Hybrid Cyclegan and Frequency Channel Attention for High-Quality Image Dehazing

Abstract

Image dehazing is critical in surveillance and automated vision systems, but existing approaches struggle to generalize across various haze situations. This paper presents a sophisticated strategy to enhancing fuzzy images that combines CycleGAN with Frequency Channel Attention, dramatically boosting clarity and usability. CycleGAN, an unsupervised deep learning system, can transform hazy images into clear ones without requiring paired datasets, making it ideal for real-world settings. The generator network learns how to map hazy and haze- free images, restoring visibility in tough conditions. A Frequency Channel Attention technique is used to improve dehazing effectiveness by allowing the model to choose focus essential visual features in both the spatial and frequency domains. This enhances the recovery of fine details and edges while lowering haze-related noise. This technique provides highquality dehazing results by combining the generative capabilities of CycleGAN with frequency-basd attention. Benchmark dataset evaluations show that the performance is superior in terms of PSNR, SSIM, and visual quality. This technology is especially useful for real-time applications like autonomous navigation, remote sensing, and visual surveillance, which solve critical issues in foggy situations.