An evaluation of denoising techniques and classification of biometric images based on deep learning

A biometric system uses pattern recognition to identify an individual based on various traits such as the face, fingerprints, etc. Adverse or uncontrolled conditions can often lead to the introduction of noise in data. In this paper, we evaluate the impact of noise on the quality of images that are input into a biometric system. Further, we analyze the role of various filters in the denoising of images. The paper proposes a robust deep learning-based framework in Pytorch to classify denoised images, which is validated by evaluating performance metrics on two popular benchmarks. Our study shows that our proposed framework helps achieve state-of-the-art accuracy without any trade-off caused due to the introduction of noise in images.