Nighttime Haze Removal Using Bilateral Filtering and Adaptive Dark Channel Prior
Tianyou Pei, Qiaoyu Ma, Pengfei Xue, Yuan Ding, Lixin Hao, Teng Yu · 2019
The weather conditions of fog and haze make the images taken at night blurred, which seriously affects the image and video quality such as video surveillance. The existing daytime dehazing methods are not suitable for night hazy scenes, night-specific dehazing methods need to be developed. In this paper, by analyzing the nature of night hazy scenes with Retinex theory, we propose a bilateral filtering approach to estimate the locally variant atmospheric illumination, which can accurately obtain the local atmospheric light as well as preserve edges. To solve the problem that the dark channel of haze-free image in the light source area is not zero, an adaptive dark channel prior (ADCP) is proposed to estimate the transmission map. Based on the difference of image value between light source area and non-light source area, the transmission can be estimated adaptively to achieve better dehazing effect. The experimental results show that the subjective visual effect of the proposed algorithm outperform the dark channel-based method. The proposed method can effectively remove haze in the night, improve the image contrast and visibility, especially for the light areas.