Weighted Generalization of Dark Channel Prior with Adaptive Color Correction for Defogging Yosuke Ueki, Masaaki Ikehara EEE Dept., Keio Univ., Yokohama, Kanagawa 223-8522, Japan Email: {ueki,ikehara}@tkhm.elec.keio.ac.jp Abstract—Images and video captured in water or fog suffer from low contrast and color distortion due to light scattering and absorption. An image formation model for hazy images is commonly used to restore both underwater images and hazy images because of the similarity between the two types of images. However, red light is attenuated faster than blue and green light in underwater, and underwater images are distorted by changes of color tone. Therefore, most current methods are specialized for either hazy images or underwater images. In this paper, we propose a novel defogging method which is efficient for both hazy images and underwater images. Our method is composed of adaptive color correction and weighted generalization of dark channel prior (WGDCP). Experimental results show that our algorithm can recover both underwater images and hazy images. Index Terms—Image processing, image enhancement, image restoration, underwater image, dehazing, defogging I. I NTRODUCTION Capturing high quality digital images by a basic or small camera is quite important for society and the demands are high. However, digital images are not always taken under good conditions but sometimes taken under bad conditions. We can use special cameras or devices to take images or videos in such condition, but these devices are usually expensive and often large and heavy. Therefore, enhancing the quality of the image which are taken under bad conditions by software have been tried. There are various types of images to enhance or recover the quality and image defogging is one of them. In hazy images, the contrast and visibility of images are often degraded due to turbid media such as haze or water. In addtion, underwater images and hazy images on land have something in common and similar methods are presently used to recover or enhance the images. Image formation models have been used to recover or enhance the quality of hazy images. Many researches have tried assmuptions to solve the image formation model for hazy images. He et al. [1] proposed the dark channel prior model (DCP) which is effective for the reconstruction of hazy images taken outdoors . DCP introduced the special assumption for haze images on the analysis of the haze images and outdoor haze-free images. They found that the lowest intensity among RGB channels in local region (dark channel) of haze-free image are often very low except the sky region and the intensity of dark channel in the haze image is not low due to the haze. Many other approaches such as . [2] have been proposed based on DCP because of its prominence. DCP has also been applied to underwater image restoration due to the similarity between (a) Underwater Image (b) Haze Image Fig. 1. Examples of images for defogging underwater images and hazy images. However, the intensity of red channel in underwater images is often much lower than other color channels because red light attenuates much faster than green and blue light. Therefore, DCP are changed to specialize for underwater image, so red channel prior [3] by Galdran et al. and UDCP [4] by S.Yang et al. have been proposed. These methods utilize only red and blue channel to calculate dark channel. There are also image formation model methods which are not based on DCP. One example, IBLA [5], achieved remarkable results for underwater image enhancement. There are also methods which are not based on image formation model. Ancuti et al. proposed an image fusion method for dehazing on land [6] and underwater image enhancement [7]. These methods involve fusing color corrected images and white balanced images by multi-scale fusion. However, it is difficult to recover both underwater images and hazy images by a single method because the attenuation of light underwater is not the same for the different wavelengths. Red light, which has longer wavelengths, is attenuated faster than blue and green light, which have shorter wavelengths. As such, underwater image enhancement or restoration present more complex and difficult problems than the dehazing of images taken on land. Y.Peng et al. focuses on expanding DCP to other types of hazy images by introducing new ambient light estimation and transmission estimation in GDCP [8]. However, the quality of images which are recovered by GDCP are not high compared to the other specialized state-of-art methods. In this paper, we propose a novel image enhancement method for both dehazing images taken on land and underwater image enhancement. Our method is motivated by GDCP [8] and WDCP [9]. WDCP solves the problems of local constant 685 978-9-0827-9705-3 EUSIPCO 2020