Nighttime image dehazing based on a modified model and saturation line prior

Abstract

The reduced visibility and contrast of images captured in hazy weather seriously affect the application of computer vision. However, most of the existing dehazing methods just focus on daytime images, and the dehazing effect on nighttime images is not obvious. Therefore, this paper proposes a method based on the modified model and saturation line prior for nighttime image dehazing. Specifically, we propose a glow term added to the nighttime imaging model to remove the glow according to the characteristics of the nighttime artificial light source. Subsequently, utilise the transmittance map obtained by the saturation line prior which improved through colour space transformation and the atmospheric light map obtained by Gaussian filtering to invert the model for haze removal. In addition, the underwater image colour compensation method is improved to be suitable for nighttime images and combined with the multi-scale retinex formula, applying a dual correction to restore image colours. The experimental results demonstrate that the proposed method can be well applied to nighttime images, restoring details and colours of nighttime images. Quantitative and qualitative comparisons verify the effectiveness of the proposed method compared with the state-of-the-art methods. Numerically, compared with the original saturation line prior method, the proposed method optimizes the average values of NIQE, AG, IE, CEIQ, and SSEQ metrics by 0.138, 2.664, 0.159, 0.199, and 2.277 respectively. Furthermore, the method can be extended to inclement weather and underwater scenes, and achieve pleasant image enhancement effects, showcasing superior robustness and practical utility performance.