In industrial production, video surveillance is a key measure to maintain the industrial production and the safety of the employees. However, in some special fields, monitoring images are often affected by dust scattering and water mist scattering that leads to fogging and blurring, and results in degraded image quality, which influence the visibility of the visual observation. In addition, due to the specialty of lighting and the scenery in these working environments, commonly used dark channel dehazing algorithms have disadvantages, such as color shift, darkness, halo, etc. The traditional method is based on the atmospheric scattering model. It allocates the same value of the atmospheric light to all the pixel, which is suitable for the case of solar illumination, because the spectrum of sunlight is relatively close to the three channels of red green and blue. However, in special environments such as the presence of artificial light illumination, where the field of view and depth of field are limited, it is irrational to consider the atmospheric light of the image as the same value. Therefore, in the traditional algorithm, the dehazing image often results in color distortion, darkness, halo, etc. In order to solve these problems and better restore the image in the special environment, this paper focuses on monitoring images of underground coal mining scenery and proposes an improved fast dark channel algorithm to address these defects. This algorithm takes into account of the color pattern of the image and combines the dehazing algorithm with it to suppress the generation of color distortion. According to the theory of image color composition, the ratio of the Red, Green and Blue channels determines the hue of the pixel, and the changes of hue causes color distortion. So, to maintain the hue unchanged, it is necessary to control the ratio of the intensity values of the three channels. In this paper, the change of the pixel-based ratios between the three color channels before and after the processing of defogging with traditional dark channel prior method is analyzed to describe the generation of color distortion. Then based on this, the compensation method is proposed in which the atmospheric light value of each pixel is calculated and corrected according to the ratio of color channels in the input foggy image. Since the intensity distribution of the fog in the image is in the low frequency, the high-frequency and low- frequency parts of the image are firstly separated and the dehazing processing is only applied to the low- frequency part in order to preserve the image details. Besides this algorithm takes more measures to avoid the details of the image being destroyed, that is, the image of transmission and initial dark channel are processed by guided filtering, so that more details can be preserved. Furthermore, this algorithm proposes customized brightness and saturation enhancement functions for underground images that compensate low brightness and low saturation of the defogged images from traditional dark channel dehazing algorithm. In addition to mining images, this paper also applies the algorithm to common dehazing image datasets and compares it with some other algorithms. The comparison metric used in this article is structural similarity, which is a common measure for the similarity of two images. In addition, we also propose a new evaluation parameter, the color similarity, which is to detect the similarity between the dehazing image and the origin image in terms of color to give the evaluation of different methods on minimizing color distortion. The results show that the algorithm proposed in this paper can effectively remove fog, correct the color distortion and halo phenomenon, and improve the brightness and color saturation effectively. The algorithm provides a new means for defogging in industries such as mining and life scenarios. This algorithm provides a new idea for image processing in non-natural lighting and low brightness environments, and provides a guarantee for the smooth progress of underground production.
