Haze scene removal via CNN, dynamic focussing and heuristic wavelet optimization
Asem Khmag · Journal of Modern Optics · 2025
Haze significantly impacts vision reliability in various image processing and computer vision applications. Most existing dehazing techniques fail to meet computer vision requirements when handling severe fog and environmental haze. The key contributions of this study are as follows: we developed a novel approach that integrates image structure with second-generation wavelets using a convolutional neural network. Second, we applied the SGWs technique to incorporate the internal structure constraints of digital images, enhancing both overall structure and fine details in the dehazed images. Lastly, we propose a deep CNN to improve the performance of the developed model and to focus on the generation of detail textures rather than distinguishing the image degradation. We evaluated the performance of the proposed method with best haze removal algorithms using quantitative metrics such as PSNR and structural similarity index. In terms of computational efficiency, our method achieved significantly lower running times compared to learning-based techniques.