This work addresses the problem of semantic foggy scene understanding (SFSU). Although extensive research has been performed on image dehazing and on semantic scene understanding with clear-weather images, little attention has been paid to SFSU. Due to the difficulty of collecting and annotating foggy images, we choose to generate synthetic fog on real images that depict clear-weather outdoor scenes, and then leverage these partially synthetic data for SFSU by employing state-of-the-art convolutional neural networks (CNN). In particular, a complete pipeline to add synthetic fog to real, clear-weather images using incomplete depth information is developed. We apply our fog synthesis on the Cityscapes dataset and generate Foggy Cityscapes with 20550 images. SFSU is tackled in two ways: 1) with typical supervised learning, and 2) with a novel type of semi-supervised learning, which combines 1) with an unsupervised supervision transfer from clear-weather images to their synthetic foggy counterparts. In addition, we carefully study the usefulness of image dehazing for SFSU. For evaluation, we present Foggy Driving, a dataset with 101 real-world images depicting foggy driving scenes, which come with ground truth annotations for semantic segmentation and object detection. Extensive experiments show that 1) supervised learning with our synthetic data significantly improves the performance of state-of-the-art CNN for SFSU on Foggy Driving; 2) our semi-supervised learning strategy further improves performance; and 3) image dehazing marginally advances SFSU with our learning strategy. The datasets, models and code are made publicly available.

Semantic Foggy Scene Understanding with Synthetic Data

This work addresses the problem of semantic foggy scene understanding (SFSU). Although extensive research has been performed on image dehazing and on semantic scene understanding with clear-weather images, little attention has been paid to SFSU. Due to the difficulty of collecting and annotating foggy images, we choose to generate synthetic fog on real images that depict clear-weather outdoor scenes, and then leverage these partially synthetic data for SFSU by employing state-of-the-art convolutional neural networks (CNN). In particular, a complete pipeline to add synthetic fog to real, clear-weather images using incomplete depth information is developed. We apply our fog synthesis on the Cityscapes dataset and generate Foggy Cityscapes with 20,550 images. SFSU is tackled in two ways: (1) with typical supervised learning, and (2) with a novel type of semi-supervised learning, which combines (1) with an unsupervised supervision transfer from clear-weather images to their synthetic foggy counterparts. In addition, we carefully study the usefulness of image dehazing for SFSU. For evaluation, we present Foggy Driving, a dataset with 101 real-world images depicting foggy driving scenes, which come with ground truth annotations for semantic segmentation and object detection. Extensive experiments show that (1) supervised learning with our synthetic data significantly improves the performance of state-of-the-art CNN for SFSU on Foggy Driving; (2) our semi-supervised learning strategy further improves performance; and (3) image dehazing marginally advances SFSU with our learning strategy. The datasets, models and code are made publicly available.

Images captured in hazy or foggy weather conditions are seriously degraded by the scattering of atmospheric particles, which directly influences the performance of outdoor computer vision systems. In this paper, a fast algorithm for single image dehazing is proposed based on linear transformation by assuming that a linear relationship exists in the minimum channel between the hazy image and the haze-free image. First, the principle of linear transformation is analyzed. Accordingly, the method of estimating a medium transmission map is detailed and the weakening strategies are introduced to solve the problem of the brightest areas of distortion. To accurately estimate the atmospheric light, an additional channel method is proposed based on quad-tree subdivision. In this method, average grays and gradients in the region are employed as assessment criteria. Finally, the haze-free image is obtained using the atmospheric scattering model. Numerous experimental results show that this algorithm can clearly and naturally recover the image, especially at the edges of sudden changes in the depth of field. It can, thus, achieve a good effect for single image dehazing. Furthermore, the algorithmic time complexity is a linear function of the image size. This has obvious advantages in running time by guaranteeing a balance between the running speed and the processing effect.

Fast Image Dehazing Method Based on Linear Transformation

Low-light is an inescapable element of our daily surroundings that greatly affects the efficiency of our vision. Research works on low-light has seen a steady growth, particularly in the field of image enhancement, but there is still a lack of a go-to database as benchmark. Besides, research fields that may assist us in low-light environments, such as object detection, has glossed over this aspect even though breakthroughs-after-breakthroughs had been achieved in recent years, most noticeably from the lack of low-light data (less than 2% of the total images) in successful public benchmark dataset such as PASCAL VOC, ImageNet, and Microsoft COCO. Thus, we propose the Exclusively Dark dataset to elevate this data drought, consisting exclusively of ten different types of low-light images (i.e. low, ambient, object, single, weak, strong, screen, window, shadow and twilight) captured in visible light only with image and object level annotations. Moreover, we share insightful findings in regards to the effects of low-light on the object detection task by analyzing visualizations of both hand-crafted and learned features. Most importantly, we found that the effects of low-light reaches far deeper into the features than can be solved by simple "illumination invariance'". It is our hope that this analysis and the Exclusively Dark dataset can encourage the growth in low-light domain researches on different fields. The Exclusively Dark dataset with its annotation is available at this https URL

Getting to Know Low-light Images with The Exclusively Dark Dataset

Images captured under poor illumination conditions often exhibit characteristics such as low brightness, low contrast, a narrow gray range, and color distortion, as well as considerable noise, which seriously affect the subjective visual effect on human eyes and greatly limit the performance of various machine vision systems. The role of low-light image enhancement is to improve the visual effect of such images for the benefit of subsequent processing. This paper reviews the main techniques of low-light image enhancement developed over the past decades. First, we present a new classification of these algorithms, dividing them into seven categories: gray transformation methods, histogram equalization methods, Retinex methods, frequency-domain methods, image fusion methods, defogging model methods and machine learning methods. Then, all the categories of methods, including subcategories, are introduced in accordance with their principles and characteristics. In addition, various quality evaluation methods for enhanced images are detailed, and comparisons of different algorithms are discussed. Finally, the current research progress is summarized, and future research directions are suggested.

/pdf/an-experiment-based-review-of-low-light-image-enhancement-5dhscyutxx.pdf

An Experiment-Based Review of Low-Light Image Enhancement Methods

Adaptive image enhancement method for correcting low-illumination images

Dehazing for images with large sky region

Due to the scattering of atmospheric particles, images captured under hazy conditions suffer from contrast attenuation and color distortion, which severely affect the performance of machine vision systems. Various types of methods have been developed to improve the clarity of images. However, these methods are typically challenging to apply in real-time systems. We present a fast, single-image dehazing method based on the atmospheric scattering theory and dark channel prior theory. The transmission map is approximately estimated using a fast average filter, the subsection mechanism is designed to avoid the high brightness of the sky region in the recovered image, the region projection method is adopted to obtain the atmospheric light, and image color compensation is implemented using the Weber–Fechner law. Our experimental results show that this algorithm can restore images to a clear and natural state and ensure the balance of quality and the speed of image restoration. Therefore, the algorithm can be used in real-time systems.

Xiaojin Wu

Papers

Fast Image Dehazing Method Based on Linear Transformation

An Experiment-Based Review of Low-Light Image Enhancement Methods

Adaptive image enhancement method for correcting low-illumination images

Dehazing for images with large sky region

A Fast Single-Image Dehazing Method Based on a Physical Model and Gray Projection