Page 132 - 2024-Vol20-Issue2
P. 132
128 | Fathi & Aziz
II. LITERATURE REVIEW
Fig. 1. Classification of Biometric traits Databases are the basis for hand biometrics research. There
are several hand image datasets created by researchers, which
thickness, area of the fingertips, circle radii on the fingers and can be obtained from a scanner [9], a digital camera [10], or a
palms, and angles between fingers [3]. Images of the hand in USB camera [11]. Most hand biometric systems work with
general, or their geometry are not used in kinship recognition 2D images [?]. The objectives of datasets establishment were,
methods, but face images, and DNA analysis are common special purpose (ID, gender recognition, etc.) or to make
methods. DNA acquisition of sample has disadvantages such datasets available in the public domain to help researchers.
as long processing time, privacy issues, storage space, and Contact-based and contact-free-based hand geometry systems
lack of real-time matching [5]. are the two categories. Pegs control the user’s hand place-
ment in contact-based, also known as peg-based. The freedom
Kinship recognition is a very important manner, especially of putting the hand within the device in contact-free, also
after disasters, for criminal evidence, for missing children, and known as peg-free, might be viewed as a significant gain in
family relative verification [6] . In some cases, a person’s face practice. In a recent work-study, based on the contact-free-
may be smashed or injured, therefore, its image cannot be used based hand geometry biometrics approach presented in [11],
for kinship recognition. Here in this study, we are attempting a large dataset of human hand images using USB Document
to employ hand images for kinship recognition (especially Camera, (11K Hands), which contains dorsal and palmar
the detection of the relatedness of an individual to a certain sides of human hand images with a size 1600*1200. Another
family) to offer additional assistance tools or to be fused with dataset was presented in [10], denoted as U-HD 1, It con-
kinship recognition based on facial images. The first barrier tains the right and left hand frontal and dorsal images of 57
in this way was the absence of hand-image’s dataset based persons (males and females), using a Samsung digital cam-
on kinship ground truth. To produce an image dataset, there era. The researchers in [12] used a scanner to collect 180
were many parameters that should be selected properly and images of palms with a size of 2528*1800 pixels. Table I
fine-tuned to get useful data like, image capturing setup, per- summarizes some literatures that deals with the use of hand
sons’ information selection and saving, and image coding. images dataset and their purpose. Numerous earlier efforts
This is called ”image acquisition stage ”, that is a crucial step have demonstrated success in a variety of applications using
in any biometric framework [7], after the image is captured, the databases indicated above and others, including identi-
pre-processing is carried out to only get hand area informa- fication, hand recognition, age determination, and gender
tion. Feature extraction is the next stage, and it is a distinctive determination. [10, 13–15]. The earliest successful device
form of reduction of dimensionality in image processing, can using the hand geometry technique was ”Identimat,” which
be performed using handcrafted methods or deep learning was implemented during the 1970s [14]. With the increas-
methods. This structure of hand consists of the length of the ing demand for reliable and automatic solutions, biometric
fingers, the width of the palm, the thickness of the palm, and recognition is becoming ever more widely deployed in many
the width of the fingers at various points. Even though these commercial, government, and forensic applications. In 2015,
indicators do not greatly differ throughout the population, they a work for verification purposes was presented [16]; Using the
can nevertheless be used to various situations [7]. The resul- IIT Delhi Touchless Palmprint Database, the system achieved
tant feature vector from this stage will be used as an input to an accuracy of 95.5%. Hussein et al., achieved the same accu-
an artificial neural network for class prediction. As compared racy for identification and verification purposes; they worked
to other approaches, hand geometry offers numerous benefits, on a hand image dataset of 35 persons and classified images
including a low computational methodology, a short template using an artificial neural network [13]. In Gender Recognition
size, and a user-friendly design [8]. Challenges, Mahmoud Afifi, tested a set of state-of-the-art
techniques on the ”11k Hands” dataset for gender classifica-
tion, and the results showed that dorsal hand images possess
distinctive features that could help with gender recognition
and biometric identification problems [11]. Another work
showed 90% accuracy in human gender recognition using the
U-HD 1 hand image dataset [10]. A technique for predicting
human age using hand photographs was reported in 2020; the
results included a categorization into 17 age groups ranging
from 18 to 75; the model employed was trained using the
”11K Hands” dataset; and the approach demonstrated a 96.5%
