GlobalContextEnhancer: A Lightweight Framework for Fingerprint Denoising and Restoration on Resource-Constrained IoT Devices

Fingerprint recognition is used for the identification of unique and complex texture features that can be taken from the parts present on the epidermis.It has become popular in all personal privacy and security applications.Nevertheless, fingerprint acquisition or obtaining fingerprints from a person is influenced by several factors such as skin conditions, uneven capture by machine, fingerprint overlap and cluttered background.Fingerprint images are often corrupted due to these problems to introduce a significant amount of noise and loss of information.Thus, the performance efficiency and recognition rate error of the subsequent systems would be negatively affected.So, it is necessary to enhance the performance system through denoising and restoration approaches of fingerprint images.To remedy these issues this study proposes a new fingerprint image restoration technique built on an autoencoder model incorporating two new modules: Multi-Scale Fusion Module (MSF) and Global Feature Extraction Module (GFE).The use of the MSF module with feature pyramid structures allows the model to exploit more local features which lead to less feature information gap between different scales.The GFE module's aim is to work in the frequency domain to take the compact representation of low-frequency components as a way to extract the global structural characteristics of fingerprint images.Using these components can help one avoid the loss of details and keep the computations light.The experimental results demonstrate that the proposed framework achieves better restorations on mainstream fingerprint datasets at a low computational cost.This offers a workable answer for fingerprint image handling in restricted situations.