Large-scale datasets centered on images have driven advancements in deep learning-based computer vision applications. While there is an abundance of datasets containing images depicting favorable weather scenes, datasets featuring images of adverse weather conditions, especially the presence of haze, are scarce due to challenges in their collection. In response, we leverage the advantages of deep learning techniques to introduce a novel approach for facilitating the rendering of realistic and diverse hazy images, named RenHaze. To be specific, RenHaze adopts a denseness parameter \(\omega\) to control the haze level of output images and consists of five subnets, including a content exploitation (CE) subnet, a depth exploitation (DE) subnet, a haze exploitation (HE) subnet, an image generation (IG) subnet, and an image discernment (ID) subnet. The CE, DE, and HE subnets are responsible for extracting features from the source clear image, depth image, and reference hazy image, respectively, and then providing them for the IG subnet. The IG subnet is used to perform image translation in a coarse-to-fine manner, while the ID subnet is employed to discern the realism of the rendered image and provide feedback to the IG subnet for generating the desired output. Extensive experiments demonstrate the superiority of the proposed model over competing IG methods in terms of the realism and diversity of synthesized hazy images, as well as its effectiveness in boosting the performance of computer vision tasks such as object detection and semantic segmentation in real-world hazy environments.

In recent years, object detection has significantly advanced by using deep learning, especially convolutional neural networks. Most of the existing methods have focused on detecting objects under favorable weather conditions and achieved impressive results. However, object detection in the presence of rain remains a crucial challenge owing to the visibility limitation. In this article, we introduce an amalgamating knowledge network (AK-Net) to deal with the problem of detecting objects hampered by rain. The proposed AK-Net obtains performance improvement by associating object detection with visibility enhancement, and it is composed of five subnetworks: rain streak removal (RSR) subnetwork, raindrop removal (RDR) subnetwork, foggy rain removal (FRR) subnetwork, feature transmission (FT) subnetwork, and object detection (OD) subnetwork. Our approach is flexible; it can adopt different object detection models to construct the OD subnetwork for the final inference of objects. The RSR, RDR, and FRR subnetworks are responsible for producing clean features from rain streak, raindrop, and foggy rain images, respectively, and offer them to the OD subnetwork through the FT subnetwork for efficient object prediction. Experimental results indicate that the mean average precision (mAP) achieved by our proposed AK-Net was up to 19.58% and 26.91% higher than those produced using competitive methods on published iRain and RID datasets, respectively, while preserving the fast-running time of the baseline detector.

Despite the recent improvement of object detection techniques, many of them fail to detect objects in low-luminance images. The blurry and dimmed nature of low-luminance images results in the extraction of vague features and failure to detect objects. In addition, many existing object detection methods are based on models trained on both sufficient- and low-luminance images, which also negatively affect the feature extraction process and detection results. In this article, we propose a framework called Self-adaptive Feature Transformation Network (SFT-Net) to effectively detect objects in low-luminance conditions. The proposed SFT-Net consists of the following three modules: (1) feature transformation module, (2) self-adaptive module, and (3) object detection module. The purpose of the feature transformation module is to enhance the extracted feature through unsupervisely learning a feature domain projection procedure. The self-adaptive module is utilized as a probabilistic module producing appropriate features either from the transformed or the original features to further boost the performance and generalization ability of the proposed framework. Finally, the object detection module is designed to accurately detect objects in both low- and sufficient- luminance images by using the appropriate features produced by the self-adaptive module. The experimental results demonstrate that the proposed SFT-Net framework significantly outperforms the state-of-the-art object detection techniques, achieving an average precision (AP) of up to 6.35 and 11.89 higher on the sufficient- and low- luminance domain, respectively.

Images are often corrupted by natural obscuration (e.g., snow, rain, and haze) during acquisition in bad weather conditions. The removal of snowflakes from only a single image is a challenging task due to situational variety and has been investigated only rarely. In this article, we propose a novel snow removal framework for a single image, which can be separated into a sparse image approximation module and an adaptive tolerance optimization module. The first proposed module takes the advantage of sparsity-based regularization to reconstruct a potential snow-free image. An auto-tuning mechanism for this framework is then proposed to seek a better reconstruction of a snow-free image via the time-varying inertia weight particle swarm optimizers in the second proposed module. Through collaboration of these two modules iteratively, the number of snowflakes in the reconstructed image is reduced as generations progress. By the experimental results, the proposed method achieves a better efficacy of snow removal than do other state-of-the-art techniques via both objective and subjective evaluations. As a result, the proposed method is able to remove snowflakes successfully from only a single image while preserving most original object structure information.

Visibility restoration techniques are widely used for information recovery of hazy images in many computer vision applications. Estimation of haze density is an essential task of visibility restoration techniques. However, conventional visibility restoration techniques often suffer from either the generation of serious artifacts or the loss of object information in the restored images due to uneven haze density, which usually means that the images contain heavy haze formation within their background regions and little haze formation within their foreground regions. This frequently occurs when the images feature real-world scenes with a deep depth of field. How to effectively and accurately estimate the haze density in the transmission map for these images is the most challenging aspect of the traditional state-of-the-art techniques. In response to this problem, this work proposes a novel visibility restoration approach that is based on Bi-Histogram modification, and which integrates a haze density estimation module and a haze formation removal module for effective and accurate estimation of haze density in the transmission map. As our experimental results demonstrate, the proposed approach achieves superior visibility restoration efficacy in comparison with the other state-of-the-art approaches based on both qualitative and quantitative evaluations. The proposed approach proves effective and accurate in terms of both background and foreground restoration of various hazy scenarios.

Background subtraction for motion detection is often used in video surveillance systems. However, difficulties in bootstrapping restrict its development. This article proposes a novel hybrid background subtraction technique to solve this problem. For performance improvement of background subtraction, the proposed technique not only quickly initializes the background model but also eliminates unnecessary regions containing only background pixels in the object detection process. Furthermore, an embodiment based on the proposed technique is also presented. Experimental results verify that the proposed technique allows for reduced execution time as well as improvement of performance as evaluated by Recall, Precision, F1, and Similarity metrics when used with state-of-the-art background subtraction methods.