Fog and haze degrade image quality by scattering light, reducing contrast, and obscuring critical details in applications like autonomous driving and surveillance. This comprehensive survey examines research works on foggy image enhancement, systematically categorizing methods into traditional physics-based approaches (e.g., Retinex, Dark Channel Prior), learning-based techniques (e.g., GANs), and hybrid fusion strategies (e.g., LiDAR-RGB). Traditional methods like variable-filter Retinex and wavelet-YUV decomposition achieve real-time processing (125 ms) but struggle with dense fog and color distortion. Learning-based approaches, such as CycleGAN and physics-guided models, improve generalization and achieve superior quality (PSNR = 28.5 dB) but require significant computational resources (3.09 s processing time). Hybrid techniques combine sensor data with image processing, achieving excellent structural preservation (SSIM = 0.88) but face hardware complexity constraints. Performance analysis reveals fundamental trade-offs between speed, quality, and deployment complexity. Key unresolved challenges include extreme fog handling, real-time edge deployment, and cost-effective multi-modal fusion.