Nighttime Single Image Dehazing via Pixel-Wise Alpha Blending

Teng Yu,Chenglizhao Chen,Guowei Yang,Pu Miao,Huan Yang,Kang Song

doi:10.1109/access.2019.2936049

Abstract

In this paper, we propose a novel method to address the nighttime single image dehazing problem. Estimation of the ambient illumination map and transmission map are the key steps of modern dehazing approaches. For hazy scenes at night, ambient illumination is usually not globally isotropic as a nighttime scene typically contains multiple light sources. Frequently, Light source regions and non-light source regions exhibit distinct color features. However, existing nighttime dehazing methods have been attempting to process these two regions based on identical prior assumptions. Moreover, the commonly-used local maximum pixel method tends to over-estimate the ambient illumination. These two drawbacks result in color distortions and halo artifacts around the light source regions in the output images. In this work, we present a pixel-wise alpha blending method for estimating the transmission map, where the transmissions estimated from dark channel prior (non-light source region) and the proposed bright channel prior (light source region) are effectively blended into one transmission map guided by a brightness-aware weights map. Based on the Retinex theory, a channel difference guided filtering method is proposed to estimate the ambient illumination, which produces a spatially variant low-frequency passband that selectively retains the high-frequency edge details. Extensive experiments on the benchmarks demonstrate that our method outperforms the state-of-the-art methods for nighttime image dehazing, especially in terms of color consistency and halo artifacts reduction in the dehazed images.

Highlights

Images or videos captured at foggy or hazy weathers are usually degraded due to the presence of suspended particles and water droplets in the air
The widely used dark channel prior in the existing methods is not valid in the light source regions, since no dark channel exists in white lights area with high intensity values. These methods tend to produce unwanted artifacts, such as color distortions and halos, in the output images. To address these existing problems resulted by inaccurate ambient illumination estimation and invalid prior assumption in light source regions, we present a novel method for nighttime dehazing
In experiments of our method, the size of dark channel prior (DCP) and bright channel prior (BCP) is fixed to 15 × 15

Summary

INTRODUCTION

Images or videos captured at foggy or hazy weathers are usually degraded due to the presence of suspended particles and water droplets in the air. By proposing a channel difference map as the guidance for filtering, the output can obtain the local low-frequency passband as well as retain the edge details This method is more theoretical sound and provides a novel way for ambient illumination estimation at nighttime hazy scenes. Pei et al [22] map the nighttime hazy images to a daytime one by performing a color transfer technique, followed by a modified dark channel prior for haze removal This method improves the visibility in the dehazed images, whereas the overall appearance seems unrealistic. Note that the light spots appear dark in the illumination map, this is reasonable since the haze-free variable J of a light source is frequently high, leading to the direct attenuation term J (x)t(x) is almost close the hazy image I , the scattering term A(x)(1 − t(x)) can be regarded as a very small value that close to 0

BRIGHT CHANNEL PRIOR

PIXEL-WISE ALPHA BLENDING

EXPERIMENTAL EVALUATION RESULTS

DISCUSSIONS AND CONCLUSION