prevents baby swaps in hospitals, Identifying Missing Children, and Civil ID Programs. This paper introduces an approach for Multimodal Biometric Recognition for Toddlers and PreSchool Children. The focus is on ﬁ nding a conclusion regarding identifying a child of different ages using as few biometric features as possible, for instance, ﬁ ngerprints, faces, iris, etc. The proposed approach uses a dataset containing iris and ﬁ ngerprint modalities of more than 100 children (aged 18 months to 4 years) and creates a new database. There are two phases in the proposed approach: Enrollment and Testing. During the enrollment phase, there are four steps, after which the extracting features are stored in the new database. There are ﬁ ve stages in the testing module; after creating the template, it compares it with the created database and ﬁ nds the matching child. A Fingerprint Extraction Tool is also proposed for extracting features from ﬁ ngerprint images. The experimental results show that the proposed approach enhances performance by 14.75 e 18.75%.


Introduction
T he defining characteristics of a person can be divided into two categories: psychologically and physically. Psychology focuses not only on the mind and behavior of a person but also on their mental, emotional, and social development. Erik Erikson was one of the first scholars to describe adolescent identity problems. Anna Freud worked with many children, unlocking different original forms of their psychological lives. For example, she was the first to describe the phenomenon of 'identification with the aggressor' (Ego, 1936). Identifying an individual's physical characteristics can be achieved by capturing fingerprints, palms, hands, retinas, iris, and facial features and then processing them using computers. Biometric systems based on adults' characteristics are known to have performed admirably.
Although biometrics work brilliantly to confirm and identify adults, validating and identifying infants, newborns, and young children with high accuracy is a challenge. However, due to the lack of security operations in the hospitals and the intensive care units (ICU), it has become necessary to identify newborns, infants, and young children Patil and colleagues (Patil et al., 2019). More than 678 million children worldwide under 5 years old in 2021, and an average of 345 000 newborns are added to the population daily. Most of those children are in developing countries where there is not any kind of identification. Fig. 1 shows the total population under 5 years old, based on historical estimates and projections through 150 years, starting with 1950, based on The United Nations Mediterranean Fertility Scenario (https://ourworldindata.org/ grapher/under-5-population?country ¼~OWID_ WRL.).
Children need to be identified in high demand. Children's biometrics are necessary for many emerging applications, such as keeping track of vaccination schedules.~25 million children under 5 do not receive essential vaccinations yearly, and nearly 6.6 million may die from vaccine-preventable diseases. Roughly half of the children under 5 are chronically ill (dwarfism), and almost 14% suffer from acute malnutrition in developed countries. Tracking childhood vaccinations; in developed countries, more than 5 million children die annually of vaccine-preventable diseases Yoon and Jain (Yoon and Jain, 2015;Committing to child survival, 2013). Concerning identifying missing children, more than 800 000 children are lost in the United States annually -a child almost every 40 s and preventing infants' babies from swapping in hospitals. In developing countries, changing newborns after birth is problematic due to insufficient maternity ward facilities and hospital overcrowding Yoon and Jain, Best-Rowden and Jain (Yoon and Jain, 2015;Best-Rowden and Jain, 2018). Biometrics helped health workers in developing countries find patient records. According to country rules, the United Nations World Food Program (WFP) has helped deliver food to needy people Basak and colleagues (Basak et al., 2017).
The most commonly used biometric traits for recognition systems, such as fingerprint Rugani (Rugani et al., 2022), iris Corby and colleagues (Corby et al., 2006), face Siddiqui and colleagues (Siddiqui et al., 2018), palmprint (Lemes and Bellon, 2011;Weingaertner et al., 2011), and footprint (Weingaertner et al., 2011;Kotzerke et al., 2013), are often focused on adults. Aside from the fact that infants' and young children's biometrics have a low rate of accuracy, another thing is that their physical features change as they grow. Consequently, it is unpredictable and subject to rapid change. In addition, some physical biometrics is expected to shift in size Patil and colleagues, Yoon and Jain (Patil et al., 2019;Yoon and Jain, 2015). There are several physical biometrics types, shown in Table 1, and their differences.
A fingerprint is defined as a mark left on a surface by rubbing the edges of a human finger Hemanth and Balas (2018). The ridgeline structures in newborns are three to five times smaller than in adults. In newborns, the skin has ridges that can easily deform when in contact Castellanos (1953). The skin condition of oily or wet fingers of younger children and hence the image quality may be influenced by 'the habit of sucking their fingers' Jain and colleagues (Jain et al., 2017).
The thin annular structure iris in the eye is responsible for controlling the diameter and size of the person. Attention issues and the inability to remain still are necessary to take the image many times. Despite having their heads fixed or closed, they keep moving their eyes and refuse to capture their data Morgan and colleagues, Galbally and colleagues (Morgan et al., 2018;Galbally et al., 2018).
Face recognition is more common; however, it faces many challenges: a change in the reaction, lighting, or environment, uncooperative subjects because of constant movement, until the exact moment when the baby is relaxed and calm with eyes open but not crying or moving a lot. In some situations, the capture procedure took nearly 45 min to create a comfortable environment for the newborn Bharadwaj and colleagues, Tiwari and Singh (Bharadwaj et al., 2010;Tiwari and Singh, 2012).
A palm print is an acquired image of the palm area (Tiwari and Singh, 2012;Zhang, 2004). This technique has many limitations; for example, the newborn must hold still during the capture time, which is 2 min. Besides, the results lack variance, and pressure is required on the hand of the newborn to keep it from moving and uncooperative child. Nevertheless, most studies estimated that 1500 PPI (pixel per inch) would be enough to capture print Lemes andcolleagues, Weingaertner and colleagues (Lemes et al., 2011;Weingaertner and Bellon, 2011).
A footprint is a mark left by a foot on a surface. Biometric recognition technology is the subject of many opinions. Children are identified using a footprint recognition system in many studies. However, two people are needed to gain a print from a child, one to take a picture, and one to hold and calm the infant or the child. It is difficult to match infant footprints due to low image capture quality, very small and fragile ridges, and taking an image at the wrong time of the day. The skin begins to dry and crack a few days after birth, and the ridge pattern that can help identify becomes obscur-edKotzerke and colleagues (Kotzerke et al., 2019).
All biometric features have their own challenges; however, some are in common. For example, infants and newborns are uncooperative subjects. They may refuse to open their hands or eyes or move their legs. The rapid growth, which constantly changes all physical attributes, is another additional challenge. One also needs to consider the protectiveness of the parents and the newborn's fragility. Therefore, ergonomics is very important and should be taken into consideration.The proliferation of biometric features in person identification today has created a promising field that could influence the identification of young children in the future. The development of machine learning techniques, particularly deep learning, opens up a great deal of potential for automating the extraction and classification of features from images. Furthermore, multimodal fusion enhanced the robustness of the recognition system, enabling it to make a recognition decision, although one or more biometric features are not successful. The process of feature extraction from images essentially focuses on reducing the necessary resources to delineate a broad dataset.
The main contributions of this paper are.
(1) Development of a novel approach, Multimodal Biometric Recognition for Toddlers and Pre-School Children (M 2 BRTPC), for child identification using minimal biometric features. (2) Proposal of a new Fingerprint Extraction Tool (FPET) to enhance the accuracy of the fingerprint modality.
(3) Creation of a new database, the New CMBD, for archiving outputs from the applied software for both fingerprint and iris modalities in a binary format. (4) Execution of comprehensive experiments that validate the superiority of the proposed system in iris and fingerprint recognition compared with existing research. (5) Insightful analysis of the trade-off between iris and fingerprint biometrics, offering a comprehensive understanding of their benefits and limitations.
The rest of this paper is discussed in the following order: The related works are discussed in the next section. Then, Section 3 explains the proposed approach (M 2 BRTPC). Then, Section 4 presents the High (Removal of the shoes and the socks and allow the technician to hold the foot and press) Unknown Minor experimental protocols that have been used. Then, Section 5 shows the results of the proposed approach compared with other related works. Finally, in the end, Section 6 concludes the paper.

Related works
Research into child fingerprint recognition dates back to 1899, with the publication of Galton's pioneering paper Rugani and colloeagues (Rugani et al., 2022). Galton collected fingerprints from a single child using an ink-on-paper method from birth to age 4.5 years. The study spanned four sessions per year, during which fingerprints were gathered from 309 children (aged between 0 and 5 years). Galton ultimately concluded that 'the print of a child at the age of 2.5 years would serve to identify him ever after.
In 2004, Netherlands Organisation for Applied Scientific Research (TNO; Dutch: Nederlandse Organisatie Voor Toegepast Natuurwetenschappelijk Onderzoek) conducted a study to evaluate the viability of using the biometric features for Dutch travel documents. The study concluded that 'it was not possible to obtain clear fingerprints from children under 4 years old' due to the subtlety of the fingers' ridge pattern Jain and colleagues (Jain et al., 2016). In 2007, a project, 'Biometrics Data Experimented in Visas (BIODEV II)', was started by eight European member states. According to the fingerprints of more than 300 children captured in Damascus (Syria) and Ulaanbaatar (Mongolia), they concluded that it is difficult to obtain fingerprints for children under the age of 12 years old Rahmun and colleagues (Rahmun and Bausinger, 2010). In 2008, Weingaetar and colleagues, Basak and colleagues (Basak et al., 2017;Weingaertner et al., 2011) compared footprint performance to palm prints for 106 newborns, and they noticed that the quality of palm prints was better than footprints.
In 2013, the European Commission's Joint Research Center published a technical statement on child fingerprinting. The study was based on the fingerprints of more than 2600 children aged 0e12 years collected using a 500-dpi fingerprint reader while passport processing by the Portuguese government. They concluded that identifying fingerprints of children under the age of 6 is difficult ('Fingerprint Recognition for Children, 2013). In 2016, fingerprint data of 206 children aged 0e5 years were collected at four stages over a year. Their work aims to investigate the persistence of fingerprint recognition. However, they only scored each subject's thumb samples instead of all fingers Jain and colleagues (Jain et al., 2017). In 2017, research was conducted using Children Multimodal Biometric Database (CMBD) with three biometric modalities irises, fingerprints, and face Basak and colleagues (Basak et al., 2017). As shown in Table 2, the database contains 106 samples of iris, fingerprint, and face from 106 subjects. The experimental results were (i) in separate left and right iris samples. The result was near 99%. (ii) They found perfect 100% accuracy on the database on a fusion. (iii) In separate fingers, the results between different fingers show (45.28e71.71%). The fusion of ten fingers shows 97%, and in combined match scores of 10 fingerprints, two irises, and a face, the average genuine accept rate (GAR) at 0.1% false accept rate (FAR) was 100%.
In 2018, a study was conducted to identify children between the ages of 4 and 12 using iris recognition. Researchers organized three data collection events, each~6 months apart. The commercial product, Verieye, was employed for data analysis, focusing on left-eye images. If an image was compared with itself, the score was 1557 (100%). However, if the images differed, the score dropped to 0 (0%). When comparing images from the first session (collection 1) to themselves, the average score was 790.3 (50.8%). The average score for comparisons between collection 1 and images from the second session (collection 2) was 380.6 (24.4%), while the average for comparisons between collection 1 and images from the third session (collection 3) was 352.5 (22.6%) Morgan and colleagues (Morgan et al., 2018).
In 2020, an approach using Fingerprint Biometrics from Infants to Adults was introduced in Javier and colleagues (Javier et al., 2020) to identify a person from a very young age. They run experiments on a dataset of more than 200 000 fingerprints from about  (Yaseen et al., 2021), research was held and proposed an approach that uses three biometric modalities. They concluded that it is easy to obtain ear biometrics from birth. Also, it is possible to improve a hardware device to get fingerprints with more features and details from infants so that the age of the participants does not exceed 6 weeks. Iris biometrics can also be used to successfully match individuals from 6 weeks ago. Table 3 concludes the comparison between the related works.
A thorough review of extant literature shows that most research studies have primarily relied on a single biometric trait for child identification, often utilizing freely available software. Notably, several studies have used low-to medium-quality hardware for image capture in conjunction with freely accessible feature extraction software. This paper seeks to fill the research gap by presenting a novel approach that merges multiple biometric traits and introduces a new FPET alongside a generated database, thereby promising to enhance child identification with improved results.

The proposed approach (M 2 BRTPC)
This study used a Novel modified Multimodal Biometric Recognition for Toddlers and PreSchool Children (M 2 BRTPC) approach. The M 2 BRTPC uses the most effective biometric features recognition fingerprint, iris, and fusion between them that helps the identification of newborns. The M 2 BRTPC approach schematic diagram is depicted in Fig. 2.
The approach consists of two phases; the first one is the Enrollment Module, which consists of four procedures; image acquisition, image preprocessing, feature extraction system, and template generated and used to create the database. The testing module is the second phase in the proposed approach and consists of five procedures; image acquisition, image preprocessing, Feature Extraction System, Template Generate, and Template Matching System.
The proposed approach concentrates recognition on fingerprint and iris because face recognition shows the lowest GAR among the three modalities. In addition, young children's faces may not be robust for recognition, and their features change rapidly. The M 2 BRTPC approach creates a new database from the CMBD dataset Basak and colleagues (Basak et al., 2017).

Image acquisition
The CMBD dataset contains five samples for each child, comprising left fingerprint impressions, right fingerprint impressions, two thumb impressions, and five samples for both irises. Table 4 summarizes the statistics of the images in the CMBD dataset for both fingerprint and iris modalities that have been utilized. The number of children with data in both modalities across both sessions is 108. In the enrollment module, the images of fingerprints and irises from session one are used to create the database (New CMBD). The 108 fingerprint and iris images from session two serve as probes for testing the newly created database.

Image preprocessing
A preprocessing is applied to the input images to extract the fingerprint and iris patterns. Some techniques include contrast and illumination correction, noise filtering, and sharpening. In addition, the edge detection technique enhances and clarifies the image's ridges. WSQviewer (http:// www.cognaxon.com/index.php?page ¼ wsqview.) application converts photos to a format appropriate to current international standards for minutiae extraction and comparison, the FBI standard for fingerprint images, and all ISO-compliant fingerprint technologies accept it.

Feature extraction system
The feature extraction system consists of two parts: the first is for iris extraction, and the second is for fingerprint extraction. The VeriEye SDK (Verieye. http, 2017) is employed with some modifications for feature extraction of the selected iris images to enhance the precision of feature extraction and matching. Concurrently, the newly proposed FPET extracts features from fingerprint images. Upon completion of the training module phase, it is concluded that when matching children's fingerprints, it can sometimes be advantageous to trick the extractor with a lower DPI (Dots per Inch) setting to account for the small ridges on children's fingers. In the testing module, a visitor class is created for the test subject, then the finger position is extracted from the image, and the image with the finger type is assigned to the visitor. Subsequently, features are extracted from the image and a match is sought within the database. Fig. 3 presents the pseudocode for the steps involved in the FPET tool within the Testing Module. The FPET utilizes the Automated Fingerprint Identification System (AFIS), which was developed by the Federal Bureau of Investigation (FBI) in 1974. The AFIS facilitates large-scale biometric identity solutions, thereby making the software more versatile for various applications and fields, such as hospitals, schools, and vaccination centers. This is achieved by creating a database from the gallery images (in the Enrollment Module) and subsequently searching for a match in the generated database (in the Testing Module).

Template generate
The selected images from the gallery session are uploaded into the software at this stage. Then, following the extraction of features and details from the images, a template is generated for each image. This template is stored encrypted and linked to the respective child in the database during the Enrollment Module phase. In the Testing Module phase, a template is generated for each selected image from the probe session or for each uploaded image, and a search is conducted for a match within the generated database. Fig. 4 displays a sample of fingerprint and iris images for child-7001 before and after encryption.

Database
The newly generated database (New CMBD) encompasses the images for fingerprints and irises processed in the previous steps. These images are converted and encrypted into binary data. The New CMBD contains two databases: the first for  experiments 1 and 3 and the second for experiment 2. In the first database, one image out of five is selected as a gallery for each child's finger and iris. In contrast, in the second database, all five out of five images are selected as a gallery for both finger and iris. Fig. 5 presents a pseudocode for the creation of these databases. These databases are used for various experiments, which will be elaborated upon in the subsequent section.

Template matching system
In this phase, the template generated from the probe image in the second session is compared with the database developed using gallery images from the CMBD dataset of the first session. The comparison utilizes the settings for the image's DPI (Dots per Inch) and the system's configured threshold. The system automatically determines the  optimal threshold based on the experiment subjects' results. If the score surpasses the threshold, the system displays the corresponding child along with the matching ratio; otherwise, it discards the match.

Experimental works
The generated database (New CMBD) is used in the proposed M 2 BRTPC approach. It contains fingerprint and iris images for 108 subjects common across two sessions. The proposed approach (M 2 BRTPC) is tested using 108 subjects representing 92.5% of the images of the CMBD dataset; the rest of the images (7.5%) and other datasets are used for the training.
Photos from the first session form the gallery, while images from the second session serve as test images. Right and left irises are dealt with as separate modalities, and the following experiments are performed uniquely. Like fingerprints, each finger is treated as a different modality, and experiments are performed on every finger separately. In Fig. 6, finger names are shown concerning human body parts.
In the subsequent subsections, we will discuss three distinct experiments. The first and second experiments are unimodal, utilizing only one finger or one iris, whereas the third experiment is multimodal, employing a fusion of fingers, irises, or a combination of both. In unimodal experiments, right and left irises are treated separately. Similarly, each finger is considered a separate fingerprinting modality, with experiments being conducted independently on each. Multimodal experiments are performed to amalgamate information from multiple modalities to yield superior results. Table 5 provides a summary of the experimental protocol implemented in this study.

Experiment 1
In this experiment, we utilized the first database, which comprises the first sample out of five collected during session one (consisting of fingerprints and iris samples), and designated it as the gallery. All five samples from session two were treated as queries during the testing phase. This is delineated in Table 6.

Experiment 2
In this experiment, the second database is utilized. All five samples from session one are treated as gallery images for both fingerprints and iris. Conversely, all five samples from session two are considered queries in the testing phase. This setup is depicted in Table 7. When compared with Experiment 2, Experiment 1 has some flaws. The performance may falter if the selected gallery image is of poor quality. To determine whether there is any improvement in the GAR values, all five samples from session one are utilized as a gallery in Experiment 2.

Experiment 3
In this experiment, an attempt was made to fuse multiple biometric traits to determine whether there was any improvement in children's identification compared with the results from a single modality in the first experiment. The same database was used. This protocol includes three modalities: the first combines the left and right index fingers, the second combines all ten fingers, and the third combines all ten fingers with the two irises. The protocol is detailed in Table 8.

Results and analysis
The results obtained from the experiments will be discussed and analyzed in the following subsections and compared with the related works.

The iris experiments
From the iris experiments.
(1) The left iris and right iris had GAR above 98%.
This demonstrates that the iris is an extremely  powerful identification tool that works almost flawlessly with databases and its persistence has been exceptionally high over the years.
(2) Blur and illumination are significant parameters that substantially affect the performance of children's recognition using iris images. (3) When the left iris is fused with the right iris, any parameter negatively impacting the recognition performance is diminished, and a perfect 100% GAR is achieved at a 0.1% False Accept Rate (FAR). This indicates no instances in the database where a false match is returned when the left and right irises are merged. (4) Neither the left nor right iris exhibits clear dominance over the other. Therefore, iris matching is equivalent in both eyes on an individual level. However, merging both for matching delivers superior overall accuracy. This answers the crucial question: the combination of both irises guarantees the identification of a child over a long span of years.

The fingerprint experiments
From the fingerprint experiments using the new (FPET).
(1) In fingerprint verification, the GAR ranges from 60.19 to 82.41% for different fingers at a 0.1% FAR. The little fingers, with 60.19% accuracy for both left and right, have the least accuracy.
(2) We observe that the average accuracy of the index and middle fingers is significantly higher than that of the ring and little fingers. The little finger records the lowest due to its smaller size. (3) The thumb demonstrates accuracy similar to the index and middle fingers but superior to the little finger. (4) Both hands show nearly identical accuracy; this depends on the finger position, skin condition, and the hardware capturing device. (5) In Experiment 2, improved matching performance for fingerprint verification for all fingers was noticed. For example, it has a GAR ranging from 83.33% to 96.3% for different fingers at the same FAR of 0.1%. Table 9 presents Experiments 1 and 2's results for all fingers and irises.

Comparison between iris and fingerprint
(1) Compared with the iris, fingerprints are significantly less accurate and less persistent. As expected, the iris clearly outperforms fingerprints in matching.
(2) Iris accuracy exhibits a lower standard deviation compared with fingerprints. This suggests that the iris is more robust in terms of matching over many years. (3) Even though the iris yields superior results, it has more limitations compared with fingerprints. Obtaining accurate iris data requires an adult's approval and significant cooperation from children, which can lead to poor image quality due to factors like closed eyes, looking away from the center, and poor illumination. (4) While acquiring iris data demands expensive equipment, it delivers superior results. Conversely, capturing fingerprints is relatively   easier with much cheaper equipment, but the trade-off is lower accuracy.

Multimodal fusion experiments
Table 10 presents the conclusions obtained by applying the fusion with the modalities in the explained protocol of GAR at 0.1% FAR.

Left Index and Right Index fingers
Larger fingers (index, middle, and thumb) performed significantly better than the others. This indicates that combining data from these fingers could enhance matching performance. This paper demonstrates a significant improvement in matching accuracy when fusing data from the Left Index and Right Index fingers, achieving an accuracy of 96.3% at 0.1% FAR. This suggests that a child can be almost certainly identified using just these two fingers.

Ten fingers
In this database, when all ten fingers are merged together, the accuracy was 100% GAR at a 0.1% FAR. The result is more robust when using ten fingers instead of just two.

Ten fingerprints, two irises
Combining all ten fingers with both irises results in a perfect match of 100% GAR at a 0.1% FAR. Additionally, this indicates that fingerprint matching can succeed when iris matching fails. It is extremely rare for independently collected iris and fingerprint modalities to fail to identify the subject when all fingers are merged with both irises.

Comparing with related works
The primary comparison study will be conducted between M 2 BRTPC and Multimodal Biometric Recognition for Toddlers and Preschool Children Basak and colleagues (Basak et al., 2017). Since both have the same experimental protocol, it will be straightforward to compare the results in each experiment for different modalities. All experiments for the approach yield superior results for fingerprint recognition, while iris recognition produces nearly identical outcomes. The proposed approach presents a modified system with better results in fingerprint and fusion modalities. Based on Experiment 1, at the same false accept rate of 0.1% FAR, Fig. 7 compares fingerprint and iris recognition using the proposed approach (M 2 BRTPC) with the aforementioned research. Their average result was 55.9%, while M 2 BRTPC's average was 74.6%, demonstrating an increase in accuracy of 18.7%. Fig. 8 compares fingerprint and iris recognition between M 2 BRTPC and the other research in experiment 2 at 0.1% FAR. According to them, the average result was 76.4%, while the average result for M 2 BRTPC was 91.1%, with better accuracy of 14.7%. Fig. 9 compares recognition using fingerprint and iris between the Multimodal Biometric Recognition Technique for Preschool Children (M 2 BRTPC) approach and other research in Experiment 3 (Fusion Modalities) under different parameters at a 0.1% False Accept Rate (FAR). In this approach, the facial modality was excluded due to its low accuracy performance and limited persistence over time. Once again, the results of the proposed approach outperformed those of the other research studies. In the first fusion protocol, which combines the left index finger with the right index finger, the GAR was 96.3% at 0.1% FAR, whereas in the other research, the GAR was 87.71% at the same FAR. The next fusion protocol, which combines all ten fingers, achieved a GAR of 100% at 0.1% FAR, compared with a GAR of 89.96% in the other research at the same FAR. Lastly, the final fusion protocol, which combines the iris data with all ten fingerprints, achieved a GAR of 100% at 0.1% FAR. This superior performance can primarily be attributed to two factors: the use of the FPET, which significantly enhanced the performance of the fingerprint matching modality and the high accuracy of iris matching. These results also indicate that even if iris matching fails to identify a subject, fingerprint results may be able to do so.
Multi-model biometrics Recognition using different transforms is a research focusing on giving a secure system using (fingerprint þ Iris recognition) Mahilraj and colleagues (Dr Mahilraj, 2021). They used three transforms for the feature extraction (Haar transform, Fourier Transform, and Laplace Transform) from the adults' images. They found that the Haar transform, which gives better results for the multi-model biometrics recognition, produced 98.5% in the aspects of accuracy.
Another comparison study will be with a Novel Multi Fuzzy Technique for Face Recognition of Newborns Arul Raj and Balasubramanian (Arul Raj and Dr Balasubramanian, 2021). They present Genetic Convolutional Neuro Multi-Fuzzy (GCNMF). It was based on a fuzzy system using CNN and NF with potential high points. A similar dataset (CMBD) was used to evaluate and analyze the proposed approach. The study concluded that GCNMF improves classification efficiency and discrimination accuracy by 99.05 percent. scope is restricted by the age range of the children involved, which limits the application to toddlers and preschoolers aged between 18 months and four years. Moreover, the database created for this research, although comprehensive, might still lack the robustness of larger, more diverse databases. Additionally, the proposed FPET is novel and lacks extensive validation in varied practical scenarios. Furthermore, while the study does offer improvements over existing methods, it does not claim absolute accuracy or precision, acknowledging the inherent challenges in biometric identification. Finally, the study assumes the practicality of multimodal biometric traits in realworld situations without accounting for potential data collection and processing challenges. The applicability of the framework presented in this paper is multi-faceted and broadly beneficial.
(1) Child Identification and Safety: This system could significantly enhance child identification processes and contribute to child safety. It could be employed in hospitals to prevent baby swaps, help in finding missing children, and bolster civil identification programs.  (2) Long-term Consistency: The use of iris and fingerprint data provides consistent identification over a child's growth period. These biometric traits demonstrate significant persistence and resistance to aging effects, making the framework applicable for long-term child identification. (3) Legal Recognition: The framework facilitates early legal recognition of children. By identifying children using distinct biometric features, we ensure they are recognized by law and their rights are protected from an early age. (4) Education and Welfare: In education and welfare systems, accurate identification of children is essential. This approach could be used to track school attendance, administer social services, and monitor the well-being of children in institutional care. (5) Research and Development: The new database created for this study, featuring binary data of children's iris and fingerprint modalities, presents a valuable resource for further research and development in child biometric identification systems. (6) Crime Prevention: By establishing a child's identity beyond doubt, the system could aid in crime prevention, such as child trafficking, identity theft, and abduction.

Conclusions
There is a growing need for applications and problems where different biometrics could be used for identification; the proposed M 2 BRTPC is interested in increasing the comprehensiveness of the biometrics system for different ages. In general, current biometric systems do not register and document young children, Toddlers, and preschool children, while the need for biometric systems for young children is enormous. M 2 BRTPC presents a modified approach to multimodal biometrics with the database of young children. It generates a new database using iris and fingerprint images from the CMBD dataset (ignoring face modality based on their recommendations). The results of experiment 1 for the modified approach are better than the compared research by 18.75%; meanwhile, the results of experiment 2 for the modified approach are better than by 14.75%. Combining two fingers can be sufficient to identify a child, while incorporating all ten fingers perfectly identifies a child. Also, the iris biometric modality can present the highest real acceptance rate, especially while combining the right with the left iris. However, collecting data from young children demands special care and patience.
Many challenges are related to biometrics for young children. The main one is data acquisition; a good hardware scanner with a better pixel-per-inch (PPI) can help capture more detailed images. For example, the acquisition of iris images depends on blur and lighting. Also, it is an essential modality in recognizing children's performance. The next one is image clarification; many parameters affect image quality; a preprocessing algorithm is used to remove unnecessary data and clarify the image and can help improve the feature extraction from each image.

Author contributions statement
Dr. Mostafa, Organized and helped in the analysis of the results of manuscript. Dr. Labib, Get the Data set and help in modifying the algorithm, analyzing the simulation results, and writing the manuscript. Dr. Mahmoud Badawy, Suggested this search point and help in analyzing the simulation results. Eng. M. Behzad, Write the simulation algorithm, and prepare the draft of the manuscript.