A Point Light Source Interference Removal Method for Image Dehazing

Yanyang Yan, Shengdong Zhang, Mingye Ju, Wenqi Ren, Rui Wang, Yuanfang Guo; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2020, pp. 874-875

Single image haze removal has been a challenging problem and the performance of the most existing dehazing methods is degraded when point light sources exist in the hazy image. In this paper, we propose a point light source interference removal method (PLiSIR) to reduce the interferences when estimating the atmospheric light. According to our observation, the pixel intensity around the point light sources can be modeled approximately by Gaussian distribution. The locations of the interfered pixels are obtained reasonably regardless of the specific number of light sources. A binary masking map is then created for distinguishing whether the pixel is affected by light sources and thus PLiSIR can be adopted to dehazing algorithms by removing the interfered pixels, during the estimation of the atmospheric light. To demonstrate how to apply PLiSIR to different algorithms, we select the dark channel prior dehazing method (DCP) and the color attenuation prior dehazing method (CAP) as two carrier methods and introduce the adaptations accordingly. Experimental results indicate that the PLiSIR can assist DCP and CAP to better estimate the atmospheric light, and thus generate better dehazing results compared to the original DCP and CAP methods. Moreover, PLiSIR also helps DCP and CAP to simplify the parameter adjustment process of the guided filter. At last, we compare our modified DCP approach (which we refer to PLiSIR-DCP) with the state-of-the-art nighttime dehazing algorithm to present an approach which is suitable for both daytime and nighttime haze removal.