Underwater captured images are usually degraded by low contrast, hazy, and blurry due to absorbing scattering, which limits their analyses applications. To address these problems, a red channel prior guided variational framework is proposed based on the complete underwater image formation model (UIFM). Unlike most of existing methods that only consider direct transmission backscattering components, we additionally include forward scattering component into UIFM. In framework, successfully...

In this study, a novel underwater colour image enhancement approach based on hue preserving is presented by combining hue–saturation–intensity (HSI) and HS–value (HSV) models. the proposed wavelet‐domain filtering (WDF) constrained histogram stretching (CHS) algorithms are operated HSI HSV models, respectively. The degraded first converted from red–green–blue model into model, wherein component H preserved WDF algorithm executed S I components. Similarly, further kept invariant as well CHS...

With the intensification of global climate change and frequent occurrence forest fires, development efficient precise fire monitoring image segmentation technologies has become increasingly important. In dealing with challenges such as irregular shapes, sizes, blurred boundaries flames smoke, traditional convolutional neural networks (CNNs) face limitations in segmentation, including flame edge recognition, class imbalance issues, adapting to complex scenarios. This study aims enhance...

The mirror detection problem is important as mirrors can affect the performances of many vision tasks. It a difficult it requires an understanding global scene semantics. Recently, method was proposed to detect by learning multi-level contextual contrasts between inside and outside mirrors, which helps locate edges implicitly. We observe that content reflects its surrounding, separated edge mirror. Hence, we propose model in this paper progressively learn similarity while explicitly...

Previous video salient object detection (VSOD) approaches have mainly focused on the perspective of network design for achieving performance improvements. However, with recent slowdown in development deep learning techniques, it might become increasingly difficult to anticipate another breakthrough solely via complex networks. Therefore, this paper proposes a universal scheme obtain further 3% improvement all state-of-the-art (SOTA) VSOD models. The major highlight our method is that we...

Detecting oriented objects along with estimating their rotation information is one crucial step for image analysis, especially remote sensing images. Despite that many methods proposed recently have achieved remarkable performance, most of them directly learn to predict object directions under the supervision only (e.g., angle) or a few several coordinates) groundtruth (GT) values individually. Oriented detection would be more accurate and robust if extra constraints, respect proposal...

The Chan-Vese model is very popular for image segmentation. Technically, it combines the reduced Mumford-Shah and level set method (LSM). This segmentation problem solved interchangeably by computing a gradient descent flow expensively tediously re-initializing function (LSF). Though many approaches have been proposed to overcome re-initialization problem, low efficiency this still not effectively. In paper, we first investigate relationship between L1-based total variation (TV) regularizer...

DETR has set up a simple end-to-end pipeline for object detection by formulating this task as prediction problem, showing promising potential. However, despite the significant progress in improving DETR, paper identifies problem of misalignment output distribution, which prevents best-regressed samples from being assigned with high confidence, hindering model's accuracy. We propose metric, recall samples, to quantitively evaluate problem. Observing its importance, we novel Align-DETR that...

Visualizing large-scale online social network is a challenging yet essential task. This paper presents HiMap, system that visualizes it by clustered graph via hierarchical grouping and summarization. HiMap employs novel adaptive data loading technique to accurately control the visual density of each view, along with optimized layout algorithm two kinds edge bundling methods, effectively avoid clutter commonly found in previous visualization tools. also provides an integrated suite...

Single image dehazing and denoising models can simultaneously remove haze noise with high efficiency. Here, the authors propose three variational combining celebrated dark channel prior (DCP) total variations (TV) for denoising. The firstly estimate transmission map associated depth using DCP, then design colour based on this estimation layered variation (LTV) regulariser, multichannel (MTV) (CTV) respectively. In order to improve computation efficiency of models, their fast split Bregman...

In this paper, we present a system to detect and count the number of vehicles in traffic surveillance videos based on Fast Region-based Convolutional Network (Fast R-CNN). R-CNN is state-of-the-art object detection network, which takes an entire image set proposals as input, produces bounding-box positions with probability estimates over classes output. First, fine-tune pre-trained net images captured from for accuracy improvement. Second, define series rules bounding boxes screening vehicle...

Active contour models are very popular in image segmentation. Different features such as mean gray and variance selected for different purpose. But with intensity inhomogeneities, there no segmentation using the active model. The images inhomogeneities often occurred real world especially medical images. To deal difficulties raised a new model higher-order diffusion method is proposed. With addition of gradient Laplace information, can converge to edge even inhomogeneities. Because...