This dissertation establishes a physics-informed computational imaging framework for removing optical artifacts across diverse imaging systems. Addressing two critical challenges——optical aberrations in virtual reality head-mounted displays and lens flare/glare in low-light photography——the research demonstrates how physical principles can systematically guide computational solutions. For VR HMDs, we develop a real-time pre-correction system that combines wavefront modeling through Zernike polynomials with optimized L2-regularized deconvolution, achieving 19.5 ms processing throughput at 1080p resolution while significantly improving visual fidelity without hardware modifications. For nighttime flare and glare removal, we introduce the Physically-Based Flare and Glare (PBFG) dataset, generated through computational rendering that accurately simulates diffraction and scattering phenomena. Leveraging this dataset, we develop two specialized networks: FlareHRF incorporates hybrid receptive fields through parallel convolutional attention and frequency-aware feed-forward networks, while FGRNet employs spatial-frequency enhancement with histogram matching to maintain visual consistency. Both architectures demonstrate remarkable capability in removing complex artifacts, including off-frame light sources and weak streaks, with FGRNet achieving up to 3 dB PSNR improvement in glare regions. Comprehensive experiments validate our physics-informed approach across both applications, showing state-of-the-art performance in artifact removal while maintaining physical plausibility and computational efficiency. This work provides a coherent methodology for integrating physical optics with computational imaging, with direct applications in immersive displays, autonomous driving, and computational photography.