Image Dehazing Based on Local and Non-Local Features

Image dehazing is a traditional task, yet it still presents arduous problems, especially in the removal of haze from texture and edge information an image. The state-of-the-art methods may result loss some visual informative details decrease quality. To improve quality, novel model proposed, based on fractional derivative data-driven regularization terms. In this model, contrast constrained adaptive histogram equalization method used as data fidelity item; applied to avoid over-enhancement noise amplification; proposed terms are adopted extract local non-local features Then, solve half-quadratic splitting used. Moreover, dual-stream network Convolutional Neural Network (CNN) Transformer introduced structure regularization. Further, estimate atmospheric light, light veil proposed. Extensive experiments display effectiveness method, which surpasses for most synthetic real-world images, quantitatively qualitatively.