Search

Nowadays, kinship verification is an attractive research area within computer vision. It significantly affects applications in the real world, such as finding missing individuals and forensics. Despite the importance of this research topic, it still faces many challenges, such as low accuracy and illumination variations. Due to the existence of different classes of feature extraction techniques, different types of information can be extracted from the input data. Moreover, the fusion power produces complementary information that can address kinship verification problems. Therefore, this paper proposes a new approach for verifying kinship by fusing features from different perspectives, including color-texture and color features in different color spaces. Besides using promising methods in the field, such as local binary pattern (LBP) and scale-invariant feature transform (SIFT), the paper utilizes other feature extraction methods, which are heterogeneous auto-similarities of characteristics (HASC), color correlogram (CC), and dense color histogram (DCH). As far as we know, these features haven't been employed before in this research area. Accordingly, the proposed approach goes into six stages: preprocessing, feature extraction, feature normalization, feature fusion, feature representation, and kinship verification. The proposed approach was evaluated on the KinFaceW-I and KinFaceW-II field standard datasets, achieving maximum accuracy of 79.54% and 90.65%, respectively. Compared with many state-of-the-art approaches, the results of the proposed approach reflect the promising achievements and encourage the authors to plan for future enhancement. [ABSTRACT FROM AUTHOR]

Manufacturing industry often uses 3D scanning technologies to inspect their products. Some of these techniques produce a point cloud that represent a section of the manufactured product. The clouds must be aligned to the model of the product in order to check its quality. Current registration methods are usually affected by dimensional problems or volumetric anomalies. This paper proposes a new method for the registration process aligning the cloud to the model in several steps. The first step is the state-of-the-art method. The second step uses the information acquired in the first one to perform a fine registration in order to not being affected by dimensional defects or little miss alignments in the previous step providing a huge improvement in the measurement of surface defects. In this paper, several techniques are proposed in order to provide a set of tools that the final users can tune to fit their needs. The whole procedure of registration can be run in real-time conditions using the sampling and caching strategies proposed. The methods proposed are tested over more than 10,000 profiles of a rail proving they successfully align the cloud and the model providing better results in the measurement of surface defects. [ABSTRACT FROM AUTHOR]

Due to the emergence and advancement of deep learning technologies, scene text detection is becoming more widespread in various fields. However, due to the complexity of distances, angles and backgrounds, the adjacent texts in images have the problem that the detection boxes are far away from the texts, i.e., a position is not accurate enough. In this paper, we propose a text detection method centered on double ResNet-based and changed channels recursive feature pyramid, which integrates ResNet50-Mish and Res2Net50-Mish, as well as using recursive feature pyramid with changed channels. Firstly, scene images are fed into ResNet50-Mish and Res2Net50-Mish of double ResNet-based, and results are passed through a weight-based addition step to generate the fused feature maps. Secondly, the processed feature maps of double ResNet-based are sent into changed channels recursive feature pyramid to obtain feature maps with enhanced feature information. Also, the relevant segmentation results are then obtained by concatenating and convoluting. Finally, the results are given to progressive scale expansion algorithm to output the location of texts in images. The proposed model is trained and tested on ICDAR15 and CTW1500 benchmark datasets. In terms of precision values, our method outperforms or is comparable to state-of-the-art methods. In particular, experimental results achieve 91.53% precision on ICDAR15 dataset and 84.89% precision on CTW-1500 dataset. [ABSTRACT FROM AUTHOR]

The implementation of 3D reconstruction for targets in the low-light-level (LLL) environment is an immediate requirement in military, aerospace and other fields related to this environment. However, in such a photon-deficient environment, the amount of available information is extremely limited; thus, leading to the 3D reconstruction task in this environment is challenging. To address this issue, an embeddable converged front- and back-end network (EC-FBNet) is proposed in this paper, it can extract sparse information from the LLL environment by aggregating multi-layer semantic, then according to the similarity of features among object parts, to calculate the global topology structure of the 3D model. For the training approach, the EC-FBNet performs the two-stage integrated training modality. We additionally construct an embedded global inferential attention module (GIAM), to distribute the association weights among the points in the model, and thus reason out the global topology structure of the 3D model. In order to acquire realistic images in the LLL environment, this study leverages the multi-pixel photon counter (MPPC) detector to capture stable photon counting images in this environment, then packages into a dataset for training by the network. In experiment, the proposed approach not only achieves results superior to the state-of-the-art approaches, but also competitive in the quality of the reconstructed model. We believe that this approach can be a useful tool for 3D reconstruction field in the LLL environment. [ABSTRACT FROM AUTHOR]

In recent years, deep metric learning and its probabilistic extensions claimed state-of-the-art results in the face verification task. Despite improvements in face verification, probabilistic methods received little attention in the research community and practical applications. Previous metric learning benchmarks avoided comparison with probabilistic methods, and it is still unclear whether they generalize well to image retrieval tasks beyond faces. In this paper, we, for the first time, perform an in-depth analysis and unified comparison of known probabilistic methods in verification and retrieval tasks. We study different design choices and argue that many of them have limited impact, while only a few probabilistic methods surpass modern metric learning approaches. Combining the best ideas of earlier works, we propose a simple modification of the top-performing method, achieving new state-of-the-art results among probabilistic methods. Finally, we study confidence prediction and show that it correlates with data quality, but contains little information about prediction error probability. We thus provide a new confidence evaluation benchmark and establish a baseline for future confidence prediction research. PyTorch implementation is publicly released. [ABSTRACT FROM AUTHOR]

Image generation has always been one of the important research directions in the field of computer vision. It has rich applications in virtual reality, image design, and video synthesis. Our experiments proved that the proposed multi-style image generative network can efficiently generate high-quality images with different artistic styles based on the semantic images. Compared with the current state-of-the-art methods, the result generation speed of our proposed method is the fastest. In this paper, we focus on implementing arbitrary style transfer based on semantic images with high resolution ( 512 × 1024 ). We propose a new multi-channel generative adversarial network which uses fewer parameters to generate multi-style images. The network framework consists of a content feature extraction network, a style feature extraction network, and a content-stylistic feature fusion network. Our qualitative experiments show that the proposed multi-style image generation network can efficiently generate semantic-based, high-quality images with multiple artistic styles and with greater clarity and richer details. We adopt a user preference study, and the results show that the results generated by our method are more popular. Our speed study shows that our proposed method has the fastest result generation speed compared to the current state-of-the-art methods. We publicly release the source code of our project, which can be accessed at https://github.com/JuanMaoHSQ/Multi-style-image-generation-based-on-semantic-image. [ABSTRACT FROM AUTHOR]

Zero-shot sketch-based image retrieval (ZS-SBIR) has attracted increasing attention in computer vision, with the aim of searching the natural images that match a given sketch under the setting of zero-shot learning. In this paper, we propose a novel ZS-SBIR method based on the specifically developed domain-aware dual attention (DADA) module and similarity loss, which utilizes prior knowledge more fully than previous work. The DADA module emphasizes the importance of different channels and spaces according to the prior knowledge of whether the input is a sketch or a natural image. To preserve more prior knowledge from the models pre-trained on ImageNet, a similarity loss function is designed to supervise inter-class similarity. Furthermore, the center loss is introduced into our model, which can bridge the representation gap between the sketch domain and the natural image domain. We perform experiments on the TU-Berlin extended dataset and the Sketchy extended dataset, and the corresponding results verify the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

3D point cloud registration is a basic task in computer vision. In recent years, various of learning-based methods have been proposed to solve this problem. These methods effectively overcome the original problem of over-reliance on initial conditions and enhance the ability of obtaining the corresponding relationship. However, few methods pay enough attention to local features and tend to cause some mismatches. Therefore, this paper proposes two networks to extract local features sufficiently. To obtain a more accurate correspondence relationship between the point clouds, we propose a feature weight allocation network (FWANet), in which the expression ability of feature is enhanced using the proposed significant feature extraction module. Besides that, we utilize an interference elimination module to remove the interference points and enhance the internal correlation of point clouds. We also propose a spatial structural generation network (SSGNet), which fully utilizes the spatial location information to determine the spatial correspondence and generate a reliable connection after concatenating multi-dimensional features. At last, a complete feature space can be effectively captured after combining our FWANet with SSGNet together. We conducted multiple experiments on ModelNet40 datasets and achieved excellent results. Experimental results on four types of data demonstrate the superiority of our algorithm against the state-of-the-art ones. Our code will be available at https://github.com/liu-zikang/registration as soon as the paper is accepted. [ABSTRACT FROM AUTHOR]

The detection of surface defects is crucial to industrial manufacturing. In recent years, numerous detection methods based on computer vision have been successfully applied in the industry. However, industrial defect detection is still full of challenges. In one aspect, most of the industrial defects are extremely small. In another aspect, even though the intra-class defects have numerous similar elements, their outward appearances differ significantly. In this paper, we propose a dual-structure attention-based multi-level feature fusion network (DaMFFN) to address these two issues. In the first attention-based multi-level feature extraction structure, we introduce novel attention pooling to preserve more detailed information about the defective features of the tiny defect by giving certain regions varying weights. In the second attention-based multi-level feature fusion structure, we propose channel attention to capture the defect feature with the greatest potential for discrimination rather than all possible defect features, which is employed to prevent the incorrect detection of intra-class defects. The experiments demonstrate that the detection performance of the DaMFFN is better than other methods in five surface defect datasets. [ABSTRACT FROM AUTHOR]

Almost all Vision Transformer-based models need to pre-train on the massive datasets and costly computation. Suppose researchers do not have enough data to train a Vision Transformer-based model or do not have powerful GPUs to implement computation for millions of labeled data. In that case, Vision Transformer-based models have no advantages over CNNs. Swin Transformer is brought forward to figure out these problems by applying the shifted window-based self-attention, which has linear computational complexity. Although Swin Transformer significantly reduces computing costs and works well on mid-size datasets, it still performs not well when it trains on a small-size dataset. In this paper, we propose a hierarchical and data-efficient Transformer based on Swin Transformer, which we call ESwin Transformer. We mainly redesigned the patch embedding module and patch merging module of Swin Transformer. We merely applied some unsophisticated convolutional components to these modules, which significantly improved performance when we trained our model on a small dataset. Our empirical results show that ESwin Transformer trained on CIFAR10/CIFAR100 with no extra data for 300 epochs achieves 97.17 % / 83.78 % accuracy and performs better than Swin Transformer and DeiT in the same training time. [ABSTRACT FROM AUTHOR]

Shape recognition is an active research topic in the field of computer vision and graphic computing. Nevertheless, existing methods are still poor in accuracy and efficiency in some extent, which greatly limits their application in computer vision system. This paper investigates the restraint of feature structure that intrinsically deteriorates recognition performance. Furthermore, we propose a fast shape recognition method based on a bi-level restraint reduction of contour coding (CC2RR), which provides more effective theoretical support for the practical application of the visual algorithm. CC2RR reduces restraint performed from contour feature extraction and expression, respectively. First, for shape contour, the restraint of contour feature extraction is reduced by transforming the direction of contour points to contour segments; second, for the encoded contour segment, the restraint of the contour feature expression is reduced; in other words, the current direction is reduced to the previous and the next direction. Guided by these insights, Hamming code distance is used to match the coding features after the twofold restraint reduction, and the results are obtained. Experimental results verify that the method significantly improves the performance, which runs up to 500 times faster than the existing description methods based on shape contours while increasing robustness. This makes the method useful in practical software system. [ABSTRACT FROM AUTHOR]

Fabric defect detection is an important part of the textile industry, aiming at the problems of many types of fabric defects, small size defects and unbalanced samples, an improved YOLOv5 fabric defect detection algorithm, FD-YOLOv5, was proposed. First, the coordinate attention module is embedded in the YOLOv5 backbone network structure to replace the bottleneck structure in the original network model. While reducing the amount of parameters and calculation, it enhances the ability of the network to extract features and improves the model's ability to detect small target defects. Secondly, a smoother Mish activation function is used in the original model convolution structure for model training, which improves the nonlinear expression ability of the model; the SIoU loss function considering the direction of the anchor box is used to improve the convergence speed and detection accuracy of the model. Finally, combining the focal loss and GHM loss functions as the target confidence loss function to solve the problem of sample imbalance in the fabric defect dataset. The experimental results based on the public fabric defect dataset of Aliyun TianChi shows that the mAP@.5 and mAP@.5:.95 of the improved algorithm are 65.1% and 30.4%, respectively, which are 8.3% and 3.2% higher than the original model, respectively, and the parameter amount, calculation amount and weight of the model are reduced by 8.4%, 11.2% and 14.3%, respectively, compared with the original model. Even compared with the state-of-the-art YOLOv7 model, the mAP@.5 value of the proposed model is improved by 6.5%. Although the FPS value is lower than YOLOv7 model, it also achieves a detection speed of 79 frames per second, which can meet the real-time demand. The experimental results demonstrate the effectiveness of the method in this paper, which can provide a reference for the automatic detection method of fabric defects. [ABSTRACT FROM AUTHOR]

Edge detection is one of the most fundamental and critical operations in image analysis and computer vision. This paper proposes an adaptive edge detection method that combines multi-scale closest neighbor operator with grid partition technique (MSCNOGP). The multi-scale closest neighbor operator can be used to remove both noisy data and small area textures, while the grid partition technique can improve the precision of the edges. By utilizing the concepts of both twin edge pixel and grid divergence, the resulting edges can be further improved. Compared with prevailing traditional methods, the MSCNOGP method achieves both the best precision and almost the best visual effect, where both line segment fitting and ellipse fitting are applied for testing different edge detection methods. The performance on the F-measure score of the MSCNOGP method is much better than those of prevailing traditional methods. [ABSTRACT FROM AUTHOR]

Object detection from UAV (unmanned aerial vehicle) images is a crucial and challenging task in the field of computer vision. The task suffers from the difficulties of small dense objects, low pixel occupation of objects, and features that are not easily extracted in images. In this paper, we proposed a multilayer feature fusion algorithm named SCA-YOLO (spatial and coordinate attention enhancement YOLO) for small object detection with hybrid attention mechanisms. It uses the single-stage detection algorithm YOLOv5 as the base framework. Firstly, a hybrid attention module with associated coordinate attention is designed to enhance the feature extraction of small objects. Secondly, to address the problem that small objects are vulnerable to being disturbed by the complex background information on UAV images, an improved SEB (simple and efficient bottleneck) module is designed to further distinguish foreground and background features. Thirdly, a multilayer feature fusion structure is built to perform channel stitching of shallow and deep feature maps, as well as to enrich the semantic information of shallow features by adding horizontal jump connections. Finally, experiments are conducted on the VisDrone2020 dataset, which involves a large number of small objects photographed by drones. In addition, we also conduct extended experiments on the DOTA dataset and PASCAL VOC dataset. Comparative experimental results indicate that the proposed method considerably improves the accuracy of small object detection on multiple benchmark datasets. [ABSTRACT FROM AUTHOR]

In order to address the problem of low efficiency and accuracy of artificial visual inspection for the quality of toy sets, this paper investigates the defect detection of toy sets based on machine vision. Firstly, an improved U-Net (IU-Net) is developed by introducing the Squeeze-and-Excitation (SE) block into the U-Net and modifying its objective function. Then, a convolutional neural network (CNN) is constructed for feature extraction and defect matching. Finally, a defect detection model for toy sets is created by integrating these two models. In addition, when it comes to a new-variety toy set, the IU-Net model is trained through transfer learning with a small number of new-variety data samples. This can reduce the dependence on new-variety data samples and improve the efficiency of model training process. The experimental results show that the defect detection accuracy of the proposed method is 100% and 99.43% for the whole toy set and single component, respectively, which meets the requirements of industrial automation quality inspection of toy sets. [ABSTRACT FROM AUTHOR]

Accurately localizing bare-hands is crucial for human–computer interaction systems. In real-time systems, computational time plays a pivotal role in achieving this task within a specified timeframe. The localization process may involve detection, tracking, or both, depending on the framework's needs. Most studies on bare-hand detection and tracking have been conducted in controlled environments. However, localizing bare-hands in uncontrolled environments are challenging due to the complexity of variations, such as changes in illumination, rotation, occlusion, scale, pose, speed, and impostor bare-hands. These factors can significantly impact the performance of the models, along with background feature domination effect and motion blur, which further complicate localization. To address these challenges, this paper presents a comprehensive analysis of the most significant deep learning-based hand localization models. We have customized an object detector as a bare-hand localization model by incorporating detection and tracking modules, providing computationally efficient performance while addressing the variations. To evaluate the models' efficiency across a range of scales, we have proposed a scale-based bare-hand database with varying scales from 20 to 200 cm. We have also evaluated these models on various bare-hand benchmark databases. [ABSTRACT FROM AUTHOR]

Recovering three-dimensional structure from images is one of the important researches in computer vision. The quality of feature matching is one of the keys to obtaining more accurate results. However, as different objects or different surfaces of objects have similar images with the same elements and different typography, the camera pose estimation will be wrong and the task will fail. This paper proposes a new mismatch elimination algorithm based on global topology consistency. We first formulate the matching task as a mathematical model based on the global constraints, then convert the feature matching into grid matching, calculate the confidence of the grids according to the changes in the angle and displacement between correspondence grid vectors, and remove the mismatches with low confidence. The experiments have demonstrated that our proposed method performs better than the state-of-the-art feature matching methods to accomplish outlier match rejection in the task of similar image matching and could be helped to obtain the correct camera pose to reconstruct more complete and more accurate object models. [ABSTRACT FROM AUTHOR]

With the extensive application of machine vision in the manufacturing industry, target region recognition in complex industrial scenes is becoming a vital research territory. In the automatic polishing of molds, polishing red lead, as an auxiliary tool for polishing positioning, can intuitively determine the areas to be polished. Its bright color information are very suitable for vision-based recognition. Due to the interference of the near color in the polishing environment, the traditional color recognition method has the appearance of over-segmentation. In this paper, we propose a novel near-color recognition method via SVM based on threshold and feature transform (TFTSVM) to improve the identification accuracy of polishing red lead. Specifically, this method adopts a threshold-based color recognition algorithm to extract two kinds of color features of red lead color and its near color in HSV color space and skillfully finds it is distinguishable in three dimensions. To reduce the computational complexity, a machine learning segmentation model is constructed, which realizes dimension reduction by integrating the feature transformation method of sample transformation and projection transformation to achieve the best segmentation effect. Experimental results on self-established dataset demonstrate that the proposed method has an excellent identification effect on the red lead area in the field polishing environment and also shows good robustness under the condition that there are reflections on the mold surface. It meets the requirements of mechanical arm polishing and improves the safety and reliability of automatic polishing. In addition, we also compare different machine learning algorithms and advanced studies to verify the correctness of the algorithm. This method also provides a reference for realizing near-color recognition in complex industrial environments. [ABSTRACT FROM AUTHOR]

With the continuous development of 3D data acquisition technology in recent years, it is more and more convenient to obtain the point cloud data of large scenes, which contains a variety of rich information. How to effectively and accurately classify and segment point cloud data of large scenes has become a research hot-spot in the field of computer vision. In this paper, we study the method based on clustering, make full use of the spatial location and context information, and propose a new network framework, FeatureB2SENet. In the 2D and 3D projection feature calculation, we generate a 32 × 32 × 1 feature image for each point and input it into the convolution neural network to process the feature image. Finally, a comprehensive verification analysis is carried out on GML _ A, GML _ B and Vaihingen data sets, which proves that the classification effect is better. [ABSTRACT FROM AUTHOR]

With the gradual growth of deep learning in machine vision, efficient extraction of 3D point clouds becomes significant. The raw data of the 3D point cloud are sparse, disordered, and immersed in noise, which makes it difficult to classify and segment. Whether 3D point clouds can be classified and segmented or not, the local feature is an essential ingredient. Therefore, this paper proposes a GateNet-based PointNet++ network (G-PointNet++). G-PointNet++ extracts local features more accurately than PointNet++ by suppressing irrelevant features and emphasizing important features. Meanwhile, it refines the feature adaptively. Besides, the SENet and attention mechanism are introduced into PointNet++. G-PointNet++ was evaluated on the public ModelNet dataset, ShapeNet dataset, and S3DIS dataset, and its effectiveness in classification and segmentation tasks was verified. In the classification task, G-PointNet++ achieves an overall classification accuracy (OA) of 95.5% on ModelNet10 and 93.3% on ModelNet40. In the segmentation task, the mIoU of G-PointNet++ reaches 85.5% on ShapeNet. These experiments show that G-PointNet++ achieves better performance and saves more time than PointNet++, and its overall accuracy is higher than that of PCT network on ModeNet40. [ABSTRACT FROM AUTHOR]

Rice which is a staple food crop in most Asian countries mainly suffers from higher yield loss due to different factors, and one of the common factors that affect rice yield is rice leaf disease. Many countries do not allow the plantation of new rice varieties unless they are disease-resistant. Hence, disease control serves as an important factor for rice production and this can be effectively accomplished by identifying the rice disease at an early stage which is required for timely pesticide application and disease control. The rice leaf diseases are mainly inspected by the farmers with their naked eye which is error-prone, time-consuming, and often leads to huge yield loss if predicted wrong. This process is effectively handled by different researchers via the use of computer vision and machine learning techniques. To overcome the drawbacks associated with manual processing and low recognition accuracy of different statistical and machine learning techniques, this paper presents a novel concept that uses 3D 2D CNN for feature extraction and an improved backtracking search (IBS) algorithm-optimized deep generative adversarial network (DGAN) for classification. The 3D2D deep convolutional neural network (DCNN) is formed by integrating the 3D fast-learning block with a 2DCNN for extracting the rice disease features such as lesions and shape. The IBS algorithm-optimized GAN architecture is used for rice disease classification and the IBS algorithm is mainly adopted to overcome the instability and overfitting issues that arise in the DGAN. The proposed IBS-optimized GAN architecture offers improved accuracy of 98.7% which is relatively higher than the existing techniques such as extreme gradient boosting (XGBoost), transfer learning techniques, and support vector machine (SVM). [ABSTRACT FROM AUTHOR]

This paper is on the concept of quantum steering and quantum entanglement of two observers. The concept is applied to a multi-view computer vision system that incorporates two cameras. Three separate multi-view static camera setups are used to compare the recognition accuracy and to improve the arrangement of cameras in gesture recognition and classroom organisation applications. Prominent features which are partially hidden from a viewpoint can increase performance if given attention. In view of that, this paper proposes an entanglement-based ranking technique that updates the weights of attentive features to improve the classification rate. Principles of entanglement are also used to optimise the position of cameras such that the authoritative hidden features are visible. The proposed technique has a high recognition rate with static cameras. It also shows a low error rate in the field of view when switching to other applications. The results are validated with derangement and Bland–Altman agreement test. The entanglement approach for determining the fine-tuned position of static cameras in a recognition task outperforms many active camera networks. [ABSTRACT FROM AUTHOR]

Near-infrared (NIR) imaging can add very useful data to many visible range image processing applications. In this paper, new fusion techniques are proposed to benefit from the data of both NIR/RGB sensors for the application of scene recognition and classification. Scene recognition and classification is an important and challenging branch of computer vision. Also, image fusion is a very well-known method in image processing. In this paper, fusion of RGB and NIR images is applied to improve the performance of scene classification. The proposed fusion technique is based on modified visual salient points (MVSP). Within this process, each RGB channel is fused with NIR channel based on their visual saliency maps. The fusion parameter space is then searched for the best fusion, based on an error function that is defined to measure the degree of conformity of the fused image with the input channels. This error function is defined on the basis of the distance between pixel intensities of NIR channel and fused image. On the contrary, the gradient difference is applied for the distance between RGB channels and fused image in the error function. After finding the most consent fused image, it is applied to deep convolutional neural networks (DCNNs) to perform scene category classification using the transfer learning method. The experimental results show that the proposed method in the scene classification outperforms the prior works. The results thus illustrate how an optimized NIR/RGB fusion can bring about better classification outcomes. [ABSTRACT FROM AUTHOR]

Color images are essential in computer vision applications such as surveillance and security systems. However, due to the low light conditions, the color of the obtained image deviates from the original color. Moreover, it would produce inaccurate results that limit the performance of these applications. Therefore, this paper proposes a hue preserving uniform illumination image enhancement via triangle similarity criterion in hue–saturation–intensity (HSI) color space. Based on this, the proposed method develops translation and scaling operations to enhance the intensity and saturation. These enhancement processes maintain the hue and features while having minimal effect on the mean brightness. The proposed method and five model-based enhancement methods are evaluated from five perspectives: subjective visual evaluation, the hue and features preservation capabilities, the ability to enhance the contrast and visual information, mean brightness preservation capability, and computational complexity. Results indicate that the proposed method produces better-enhanced images than the other five model-based enhancement methods. [ABSTRACT FROM AUTHOR]

Images captured in bad weather conditions such as fog, mist, haze, etc., are severely degraded due to the scattering of the particles in the atmosphere. These images are inappropriate for various applications of computer vision, e.g., transportation, remote sensing, video surveillance object recognition, etc. Image dehazing is the process of removing the haze effect from an image so that these applications can be benefited. The physical model of haze formation is used to restore a hazy image which requires two parameters to estimate: transmission and airlight. The accuracy of the dehazing depends on the estimation of the transmission. Dark channel prior (DCP) is an effective method to compute the transmission. However, a dark channel underestimates the transmission when an object in the scene has a similar color to the atmospheric light or sky region, as a result, the dehazed image looks dark. In this paper, we explore the DCP from a new perspective and reformulate it into contrast, saturation and brightness. We proposed a method to estimate the transmission without computing the dark channel. To overcome the problem of over-enhancement and remove the haze effect, a nonlinear model based on inverse strategy is introduced. It prevents the transmission from becoming over-estimated or under-estimated. The experimental result section demonstrates the efficacy of the proposed method over the natural and synthetic hazy images along with qualitative and quantitative analysis. [ABSTRACT FROM AUTHOR]

In object detection, deciding whether each anchor box should be assigned as a positive or negative sample is a very important procedure, known as label assignment, which greatly influences the performance of detectors. Current detectors use the Intersection-of-Union (IoU) as the criterion for assigning labels (0 for negative and 1 for positive) to each anchor box while ignoring the importance of classification scores for defining samples. In this paper, we propose a novel label assignment strategy that directly assigns all ground-truths to corresponding anchor boxes based on their classification scores and determines the category belongs to each anchor box. Simultaneously, taking into account the classification accuracy and localization precision, we design a branch alignment module that enables each branch to acquire information from others as an additional supervisory signal based on the loss of different branches. Extensive competitive experiments on MS COCO benchmark demonstrate the effectiveness of our approach, which has a significant effect on the improvement of the baseline. [ABSTRACT FROM AUTHOR]

Automatic facial expression recognition plays a crucial role in computer vision, and pattern recognition. However, most existing deep learning-based facial expression classifiers usually obtain high average accuracy but have poor recognition accuracy for difficult expressions, like fear and disgust. In this paper, we propose a novel end-to-end architecture termed two-stream inter-class variation enhancement network, which learns the high-level semantic features and subtle inter-class variations in a joint fashion. More precisely, the global feature extraction network is used to extract spatial-channel semantic features, and the variations between different expressions are modeled by a distinction-reinforced network. The outputs of these two streams are weighted integrated in the expression classification network. In addition, a class balanced-weighted cross-entropy loss is designed to further improve feature discrimination. Experiment results indicate that the proposed network can significantly improve the recognition of difficult expressions and achieve a satisfactory average recognition accuracy of 73.67% on FER2013, 86.17% on RAFDB, 98.19% on CK+, and 98.85% on Oulu-CASIA, which outperforms the other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

3D point set reconstruction is an important and challenging 3D shape analysis task. Current state-of-the-art algorithms for 3D point set reconstruction employ a deep neural network (DNN) having an encoder–decoder architecture. Recently, the decoder DNNs that transform multiple 2D planar patches to reconstruct a 3D shape have seen some success. These "patch-folding" decoders are adept at approximating smooth surfaces in 3D objects. However, 3D point sets generated by these decoders often lack local geometrical details, as 2D planar patches tend to overly constrain the patch folding process. In this paper, we propose a novel decoder DNN for 3D point sets called Hyperplane Mixing and Folding Net (HMF-Net). HMF-Net uses less constrained hyperplane, not 2D plane, patches as its input to the folding process. HMF-Net has, as its core building block, a stack of token-mixing layers to effectively learn global consistency among the hyperplane patches. In addition to HMF-Net, we also propose a novel loss for 3D point set reconstruction called Weighted Chamfer Distance (WCD). WCD tries to weight, or amplify, loss from parts of shape that are highly variable across training samples by emphasizing higher point-pair distance values between a generated point set and a groundtruth point set. This helps the decoder DNN learn shape details better. We comprehensively evaluate our algorithm under three 3D point set reconstruction scenarios, that are, shape completion, shape upsampling, and shape reconstruction from 2D images. Experimental results demonstrate that our algorithm yields accuracies higher than the existing algorithms for 3D point set reconstruction. [ABSTRACT FROM AUTHOR]

Multi-modality image fusion refers to integrating series of images acquired from different sensors and obtaining a fused image which is expected to provide more comprehensive information. It plays a pivotal role in many computer vision tasks and promotes the performance of subsequent applications. Most existing approaches attempted to design appropriate fusion rules for specific image fusion task, which have a bad generalization to different fusion tasks. Apart from that, the texture details in the fused images are common blurred due to undesirable artifacts introduced from the different modalities. In this paper, we propose a generic multimodal image fusion framework by combing the visual saliency model and flexible bilevel paradigm. Specifically, we decompose input images into an intensity layer, representing large-scale intensity variations, and a detail layer, containing small textures changes. Then we fuse the intensity layer through visual saliency map to improve the contrast of an image under consideration, and design a bilevel paradigm for fusing the detail layer to obtain fine details. Furthermore, to make the fused result visual friendly, a deep prior is built in the bilevel paradigm. Besides, an elastic target-guided hyper-parameter is introduced to dominate the proportion of the textural details from the source images, which can be further adjusted in accordance with different fusion tasks. We conducted the experiments on three available datasets to demonstrate the superiority of our framework against the state-of-the-art methods quantitatively and qualitatively in a variety of fusion tasks, including infrared and visible image, near-infrared and visible image fusion and multimodal medical image fusion. [ABSTRACT FROM AUTHOR]

Multi-scale feature fusion is an important part of modern network architectures to extract more comprehensive information for most computer vision tasks, such as semantic segmentation and keypoint estimation. However, most existing multi-scale methods add fusion connections between layers or branches directly, which inevitably ignores the semantic information discrepancy between feature maps with different resolutions and depths. Moreover, inappropriate fusion connections may lead to the loss of channel-wise and spatial information. In this paper, we propose a method to enhance and refine multi-scale feature fusion for human pose estimation by employing two attention mechanisms. Specifically, we present a novel multi-head spatial attention (MHSA), which is employed to model context information of the intermediate feature maps and reinforce important local features. Meanwhile, we utilize the position channel attention (PCA) to capture long-range dependencies while retaining the important position information in the attention maps. Combining with the modules of MHSA and PCA, we design an enhanced multi-scale feature fusion network (EMF-HRNet) based on the high-resolution network (HRNet). Our proposed EMF-HRNet yields better results with repeated multi-scale information exchange and feature fusion units. Extensive experiments on two common benchmarks, COCO Keypoint dataset and MPII Human Pose dataset, show that our method significantly improves the performance of state-of-the-art pose estimation methods. [ABSTRACT FROM AUTHOR]

Hand gesture recognition is an important research field in computer vision. To effectively solve the problem of low hand gesture recognition accuracy, we propose two modules by using atrous convolution in this paper. One is Multi-Scale Fusion (MSF) module. The other is Light-Weight Multi-Scale (LWMS) module. The MSF module can be used for extracting multi-scale features at different receptive fields. The LWMS module can be considered as a kind of enhanced and expanded convolutional operation. Based on the two modules, a Hand Gesture Recognition Approach called HGRA is designed. HGRA is a hand gesture recognition approach which is based on an end-to-end CNN-based framework with two branches. One branch uses the U-Net combined with Multi-Scale Attention module to perform hand gesture segmentation in order to separate hand gestures from complex backgrounds. Then the segmentation result is used for extracting shape features. The other branch extracts visual features, such as appearance and color. The shape and the visual features obtained by the two branches are integrated to perform hand gesture recognition. Experimental results on the OUHANDS and HGR1 gesture datasets show that the proposed method has competitive performance both in hand gesture segmentation and recognition. [ABSTRACT FROM AUTHOR]

Moving object tracking is one of the key issues in the domain of computer vision. A variety of challenges are posed while tracking the object in the real-world scenario. In this paper, we have proposed a particle filtering-based algorithm to track the moving object in a complex real-world environment having shadows, dynamic entities in the background, bad weather condition and illumination variation. Specifically, we have attempted to have an effective target and scene model by adhering to the notion of feature fusion. In this regard, a novel feature fusion strategy is proposed which is scene dependent. Besides, for an efficient target model, the template for tracking is partitioned with a view to capture the local attributes of the object. In every partition, two features, namely the local binary pattern (LBP) and the mean RGB color features, are fused in a probabilistic framework. The weights for the probabilistic fusion are computed based on the local scene dynamics. The feature that strongly favors the scene in a given template partition is assigned more weightage and vice versa in the fusion process. The combined feature over all the template partitions is used to model the target. The particles that represent the state of the object evolve through a dynamic state model to track the object. The proposed tracking algorithm is successfully tested on videos considered from LASIESTA, CDnet, and DAVIS 2016 data base and it showed improved tracking accuracy as compared to existing algorithms. [ABSTRACT FROM AUTHOR]

Pose-invariant facial expression recognition is one of the popular research directions within the field of computer vision, but pose variant usually change the facial appearance significantly, making the recognition results unstable from different perspectives. In this paper, a novel deep learning method, namely, soft thresholding squeeze-and-excitation (ST-SE) block, was proposed to extract salient features of different channels for pose-invariant FER. For the purpose of adapting to different pose-invariant facial images better, global average pooling (GAP) operation was adopted to compute the average value of each channel of the feature map. To enhance the representational power of the network, Squeeze-and-Excitation (SE) block was embedded into the nonlinear transformation layer to filter out the redundant feature information. To further shrink the significant features, the absolute values of GAP and SE were multiplied to calculate the threshold suitable for the current view. And the developed ST-SE block was inserted into ResNet50 for the evaluation of recognition performance. In this study, extensive experiments on four pose-invariant datasets were carried out, i.e., BU-3DFE, Multi-PIE, Pose-RAF-DB and Pose-AffectNet, and the influences of different environments, poses and intensities on expression recognition were specifically analyzed. The experimental results demonstrate the feasibility and effectiveness of our method. [ABSTRACT FROM AUTHOR]

Scene parsing is one of the fundamental tasks in computer vision. Humans tend to perceive a scene in a hierarchical manner, i.e., first identifying the coarse category (e.g., vehicle) of a group of objects and then the fine category (e.g., bicycle, truck or car) of each of them. Despite recent tremendous progress on scene parsing, such a hierarchical semantics prior (HSP) has not been explicitly exploited. In this paper, we aim to introduce the HSP into scene parsing, by proposing a hierarchical semantics network (HSNet). Our key contribution is a bidirectional cross-level feature matching framework, which enables us to learn multi-level, hierarchy-aware features via forward feature transfer and backward feature regularization. In the forward stage, we train a coarse-to-fine module to learn fine-category features that explicitly encode hierarchical semantics information. In the backward stage, we introduce a fine-to-coarse module to collapse fine-category features to coarse-category features that are used to regularize the feature learning of our network. Experimental results on Cityscapes and Pascal Context show that our method achieves state-of-the-art performances. Our visualization also shows that our learned features capture semantic hierarchy favorably. [ABSTRACT FROM AUTHOR]

Recognition of unseen object class by a human being is always based on the relationship between seen and unseen classes, given that human has some background knowledge of the unseen object class. Zero-shot learning is a learning paradigm that tries to develop a recognition model to recognize mutually exclusive training and testing classes. A zero-shot learning model trained on labeled data can also recognize unseen classes when sufficient information about the relationship between seen and unseen classes is given. Semantic space contains semantic information about seen and unseen classes. It is an important part of zero-shot learning and acts as a bridge between seen and unseen classes. In this article, we provide a compact and comprehensive survey on zero-shot learning. First, we explain the different ways to construct semantic space along with its pros and cons. Next, we present a categorization of zero-shot learning methods from the semantic space construction point of view. Furthermore, this paper also presents performance evaluation measures with a relevant and influential zero-shot learning database. [ABSTRACT FROM AUTHOR]

Human pose estimation is a fundamental yet challenging task in computer vision. Recently, with the involvement of deep neural networks, human pose estimation has made great progresses. However, existing pose estimation networks still have some difficulties in detecting small-scale keypoints and distinguishing semantic confusion keypoints. In this paper, a novel convolutional neural network named multi-scale position enhancement network is proposed to address the above two problems. First, a multi-scale adaptive fusion unit is proposed to dynamically choose and fuse features on different scales, allowing small-scale keypoints to obtain more detailed information that is beneficial for detection. Second, we discover that although appearance-similar parts are difficult to distinguish in semantics, they differ significantly in spatial location. Therefore, a position enhancement module is designed to highlight features of real joint locations while learning more discriminative features to suppress features of similar joint regions. Finally, a global context block is applied to optimize the prediction results in order to further improve the network performance. Experiments on both single- and multi-person pose estimation benchmarks illustrate that our approach yields more accurate and reliable results. [ABSTRACT FROM AUTHOR]

Image decomposition into its structure and texture components is widely used in various image processing and computer vision applications. It is challenging to extract the structure component from an image having intricate texture since it is difficult to extract the structure from the texture that shares similar color intensity or scale. The aim of this work is to smooth the texture component from the image without affecting the significant image structures and to serve the purpose a structure- aware adaptive joint bilateral texture filtering has been employed. Main contribution in this paper is the designing of the guidance image, used in joint bilateral filtering for texture smoothing. To obtain high efficiency by using the proposed method, authors designed a scale map, which provides the size of the spatial kernel at each pixel using the characteristics of the structure and texture components. The experimental section demonstrates the supremacy of the proposed method over the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Methods for (PC) quality assessment customarily perform local comparisons between corresponding points in the "degraded" and pristine PCs. These methods often compare the geometry of the degraded PC and the geometry of the reference PC. More recently, a few methods that use texture information to assess the PC quality have been proposed. In this work, we propose a full-reference Point Cloud Quality Assessment (PCQA) metric that combines both geometry and texture information to provide an estimate of the PC quality. We use a projection technique that represents PCs as 2D manifolds in the 3D space. This technique maps attributes from the PCs onto the folded 2D grid, generating a pure-texture 2D image (texture maps) that contains PC texture information. Then, we extract statistical features from these texture maps using a multi-scale rotation-invariant texture descriptor named the Dominant Rotated Local Binary Pattern (DRLBP). The texture similarity is computed by measuring the statistical differences between reference and test PCs. The geometrical similarities are computed using geometry-only distances. Finally, the texture and geometrical similarities are fused using a stacked regressor to model the PC visual quality. Experimental results show that the proposed method outperforms several state-of-the-art methods. An implementation of the metric described in this paper can be found at https://gitlab.com/gpds-unb/pc-gats-metric. [ABSTRACT FROM AUTHOR]

Recovering 3D human pose and body shape from a monocular image is a challenging task in computer vision. In this paper, we present an optimization-based algorithm and an innovative framework to reconstruct 3D human body from a single monocular image. All the evaluation tasks are established on the basis of the classic parametric 3D body model SMPL. Firstly, a new combined objective function of SMPL parameters is proposed to involve four loss terms on 2D joints, 3D joints, facial landmarks and pose priori, respectively, which increase the reliability of the evaluation results dramatically. Furthermore, we use the estimation results given by an end-to-end regression network as the initial values of the parameters, which has been proved to speed up the optimization process. The experiments on benchmark datasets Human 3.6 M, LSP and a wild dataset demonstrate that our model achieves an accurate and robust estimation of the 3D human body and outperforms the popular competing algorithms in precision and robustness. [ABSTRACT FROM AUTHOR]

Detection of surface defects in manufacturing systems is crucial for product quality. Detection of surface defects with high accuracy can prevent financial and time losses. Recently, efforts to develop high-performance automatic surface defect detection systems using computer vision and machine-learning methods have become prominent. In line with this purpose, this paper proposed a novel approach based on Depth-wise Squeeze and Excitation Block-based Efficient-Unet (DSEB-EUNet) for automatic surface defect detection. The proposed model consists of an encoder–decoder, the basic structure of the Unet architecture, and a Depth-wise Squeeze and Excitation Block added to the skip-connection of Unet. First, in the encoder part of the proposed model, low-level and high-level features were obtained by the EfficientNet network. Then, these features were transferred to the Depth-wise Squeeze and Excitation Block. The proposed DSEB based on the combination of Squeeze-Excitation and Depth-wise Separable Convolution enabled to reveal of critical information by weighting the features with a lightweight gating mechanism for surface defect detection. Besides, in the decoder part of the proposed model, the structure called Multi-level Feature Concatenated Block (MFCB) transferred the weighted features to the last layers without losing spatial detail. Finally, pixel-level defect detection was performed using the sigmoid function. The proposed model was tested using three general datasets for surface defect detection. In experimental works, the best F1-scores for MT, DAGM, and AITEX datasets using the proposed DSEB-EUNet architecture were 89.20%, 85.97%, and 90.39%, respectively. These results showed the proposed model outperforms higher performance compared to state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Accurate segmentation is the basis of object detection, computer vision and other fields. However, the complexity of images, together with the existence of noise and other image artifacts, makes image segmentation still a bottleneck. In this paper, a dynamic relatedness model is presented and an improved fuzzy clustering algorithm is proposed. Compared with traditional algorithms, the relatedness model is measured in the process of image segmentation, and can avoid the effect of inaccurate features in noisy images. With the help of the proposed relatedness, more accurate information can be adopted to enhance the results. Simulated experiments on various images demonstrate that the proposed algorithm can achieve satisfying results and is insensitive to noise of different types. [ABSTRACT FROM AUTHOR]

Crowd counting has been a growing hot topic in the computer vision community in recent years due to its extensive applications in the fields of public safety and commercial planning. However, up to now, it has been still a challenging task in realistic scenes owing to large-scale variations and complex background interference. In this paper, we have proposed an efficient end-to-end Multi-Scale Feature Fusion and Attention mechanism CNN network, named as MSFFA. The presented network consists of three parts: the front-end of the low-level feature extractor, the mid-end of the multi-scale feature fusion operator and the back-end of the density map generator. Among them, most significantly, in the mid-end, we stack three MSFF blocks with the residual connection, which on the one hand, makes the network obtain large-scale continuous variations and on the other hand, enhances the information transmission. Meanwhile, a global attention mechanism module is employed to extract effective features in complex background scenes. Our method has been evaluated on three public datasets, including ShanghaiTech, UCF-QNRF and UCF_CC_50. Experimental results show that our method outperforms some existing advanced approaches, indicating its excellent accuracy and stability. [ABSTRACT FROM AUTHOR]

Chest radiography (X-ray) is the most common diagnostic method for pulmonary disorders. A trained radiologist is required for interpreting the radiographs. But sometimes, even experienced radiologists can misinterpret the findings. This leads to the need for computer-aided detection diagnosis. For decades, researchers were automatically detecting pulmonary disorders using the traditional computer vision (CV) methods. Now the availability of large annotated datasets and computing hardware has made it possible for deep learning to dominate the area. It is now the modus operandi for feature extraction, segmentation, detection, and classification tasks in medical imaging analysis. This paper focuses on the research conducted using chest X-rays for the lung segmentation and detection/classification of pulmonary disorders on publicly available datasets. The studies performed using the Generative Adversarial Network (GAN) models for segmentation and classification on chest X-rays are also included in this study. GAN has gained the interest of the CV community as it can help with medical data scarcity. In this study, we have also included the research conducted before the popularity of deep learning models to have a clear picture of the field. Many surveys have been published, but none of them is dedicated to chest X-rays. This study will help the readers to know about the existing techniques, approaches, and their significance. [ABSTRACT FROM AUTHOR]

Graph matching (GM) is a crucial task in the fields of computer vision. It aims at finding node-to-node correspondences between two graphs. In this paper, we propose a new GM method. We combine feature and spatial location information to construct a mixture dissimilarity matrix and compensate for the deficiency that previous methods consider only feature information. The element of this matrix is used to measure the difference in a pair of nodes. First, feature dissimilarity is measured using a new shape context descriptor. Second, spatial location dissimilarity is obtained by capturing local and global information of nodes. In accordance with the mixture dissimilarity matrix, the initial correspondences are established using the Hungarian method. Lastly, our method solves two objective functions and obtains the matching results. The experimental results for three commonly used datasets verify the efficiency and advantages of our method relative to the current methods. [ABSTRACT FROM AUTHOR]

3D non-rigid shape correspondence, as an important research topic in 3D shape analysis, is useful but challenging in computer graphics, computer vision, and pattern recognition. Despite recent success of several deep neural networks for shape correspondence, those networks cannot achieve robust results on non-rigid objects due to their local deformation complexity. This paper presents a novel and efficient shape correspondence network—Shared Optimized Res-CapsuleNet (SORCNet)—that learns point features based on enhanced descriptors to solve dense correspondence between non-rigid 3D shapes. To further improve the iterative efficiency and accuracy of the model, we design an optimized residual network structure, based on the stochastic gradient descent algorithm with momentum and weight decay (SGDW). Moreover, as the convolutional neural network does not perform well when the shape has directional variance, we present a shared capsule network structure with dual routings, which correlates the hierarchical geometric relationships of the semantic parts well to extract more informative point features. We proved that the primary capsule has a greater influence on feature extraction than the routing and decoder parts. The entire network, SORCNet, is integrated and trained/tested by taking the descriptors and Laplacian eigenbases of two shapes as input. The experiments on public datasets, such as FAUST, SCAPE, TOSCA and KIDS, demonstrate the better effectiveness, accuracy, and adaptability of our method than those of the state of the art in 3D shape correspondence. [ABSTRACT FROM AUTHOR]

Semantic segmentation is an active field of computer vision. It provides semantic information for many applications. In semantic segmentation tasks, spatial information, context information, and high-level semantic information play an important role in improving segmentation accuracy. In this paper, a semantic segmentation network with multi-path structure, attention reweighting, and multi-scale encoding structure is proposed. Firstly, three parallel structures were designed, including a pyramid spatial path with a pyramid image input, a context path composed of a lightweight backbone network, and a semantic graph path composed of spatial graph convolutional layers. Secondly, a feature fusion module was designed to perform a weighted fusion of the output features of different paths based on the channel attention mechanism. Then, the semantic segmentation dataset CamVid and Cityscapes were used for network training. Finally, ablation experiments were carried out to verify the effectiveness of the proposed network components, and analyze the computational efficiency and segmentation accuracy of the model. The experimental results show that the semantic segmentation network can improve the accuracy of semantic segmentation by combining multi-scale information, high-level semantic information, and global context information while ensuring high computational efficiency. [ABSTRACT FROM AUTHOR]

YOLO, an important algorithm for target detection, is ineffective in detecting small dynamic targets. In this paper, we utilize background subtraction, which is highly sensitive to dynamic pixels, to provide YOLO with the location and features of small dynamic targets, thus reducing the missed detection rate of small targets. This method uses background subtraction and YOLO to obtain the mask and class of the target, respectively. If the target’s mask and class can be detected, the features of YOLO and Masks data module are constructed or updated using its characteristics and class. Conversely, if only the target mask is obtained, the target mask is introduced into the features of YOLO and Masks data module for similarity detection, so as to determine the target class. Finally, YOLO performs the forced detection of the target based on the coordinates of the mask with the determined class. Validated with the SBMnet dataset, the experimental results show that for dynamic targets with three different line-of-sight distances, the method proposed in this paper improves the precision by 2.3%, recall by 3.5%, and F1-score by 3.1%. [ABSTRACT FROM AUTHOR]

Image processing and computer vision on mobile devices have a wide range of applications such as digital image enhancement and augmented reality. While images acquired by cameras on mobile devices can be processed with generic image processing algorithms, there are numerous constraints and external issues that call for customized algorithms for such devices. In this paper, we survey mobile image processing and computer vision applications while highlighting these constraints and explaining how the algorithms have been modified/adapted to meet accuracy and performance demands. We hope that this paper will be a useful resource for researchers who intend to apply image processing and computer vision algorithms to real-world scenarios and applications that involve mobile devices. [ABSTRACT FROM AUTHOR]

Low-light images usually contain high noise and low contrast. This brings bad visual feelings and hinders subsequent computer vision work. At present, many algorithms have been proposed to enhance low-light images. However, the existing methods still have some problems, such as insufficient enhancement, color distortion, or overexposure. In this paper, we propose a low-light image enhancement network based on the spatial pyramid to solve the problems existing in other methods, so as to make the enhancement result closer to the normal illumination image in brightness and color. The network is divided into two parts. Firstly, the decomposition network is designed based on Retinex theory, and the image is decomposed into the illumination image and reflection image. Then, the illumination image is processed through the three convolution kernels on the spatial pyramid module to obtain three sets of features with different scales. Next, we concatenate these three groups of features together. And the concatenated features are extracted through a convolution kernel to obtain the enhanced illumination image. Finally, the enhanced illumination image and the decomposed reflection image are multiplied pixel by pixel to obtain an enhanced image. In addition, we introduce a color loss function to solve the problem of color distortion. The experimental results show that the proposed algorithm has better visual feelings than other algorithms. We also calculate the peak signal-to-noise ratio, structural similarity index and average brightness of the enhanced results of different algorithms, and the results show that the proposed algorithm performs better. [ABSTRACT FROM AUTHOR]

Moving object detection in low-luminance Images is one of the most fundamental and difficult issues in machine vision. Therefore, in this paper, deep self-adaptive network (DSA-Net) is proposed to effectively detect moving objects in low-luminance scenes. Particularly, (1) three mechanisms are developed in this joint learning framework: graph-based unsupervised feature selection, feature representations ranking, and multiple-way feature interaction. (2) Both anti-occlusion and multi-object handling module are explored simultaneously in the unified DSA-Net model. (3) A weakly fine-tuning strategy is presented, including the easiness and group curriculum term. It leverages helpful prior-knowledge to guide the learner to select confident training samples. The experimental results show that DSA-Net outperforms the state-of-the-art methods. [ABSTRACT FROM AUTHOR]