Multiple CNN Models For Enhanced Palmprint Recognition
Loading...
Date
2024-07-14
Journal Title
Journal ISSN
Volume Title
Publisher
University of M'sila
Abstract
EN
These days, there is more talk about increasing crime, piracy, and lack of security across
different sectors. It is also very important to verify people’s identities for financial transactions,
accessing services, and mobility. Traditional security systems use pre-existing
information (like passwords or PINs) or token-based access (like keys, IDs, or badges).
However, these systems frequently cannot discriminate between fraudsters and those
who are allowed, they are less trustworthy in many environments.In this work, we
choose to investigate one of these systems, which is a deep-learning palmprint recognition
system. This system is difficult to replicate. There are several benefits, such as affordability
and simplicity of usage. Our work may be categorized into two parts for feature
extraction : transfer learning and fine-tuning, and two strategies : learning one instance
and multiple instances. Firstly, we prepare our datasets into 3 datasets to evaluate our
proposed models : left, right, and multiple instances. After that, we select three convolutional
neural network algorithms to carry out the feature extraction and classification
operation to confirm individual Recognition using both techniques : transfer learning
and fine-tuning. The PolyU palmprint database is used to evaluate the performance
of the suggested model. Our proposed method for the PolyU palmprint database using
transfer learning achieved an accuracy of 85.25% with VGG16, 87% with DenseNet121,
and 86.25% with MobileNetV2. Using fine-tuning, we achieved an accuracy of 92.50%
with VGG16, 98.75% with DenseNet121, and 98.50% with MobileNetV2. Experimental
results conclude that the proposed work obtained good performance compared to
existing methods in multi-instance scenarios.
Description
Keywords
Palmprint Recognition, CNN, Feature extraction, multi-instance, oneinstance iv