Facial Expression Recognition (FER) plays a critical role in numerous applications, ranging from healthcare diagnostics to road safety systems and marketing strategies, where real-time insights into human emotions are essential. Despite significant advancements in FER under controlled conditions such as well-lit, frontal, and unobstructed settings recognition in real-world, unconstrained environments remain a persistent challenge. Occlusions present a substantial barrier by obscuring essential facial features, thereby compromising recognition accuracy. Addressing this challenge has been the focus of extensive research, leading to the development of two primary strategies: methods that analyze the visible, unobstructed regions of the face and those that aim to reconstruct the missing or occluded facial features. In this study, we propose a novel approach based on Conditional Generative Adversarial Networks (cGANs) for reconstructing occluded optical flow information critical to FER. By conditioning the reconstruction process on latent vectors derived from partially occluded optical flow images, the cGAN effectively restores the missing information while preserving the spatial and temporal coherence of facial expressions. We demonstrate that our cGAN-based method significantly enhances FER performance by mitigating the adverse effects of occlusions.