Shadow Detection and Removal from Hand Images using Synthetic Dataset

Abstract

There is a significant demand for biometric access in terms of authentication recently. In addition to the fingerprint recognition system, contactless hand-based biometric systems, such as palm vein recognition, have emerged as notable approaches. However, palm vein recognition systems face numerous challenges in real-world environments, one of which is shadows. Shadows can disrupt the palm vein structure and lead to incorrect recognition. In this project, our goal is to enhance palm vein images by addressing the issue of shadows on hand. To achieve this, we carried out experiments involving the ST- CGAN and SP+M Net in addition to a hand extraction module for shadow detection and shadow removal specifically on the hand. In addition, rather than using the unchanged shadow mask for model training, we introduce a random shape generator to produce the random shadow mask. Furthermore, we also present a method for generating a new synthetic dataset during each training iteration, which aims to enrich the dataset and improve the training process