Conventional cameras capture image irradiance (RAW) on a sensor and convert it to RGB images using an image signal processor (ISP). The images can then be used for photography or visual computing tasks in a variety of applications, such as public safety surveillance and autonomous driving. One can argue that since RAW images contain all the captured information, the conversion of RAW to RGB using an ISP is not necessary for visual computing. In this paper, we propose a novel $\rho$ -Vision framework to perform high-level semantic understanding and low-level compression using RAW images without the ISP subsystem used for decades. Considering the scarcity of available RAW image datasets, we first develop an unpaired CycleR2R network based on unsupervised CycleGAN to train modular unrolled ISP and inverse ISP (invISP) models using unpaired RAW and RGB images. We can then flexibly generate simulated RAW images (simRAW) using any existing RGB image dataset and finetune different models originally trained in the RGB domain to process real-world camera RAW images. We demonstrate object detection and image compression capabilities in RAW-domain using RAW-domain YOLOv3 and RAW image compressor (RIC) on camera snapshots. Quantitative results reveal that RAW-domain task inference provides better detection accuracy and compression efficiency compared to that in the RGB domain. Furthermore, the proposed $\rho$ -Vision generalizes across various camera sensors and different task-specific models. An added benefit of employing the $\rho$ -Vision is the elimination of the need for ISP, leading to potential reductions in computations and processing times.

Image segmentation is a challenging task in the field of image processing and computer vision. In order to obtain an accurate segmentation performance, user interaction is always used in practical image-segmentation applications. However, a good segmentation method should not rely on much prior information. In this paper, an efficient superpixel-guided interactive image-segmentation algorithm based on graph theory is proposed. In this algorithm, we first perform the initial segmentation by using the MeanShift algorithm, then a graph is built by taking the pre-segmented regions (superpixels) as nodes, and the maximum flow–minimum cut algorithm is applied to get the superpixel-level segmentation solution. In this process, each superpixel is represented by a color histogram, and the Bhattacharyya coefficient is chosen to calculate the similarity between any two adjacent superpixels. Considering the over-segmentation problem of the MeanShift algorithm, a narrow band is constructed along the contour of objects using a morphology operator. In order to further segment the pixels around edges accurately, a graph is created again for those pixels in the narrow band and, following the maximum flow–minimum cut algorithm, the final pixel-level segmentation is completed. Extensive experimental results show that the presented algorithm obtains much more accurate segmentation results with less user interaction and less running time than the widely used GraphCut algorithm, Lazy Snapping algorithm, GrabCut algorithm and a region merging algorithm based on maximum similarity (MSRM).

In recent years, Wireless Sensor Networks with a Mobile Sink (WSN-MS) have been an active research topic due to the widespread use of mobile devices. However, how to get the balance between data delivery latency and energy consumption becomes a key issue of WSN-MS. In this paper, we study the clustering approach by jointly considering the Route planning for mobile sink and Clustering Problem (RCP) for static sensor nodes. We solve the RCP problem by using the minimum travel route clustering approach, which applies the minimum travel route of the mobile sink to guide the clustering process. We formulate the RCP problem as an Integer Non-Linear Programming (INLP) problem to shorten the travel route of the mobile sink under three constraints: the communication hops constraint, the travel route constraint and the loop avoidance constraint. We then propose an Imprecise Induction Algorithm (IIA) based on the property that the solution with a small hop count is more feasible than that with a large hop count. The IIA algorithm includes three processes: initializing travel route planning with a Traveling Salesman Problem (TSP) algorithm, transforming the cluster head to a cluster member and transforming the cluster member to a cluster head. Extensive experimental results show that the IIA algorithm could automatically adjust cluster heads according to the maximum hops parameter and plan a shorter travel route for the mobile sink. Compared with the Shortest Path Tree-based Data-Gathering Algorithm (SPT-DGA), the IIA algorithm has the characteristics of shorter route length, smaller cluster head count and faster convergence rate.

Multi-object tracking (MOT) is a task to identify objects in videos, however, objects with similar appearance or occlusion may cause frequent ID switching, which is the main challenge of current MOT. In this paper, we propose a novel self-cross graph neural network-based multi-object tracking method, which we termed as SCGTracker. This method seamlessly integrates object detection and tracking through graph neural networks, building upon the foundation of the JDE paradigm. Specifically, we construct graph structures to capture the correlation between objects in both spatial and temporal dimensions. To further tackle the frequent ID switching problem, we employ an attention mechanism to aggregate object context information within the same frame and across different frames, updating the object information via graph neural networks to derive highly distinctive appearance features. Ultimately, the obtained strongly distinguishable object appearance features serve to mitigate the issue of frequent object ID switches. In experiments conducted on the MOT17 test set, our proposed method yields promising results, achieving a 73% Multiple Object Tracking Accuracy (MOTA) and a 73.2% ID F1 score. Furthermore, it demonstrates a substantial reduction in ID switches compared with state-of-the-art methods.

With the widespread adoption of mobile multimedia devices, the deployment of compute-intensive inference tasks on edge and resource-constrained devices, particularly in the context of low-light text detection, remains a formidable challenge. Existing deep learning approaches have shown limited effectiveness in restoring images for extremely dark scenes. To address these limitations, this paper presents a novel cloud-based Low-light Attention Enhancement Generative Adversarial Network for unpaired text images (LAE-GAN) for the non-paired text image enhancement task in extremely low-light conditions. In the first stage, compressed low-light images are transmitted from edge devices to a cloud server for image enhancement. The LAE-GAN, an end-to-end network comprising a Zero-DCE and AGM-net generator, is designed with a global and local discriminator structure. The initial illumination restoration of extremely low-light images is accomplished using the Zero-DCE network. To enhance text details, we propose an Enhanced Text Attention Mechanism (ETAM) that transforms text information into a comprehensive text attention mechanism across the entire network. The Sobel operator is employed to extract text edge information, while attention is focused on text region details through constraints imposed on the attention map and edge map. Additionally, an AGM-Net module is integrated to reduce noise and fine-tune illumination. In the second stage, the cloud server makes decisions based on user requirements and processes requests in parallel, scaling with the quantity of requests. In the third stage, the enhanced results are transmitted back to edge devices for text detection. Experimental results on widely used LOL and SID low-light datasets demonstrate significant improvements in both quantitative and qualitative analysis, surpassing state-of-the-art enhancement methods in terms of image restoration and text detection.