APL: Adaptive Parameter Learning for Image Dehazing

Foggy or hazy images result from light scattering and absorption by atmospheric particles. Intensity transformation techniques offer solutions to solve this problem, but param-eters selection significantly impacts the quality of the final image. In this paper we propose the APL method, a parameter learning approach for enhancing hazy images using Convolutional Neural Networks (CNNs), resulting in a new ISP learning-based pipeline. A set of intensity transformation techniques is applied, combined with image quality metrics, to define parameters for hazy image enhancement. A CNN regression model is employed to learn about the problem and estimate parameters for the transfor-mation stage. The best dehazing parameters are determined and utilized to enhance the quality of degraded images. Experiments are conducted on three datasets of hazy images, including two datasets available in the literature and one proposed dataset of real-world foggy images. Results are evaluated comparing to other dehazing methods using full-reference (PSNR and SSIM) and non-reference (NIQE and BRISQUE) metrics, demonstrating a high accuracy in image dehazing achieved by our method. The proposed source code and dataset are available here.