Fringe Projection Profilometry (FPP) is a widely recognized technique for deriving 3D profiles from images. Despite numerous methodologies developed to determine depth in FPP, the inherent triangulation setup of camera, projector, and object often introduces substantial shadows in captured fringes, especially for complex objects. These shadows can impede algorithm performance and introduce undesirable artifacts in final depth profiles. In this work, we introduce a Transformer-based Fringe Restoration network designed to repair shadowed regions in single deformed fringe images. The network comprises an object localization module to identify object regions and a shadow repair module that utilizes reference and deformed fringes to restore shadowed areas. In addition, we construct a comprehensive pseudo-realistic dataset using Blender, a computer graphics tool, to train the proposed network. Our results demonstrate precise object region segmentation with just a single fringe image, and the proposed network achieves superior fringe restoration, as quantified by Intersection over Union (IoU) and dice score metrics. Moreover, 3D reconstruction on shadowed and shadow-free deformed fringes using standard single-shot methods exhibits enhanced performance owing to the fringe restoration network.