Search

Underwater Image Enhancement (UIE) addresses challenges in marine resource exploitation such as absorption, noise, and low contrast. This paper employs bibliometric methods and data from Web of Science to analyze UIE literature. The analysis reveals that research on UIE entered a rapid development stage starting from 2017. Three main areas of technical support for research: underwater image enhancement methods, deep learning-based methods, and underwater image restoration and depth estimation. Hotspots in research include enhancement, deep learning, restoration, feature extraction, and quality assessment. Drawing from the bibliometric analysis, this paper proposes a conceptual framework for UIE research and presents five research suggestions: enhancing method robustness and adaptability, improving real-time performance, integrating software and hardware, establishing underwater image benchmark datasets, and refining quality evaluation systems for underwater images. This study offers valuable references and guidance for the future exploration and development of the UIE field. • Bibliometric analysis of underwater image enhancement (UIE) literature using Web of Science data. • Three main technical areas: enhancement methods, deep learning, restoration/depth estimation. • Research hotspots: enhancement, deep learning, restoration, feature extraction, quality assessment. • Proposes conceptual framework and five suggestions for future UIE research direction. • Recommendations include enhancing robustness, real-time performance, software–hardware integration, benchmark datasets, refined quality evaluation. [ABSTRACT FROM AUTHOR]

This paper proposes an occupancy map-based low complexity motion prediction method for video-based point cloud compression (V-PCC). We propose to utilize the occupancy map, direction gradient, and regional dispersion to divide the attribute maps into static, complex, and common blocks. Then, we propose an early termination method for static blocks, an adaptive motion search range method for complex blocks, and an early inter prediction mode decision algorithm for affine motion region in common blocks. [Display omitted] • The attribute maps are divided into static, complex, and common blocks. • An early termination algorithm is proposed for static blocks. • An adaptive motion search range method is proposed for complex blocks. • An early inter prediction mode decision algorithm is proposed for common block. This paper proposes an occupancy map-based low complexity motion prediction method for video-based point cloud compression (V-PCC). Firstly, we propose to utilize the occupancy map, direction gradient, and regional dispersion to divide the attribute maps into static, complex, and common blocks. Then, we propose an early termination method for static blocks, an adaptive motion search range method for complex blocks, and an early inter prediction mode decision algorithm for affine motion regions in common blocks. Experiment results show that, in comparison to the test model category2 (TMC2) v15.0, called the anchor method, the average bitrate savings of Y, U, and V components of the proposed method achieve 24.27%, 32.64%, and 31.23% on 8i voxelized full bodies version 2 (8iVFBv2) dataset, respectively. Further, the time savings is 41.97% for attribute maps. Similarly, the proposed method also achieves consistent performance on Microsoft voxelized upper bodies (MVUB) dataset. [ABSTRACT FROM AUTHOR]

Domain adaptation has achieved great success in using labeled source domain samples to identify unlabeled target domain samples. Here, we aim to solve the open-set domain adaptation, which is different from the closed-set domain adaptation in that it contains categories in target domain that do not appear in source domain. To solve this problem, this paper proposes open-set domain adaptation model based on subdomain alignment, which uses variable weights for discriminative training of unknown samples in target domain. Aiming at the distribution differences between domains, the model aligns the distributions of the category subspaces of source and target domains, enhancing the distribution similarity within the subspaces of the same category. Through experiments on different domain adaptation datasets, the results show that the model proposed in this paper effectively improves the accuracy of open-set domain adaptation classification. [ABSTRACT FROM AUTHOR]

Using 3D reconstruction techniques within computer vision frameworks can result in more robust and accurate solutions. However, the main challenge lies in the high computation and memory resources required by such methods. To reduce the complexity of these frameworks, a practical solution is to use geometric landmarks instead of the entire image. Therefore, in this paper the problem of 3D shape recovery of a set of standard 2D landmarks, on a single human face image is faced using multi-dimensional scaling (MDS) approach to find a 3D embedding for a set of standard 2D points. Hence, MDS approach is used for the first time in this study to establish an unbiased mapping from 2D landmark space to the corresponding 3D shape space. A deep neural network learns the pairwise 3D dissimilarity among standard 2D landmarks. This scheme leads to find a symmetric dissimilarity matrix, to be fed into the MDS approach to appropriately recovering the 3D shape of corresponding 2D landmarks. In the case of complex input image formations like posedness or perspective projection causing occlusion in the input image, an autoencoder component is used in the proposed framework, as an occlusion removal part, which turns different input views of the human face into a profile view. The results of performance evaluation using variety of synthetic and real-world human face datasets, including NoW dataset, Besel Face Model (BFM), CelebA, CoMA - FLAME, and CASIA-3D, indicates the superiority of the proposed framework, despite its small number of training parameters, with the related state-of-the-art and recent 3D shape recovery of landmark methods from the literature, in terms of unbiasedness, independent projection, efficiency and accuracy. Further, ablation study is performed to find the best training scheme of the deep learning components. All codes and public data of our paper are publicly available for research purposes at https://github.com/s2kamyab/DeepMDS. [ABSTRACT FROM AUTHOR]

• Conditionally and directly skip all non-first pre-coding. • In I picture, for a CU with top-left location (x, y), if both x and y are multiples of 64, then each non-first pre-coding is skipped. • In I picture, for a CU with top-left location (x, y), if at least one of x and y is not multiples of 64 and a set of conditions is satisfied by a non-first pre-coding, then the non-first pre-coding is skipped. • In non-I picture, for CUs at any location, if a set of conditions is satisfied by a non-first pre-coding, then the non-first pre-coding is skipped. • Direct Non-First Pre-coding Skip (DNFPS) significantly reduce the coding complexity with negligible impact on coding efficiency. The third generation of Audio Video Coding Standard (AVS3) adopts sophisticated block partitioning structure with a multi-type tree, which includes Quad-Tree (QT), Binary-Tree (BT), and Extend Quad-Tree (EQT). During multi-type tree recursive block partitioning and pre-coding process of encoding, each Coding Unit (CU) of each Size at each Location (CUeSeL) of the picture is repeatedly pre-coded many times when visiting each branch of each tree. Although there are early termination mechanism, the number of visits for each CUeSeL is still very high and up to more than fifty. This process brings very high coding complexity. In order to reduce the coding complexity of AVS3, this paper proposes a method named Direct Non-First Pre-coding Skip (DNFPS) for a CUeSeL to directly and completely skip all non-first pre-coding under specific conditions. The conditions include, but are not limited to, those related to the first pre-coding. The contributions of the paper are as follows. In I pictures, for a CUeSeL with top-left location (x, y), if both x and y are multiples of 64 or if a set of conditions related to first pre-coding is satisfied, then each non-first pre-coding is skipped, and the coding result of the first pre-coding is reused for each non-first pre-coding skipped. In non-I pictures, each non-first pre-coding is skipped, and the coding result of the first pre-coding is reused for each non-first pre-coding skipped, provided a set of conditions related to first pre-coding is satisfied. The experimental results demonstrate that DNFPS has a negligible impact on coding efficiency under All Intra (AI) configurations, and coding runtime is reduced by 16 %. [ABSTRACT FROM AUTHOR]

• An adaptive HEVC video steganography algorithm is designed, which can effectively resist the PU shift-based steganalysis. • The introduction of multivariate STC allows for smaller modifications with the same embedding capacity. • The superiority of this algorithm over the state-of-the-art video steganography algorithm is verified through extensive experiments. Calibration-based steganalysis methods are a fatal threat to current HEVC steganography algorithms. The current steganography algorithms based on prediction unit (PU) still lack resistance to this type of steganalysis. The reason is that the improvements in HEVC codec provide more potential steganographic space as well as introduce more detectable distortion. Therefore, a HEVC video steganography method that has resistance to PU shift-based steganalysis is proposed in this paper. First, the PU shift phenomenon that exists in PU-based steganography is introduced. Then the existing calibration-based steganalysis method is enhanced. According to the encoding rules of HEVC, the distortion cost function according to the difference of rate distortion cost (RD cost) during recompression is designed, which aims to obtain better resistance to calibration-based steganalysis. Experimental results demonstrate that the steganography algorithm designed in this paper provides excellent resistance ability for calibration-based steganalysis. [ABSTRACT FROM AUTHOR]

• Proposing a shift-insensitive perceptual feature based on GM and LoG signals, and theoretically selecting a ratio parameter to balance them. • Building FR-IQA models based on the proposed QGL feature maps that achieve good performance on three benchmark databases. • Achieving the best shift-insensitive performance on images with spatial changes compared with other metrics. • Combining QGL with deep neural networks and validating that QGL can help improving the accuracy in texture classification. Most existing full-reference (FR) Image quality assessment (IQA) models work in the premise of that the two images should be well registered. Shifting an image would lead to an inaccurate evaluation of image quality, because small spatial shifts are far less noticeable than structural distortion for human observers. To this regard, we propose to study an IQA feature that is shift-insensitive to the basic primitive structure of images, i.e., image edge. According to previous studies, the image gradient magnitude (GM) and the Laplacian of Gaussian (LoG) operator that depict the edge profiles of natural images are highly efficient structural features in IQA tasks. In this paper, we find that the Quadratic sum of the normalized GM and the LoG signals (QGL) has excellent shift-insensitive property in representing image edges after theoretically solving the selection problem of a ratio parameter to balance the GM and LoG signals. Based on the proposed QGL feature, two FR-IQA models can be built directly by measuring the similarity map with mean and standard deviation pooling strategies, named mQGL and sQGL, respectively. Experimental results show that the proposed sQGL and mQGL work robustly on four benchmark IQA databases, and QGL-based models show great shift-insensitive property to spatial translation and image rotation while judging the image quality. In addition, we explore the feasibility of combining QGL feature with deep neural networks, and verify that it can help to promote image pattern recognition in texture classification tasks. [ABSTRACT FROM AUTHOR]

Images taken in low-light conditions frequently encounter visibility problems, such as severe noise, reduced brightness, and low contrast. This paper introduces an approach to enhance low-light images using the Metropolis Theorem (MT). The method begins by applying a global gamma correction to the input image, followed by transforming the globally corrected image into the HSV (Hue, Saturation, Value - V) domain. To achieve multi-scale decomposition, an application of the MT is proposed, resulting in approximation and detail sub-images of the V component. Subsequently, local gamma correction is employed on both the final approximation and detail images to enhance local contrast. The refined approximation and detail images are then combined to reconstruct the refined V component. The reconstructed image is obtained by weighting each band of the image with the refined V component. Experimental results demonstrate that the proposed method outperforms state-of-the-art methods, providing improved visual quality and more natural colors. [ABSTRACT FROM AUTHOR]

• A unified adversarial attack framework is designed to reveal the security of the state-of-the-art image tampering localization algorithms. • Two effective attack methods are proposed by relying on the optimization-based and gradient-based adversarial example strategies. • Extensive experimental evaluations verify that the proposed attacks can sharply reduce the tampering localization accuracy while preserving high visual quality for the attacked images. It is significant to evaluate the security of existing digital image tampering localization algorithms in real-world applications. In this paper, we propose an adversarial attack scheme to reveal the reliability of such deep learning-based tampering localizers, which would be fooled and fail to predict altered regions correctly. Specifically, two practical adversarial example methods are presented in a unified attack framework. In the optimization-based adversarial attack, the victim image forgery is treated as the parameter to be optimized via Adam optimizer. In the gradient-based adversarial attack, the invisible perturbation yielded by Fast Gradient Sign Method (FGSM) is added to the tampered image along gradient ascent direction. The black-box attack is achieved by relying on the transferability of such adversarial examples to different localizers. Extensive experiments verify that our attacks can sharply reduce the tampering localization accuracy while preserving high visual quality for attacked images. Source code is available at https://github.com/multimediaFor/AttackITL. [ABSTRACT FROM AUTHOR]

Despite the widespread availability of HDR (High Dynamic Range) display devices, the majority of video sources are still stored in SDR (Standard Dynamic Range) format, leading to an urgent need for UHD (Ultra High Definition) video reconstruction. A simple solution is to divide the task into two sub-tasks: video super-resolution (SR) and video Bit-Depth Enhancement (BDE). While some joint enhancement tasks led by SR have been explored in multiple dimensions, the quantitative dimension has often been overlooked. In this paper, we address the first joint task of BDE and SR and propose a succinct network called MSTG. We conduct a systematic analysis to explain how this effect can be achieved and overcome the core challenge arising from distortions exacerbated by the differing optimization directions of BDE and SR. This challenge is attributed to indistinguishable contours and detailed textures. To tackle this, we employ a dual-branch module that leverages both gradient and image information to discern between BDE and SR. Furthermore, we propose a novel model named Cross-scale Transformer (CTF) embedded with Cross-scale Attention (CSA). This model facilitates information aggregation while adaptively utilizing structural similarity features. It jointly extracts, aligns, and fuses frame features at multiple scales, forming the MST module. Experiments demonstrate that the proposed algorithm outperforms the cascaded video enhancement methods by large margins on both synthetic and realistic datasets. [ABSTRACT FROM AUTHOR]

Since the camouflaged objects share very similar colors and textures with the surroundings, there is still a great challenge in accurately locating and segmenting target objects with the varying sizes and shapes in different scenes. In this paper, we propose a novel Attention Fusion and Interaction network (AFINet) to detect the camouflaged objects by exploring the cross-level complementary information. Specifically, we first propose a Multi-Attention Interaction (MAI) module to fuse the cross-level features containing different characteristics by the attention interaction, thereby fully making use of the specific and complementary information from different levels to deal with scale variation. Furthermore, we design a Location and Boundary Guidance (LBG) module to make each side-output feature aware of where to learn, which can avoid the disturbances of the non-camouflaged regions by distinguishing the subtle differences. Comprehensive experiments and comparisons are conducted on four widely used benchmark datasets, demonstrating that the proposed network achieves state-of-the-art performance. The code and prediction maps will be available at https://github.com/ZhangQing0329/AFINet. • We propose a Attention Fusion and Interaction Network (AFINet) for COD. • We design a MAI module to fully utilize the multi-level attention maps. • We design a LBG module to directly fuse the global context and global details cues. • The proposed method performs competitively against the recent methods. [ABSTRACT FROM AUTHOR]

Compression method for GPU streaming of discrete light fields is proposed in this paper. Views on the scene are encoded with video codec to enable streaming in real time. Instead of using a classic scheme, all frames are encoded according to one reference frame. Any frame is decoded directly, in a random-access manner that is suitable for light-field rendering methods, where only few frames are necessary on the GPU. The proposed scheme reaches the best decoding quality/time ratio in comparison to other schemes, where all preceding frames need to be decompressed, and all-key-frame video that supports random access, but is extremely large. The proposed method solves the space-requirements and streaming-bandwidth issues using the GPU accelerated decoding, and enables incorporating light-field assets in real-time 3D simulations. Compared to existing methods, the proposal is easy to implement, does not depend on specific video format or extension and is efficient on consumer GPUs. [Display omitted] • Real-time streaming method for light field rendering. • Solves light field issues such as excessive memory usage and slow streaming. • GPU-accelerated approach using nowadays actively developed HW decoders. • Ready to be used with upcoming video codecs. • Better than existing methods in speed, quality, implementation, native support. [ABSTRACT FROM AUTHOR]

The goal of No-Reference Image Quality Assessment (NR-IQA) is to endow computers with a human-like ability to evaluate an image's quality without comparison to a reference. Current deep learning-based methods mainly work in the spatial domain to measure the quality, heavily rely on semantic information, and less on the degradation of the image itself, struggling to accurately judge the quality of an image in similar scenes. In this paper, we propose a novel degradation priors learning architecture to address the NR-IQA task by leveraging learnable degradation priors, along with semantic features. The multi-task learning strategy is introduced to ensure our model could obtain accurate degradation priors for the NR-IQA task. Extensive experiments on public benchmarks demonstrate that our approach outperforms state-of-the-art solutions. Besides we also collect an additional dataset namely ReD-1K to illustrate the superiority of our approach to judge the image quality in similar scenes. • We propose a novel architecture involving degradation priors and semantic features for NR-IQA. • A multi-task learning framework is proposed for NR-IQA, intergating semantic features and frequency domain degradation features. • We collect a new dataset namely ReD-1K, which consists of 537 pairs of degraded and non-degraded real images. [ABSTRACT FROM AUTHOR]

For text-independent writer identification(WI) tasks, this paper proposes a novel method based on semantic codebooks and two effective feature descriptors. This method requires constructing semantic codebooks consisting of various subimages and their labels. Moment features and high-order correlation functions are used to describe handwriting style, and the standard Euclidean distance function is used for distance measurement. Then, two-way analysis of variance(TW-ANOVA) is used to filter out character factors that affect identification accuracy. Finally, a combined measure obtained from multiple classifiers based on a fuzzy integral rule is used to determine the candidate writers. To verify the effectiveness of the system, experiments are performed using the Firemaker, IAM, CRUG-CN and Uyghur2016 databases. The evaluation results demonstrate better performance than most existing methods. [ABSTRACT FROM AUTHOR]

Nowadays, the use of video data for machine (VCM) tasks has become increasingly prevalent, with deep learning and computer vision requiring large volumes of video data for object detection, object tracking, and other tasks. However, the features required for machine tasks are different from those used by humans, and a new approach is needed to encode and compress video data for machine consumption. Video coding for machines has received considerable attention, with many approaches focusing on compressing features rather than the video itself. However, a key challenge in this process is repacking the features in an efficient and effective manner. This paper proposes a distance-based patch tiling and intra-block quilting method to repack feature sequences in a manner that is better suited for existing video codecs, based on statistical analysis of feature characteristics in the channel dimension. Experimental results demonstrate that our method achieves an 65.54% BD-rate gain compared to benchmark methods. This research has significant implications for improving the efficiency of video coding for machine applications, and future work could explore the use of feature dimensionality reduction and combination of neural network (NN) codec to optimize the repacking of features for compression. [Display omitted] • Video data accounts for most internet traffic, video consumed by machines outnumbers video consumed by human. • Feature compression is an effective solution for coding for Machines with advantages of high compression ratio, computational offloading and privacy protection, while maintaining task accuracy. • Repacking features into a pseudo video to fit into existing video codecs demonstrates a performance gain. • Features is repacked based on statistical characteristics in the channel dimension. • Feature similarity is calculated by Mean Squared Error or binary distance. [ABSTRACT FROM AUTHOR]

Universal Domain Adaptation (UniDA) is a technology that enables the intelligent model to transfer knowledge learned from labeled source domains to related but unlabeled target domains without any prior label set relationship. The key to UniDA lies in rejecting target domain-specific "unknown" samples to achieve domain alignment on shared classes. In this paper, we propose the M aximum O pen-set E ntropy O ptimization via uncertainty measure to adaptively reject "unknown" samples. Specifically, MOEO sets a transition region within the most easily confused class in the open-set space. We optimize the maximum open-set entropy of simple samples outside the transition region to further improve their confidence and separate difficult samples within the transition region by aggregating similar neighbors. Accordingly, a secure boundary is formed between shared samples and "unknown" samples, promoting domain alignment within shared classes. Experiments on four benchmarks show that MOEO outperforms the previous state-of-the-art significantly. [ABSTRACT FROM AUTHOR]

Recent works on Video Anomaly Detection (VAD) have made advancements in the unsupervised setting, known as Unsupervised VAD (UVAD), which brings it closer to practical applications. Unlike the classic VAD task that requires a clean training set with only normal events, UVAD aims to identify abnormal frames without any labeled normal/abnormal training data. Many existing UVAD methods employ handcrafted surrogate tasks, such as frame reconstruction, to address this challenge. However, we argue that these surrogate tasks are sub-optimal solutions, inconsistent with the essence of anomaly detection. In this paper, we propose a novel approach for UVAD that directly detects anomalies based on similarities between events in videos. Our method generates representations for events while simultaneously capturing prototypical normality patterns, and detects anomalies based on whether an event's representation matches the captured patterns. The proposed model comprises a memory module to capture normality patterns, and a representation learning network to obtain representations matching the memory module for normal events. A pseudo-label generation module as well as an anomalous event generation module for negative learning are further designed to assist the model to work under the strictly unsupervised setting. Experimental results demonstrate that the proposed method outperforms existing UVAD methods and achieves competitive performance compared with classic VAD methods. [ABSTRACT FROM AUTHOR]

The primary goal of image steganalysis is to detect subtle steganographic signals hidden within digital images. An essential aspect of digital image steganalysis involves the extraction of meaningful steganographic signal features. However, existing JPEG steganalysis networks have not delved deep enough into the research on extracting steganographic noise signals, especially in the preprocessing stage of the model. In this paper, we propose a novel approach called the D ilated B lind- S pot S teganalysis Network (DBS2Net), specifically designed for JPEG steganalysis. DBS2Net incorporates the dilated blind-spot convolutional network into the preprocessing stage to extract noise residuals, thereby enhancing detection accuracy. The blind-spot network estimates central outputs based on adjacent regions to effectively suppress semantic information and uses dilated convolutions to expand the receptive field. Additionally, we design the transition block to further enhance noise signals by mapping noise residuals extracted by the blind-spot network. In feature extraction, depth-wise separable convolutions are used to minimize the parameter count, while dual-view pooling layers compress features, resulting in improved detection performance. Experimental results validate the efficacy of this approach, and comprehensive analyses of JPEG steganographic noise support its success. In particular, DBS2Net outperforms state-of-the-art steganalysis methods, achieving superior detection accuracy with minimal model complexity. • Adapting denoising methods for steganalysis with enhancements. • Analyzing hidden information in images using blind-spot networks. • Utilizing transition blocks to map noise residuals and amplify subtle signals. • Employing depth-wise separable convolutions for noise reduction, maintaining lightweight models. • Enhancing detection accuracy with dual-view pooling layers, achieving state-of-the-art performance. [ABSTRACT FROM AUTHOR]

Unsupervised person re-identification (ReID) is a task that aims to retrieve pedestrians across different cameras from unlabeled data. Existing methods rely on clustering to generate pseudo-labels, but they are inevitably noisy. Although pseudo-label refinement approaches have been presented, the essentiality of patch contours is ignored. The tendency analysis of retrieval between global and patch features has not been well investigated. In this paper, we propose a Patch-based Tendency Camera Multi-Constraint Learning (PTCML) model for unsupervised person ReID. First, to explore the tendentious retrieval of global and patch features, we design a Ranking Tendency Similarity (RTS) score by gauging the distribution discrepancy of distance changes. Second, based on RTS score, we propose a Tendency-based Mutual Complementation (TMC) loss to improve the quality of global and patch pseudo-labels. Third, to resist camera variations, we propose an Adaptive Camera Multi-Constraint (ACM) loss to optimize recognition results with camera distribution constraint and instance constraint simultaneously. Finally, numerous experiments on Market-1501 and MSMT17 demonstrate that our method can significantly surpass the state-of-the-art performance. • We design RTS score to mine the complementary tendency of increased distance between global and patch features. • Based on RTS score, TMC loss is proposed to refine the pseudo-labels. • We propose ACM loss to apply camera distribution and instance constraints simultaneously. [ABSTRACT FROM AUTHOR]

Accurate 3D human pose estimation is a challenging task due to occlusion and depth ambiguity. In this paper, we introduce a multi-hop graph transformer network designed for 2D-to-3D human pose estimation in videos by leveraging the strengths of multi-head self-attention and multi-hop graph convolutional networks with disentangled neighborhoods to capture spatio-temporal dependencies and handle long-range interactions. The proposed network architecture consists of a graph attention block composed of stacked layers of multi-head self-attention and graph convolution with learnable adjacency matrix, and a multi-hop graph convolutional block comprised of multi-hop convolutional and dilated convolutional layers. The combination of multi-head self-attention and multi-hop graph convolutional layers enables the model to capture both local and global dependencies, while the integration of dilated convolutional layers enhances the model's ability to handle spatial details required for accurate localization of the human body joints. Extensive experiments demonstrate the effectiveness and generalization ability of our model, achieving competitive performance on benchmark datasets. [ABSTRACT FROM AUTHOR]

The single image dehazing (SID) is challenging because of ill-posed characteristic, since there exist multiple possible dehazed images for a single hazy image due to multiple possible values of transmission. The SID is solved using scattering model, which requires computation of two parameters (transmission and atmospheric light) for its solution. The existing methods have presented priors and techniques to estimate transmission with limited focus on structure of the transmission. This paper proposes a lower bound on transmission using structural measure. The proposed lower bound is utilized to estimate the value of transmission, while preserving structural. Further, a quality control parameter is introduced to ensure non-negative value of the transmission for images with brighter objects than atmospheric light. The accuracy and efficiency of the proposed method is established by comparing with renowned SID methods using benchmark datasets. [ABSTRACT FROM AUTHOR]

3D hand pose estimation and shape reconstruction is to recover the hand joint points and hand mesh vertices coordinates from the image. However, existing methods usually only use the high-level semantic features extracted by the backbone network to represent the hand mesh vertex features, which leads to a single representation of the hand vertices features and cannot fully utilize the feature information extracted by the network. In this paper, we propose a method for real-time 3D reconstruction of hands from a single RGB image, which enriches the 3D semantic information of the mesh vertices through multi-feature fusion. Firstly, we regress the 2D features of mesh vertices through Integral Pose Regression (IPR) and regard them as prior information to 3D features. Then we design a Multi-Scale Sampling(MSS) module to extract multi-scale information. Finally we fuse 2D prior features, multi-scale features, and high-level semantic features extracted by backbone to represent 3D initial feature. Additionally, we propose a Multi-Root(MR) loss function to address the imbalance problem caused by a single root joint. The experimental results indicate that our network achieves competitive performance on the FreiHAND and HO-3D public datasets, achieving fast inference speed with fewer parameters. • We propose a Multi-Scale Sampling (MSS) module to extract multi-scale information. • We proposed a Multi-Root (MR) loss function to alleviate the imbalance issue of joints. • Our network achieves good performance in terms of model efficiency and accuracy. [ABSTRACT FROM AUTHOR]

JPEG steganography hides secret messages by modifying DCT coefficients in cover images to ensure the undetectability. Currently, most JPEG steganography methods rely on the additive distortion model, assigning adaptive costs to individual coefficients. However, these costs are often designed within a single domain, either DCT or decompressed domain, thereby making the hiding behavior detectable by multi-domain steganalysis features. To improve the steganography security, this paper proposes an universal Dual-Domain JPEG Additive Distortion (DDJAD) framework, which minimize additive distortion in both JPEG and decompressed domain concurrently. A measure to reduce the perceptibility of coefficient modifications within pixel blocks is first formulated and then incorporated into the traditional JPEG steganography model. Finally, the new additive distortion model is solved by fine-tuning existing costs, illustrating the universality of the proposed method. Experimental results showcase the superiority of the proposed method over previous works across various quality factors and payloads, employing typical steganalysis features. • Paradigm Shift : Novel dual-domain model tackles JPEG steganography limits. • Innovative Framework : Enhanced security via new embedding framework. • Advanced Distortion Cost : Innovative cost addresses new paradigm challenges. • Steganalysis Resistance : Scheme superior in resisting steganalysis. [ABSTRACT FROM AUTHOR]

With the help of computer-assisted systems, cartoon synthesis has become convenient and efficient by reusing existing cartoon materials. However, quality evaluation of the obtained cartoon image still relies on labor-intensive subjective judgment. This accordingly raises an urgent demand for effective quality evaluation methods to automatically select a cartoon image of high quality from a set of candidates with different parameter settings. In this paper, a new blind image quality assessment metric is developed for evaluating the perceptual quality of cartoon images by considering structure and chromatic distortions. The extracted gradient-based local structure features and multiscale chromatic statistical features are integrated into one representation for an overall perceptual quality prediction. Experimental results on two benchmark cartoon image datasets, i.e., NBU-CIQAD and HFUT-CID, indicate that the proposed metric outperforms both the state-of-the-art blind quality evaluation methods designed for natural images or synthetic images. • We introduce a blind cartoon IQA metric by extracting structure and chromatic features for quality assessment. • A gradient orientation-based method is devised for detecting decorative lines of cartoon images. • We represent chromatic changes using multiscale statistics based on invariance descriptors from various aspects. [ABSTRACT FROM AUTHOR]

Face recognition (FR) technique has become a pervasive and ubiquitous part of daily lives, from unlocking our smartphones with a glance to being scanned by surveillance cameras in various outdoor locations. When people's face photos are uploaded to the cloud for face recognition processing, they often have legitimate concerns about the privacy and security of their biometric data. A number of privacy-preserving face recognition (PPFR) frameworks have been proposed to address these issues by enabling the cloud to perform face recognition without revealing the identity or features of the face photos. However, these frameworks suffer from several limitations. They rely on computationally intensive operations that increase the cost and time of face recognition, leading to less applications in the real-world scenario. Many current frameworks support only one face recognition method and cannot be extended to different models. To overcome these challenges, in this paper, we propose a PPFR framework with high recognition accuracy based on extensible feature extraction for different application scenarios. In particular, features are extracted by a selective model, such as MobileFaceNet, ResNet-18 or ResNet-50, and encrypted by a randomness-based encryption algorithm in both face owner and user. Cloud service provider (SP) performs face recognition by comparing the Euclidean distances between features received from the above two entities. Extensive experiments verify that the proposed framework has significant advantages in terms of accuracy and efficiency. • In the proposed framework of PPFR, the accuracy of face recognition remains as high as we expect regardless of in the plaintext or encrypted domain. • The feature extraction component can be customized to meet specific requirements by selecting different baseline models, such as MobileFaceNet, ResNet-18, and ResNet-50. • The proposed PPFR solution outperforms the baseline frameworks in terms of data transmission volume, computational efficiency, and recognition accuracy. [ABSTRACT FROM AUTHOR]

Restoring image quality in low-light environments is an intriguing topic. While deep learning models have made significant strides in low-light enhancement, most models do not take into account the inherent characteristics of objects themselves. In this paper, we use the characteristics of the image itself to construct a Signal-to-Noise Ratio (SNR) map that guides the signal space variation to dynamically stretch the pixel values. Specifically, we propose a novel signal-to-noise ratio image-guided enhancement framework that uses the feature information of the original image to guide spatial variations in the image. It involves step-wise guidance for image feature fusion, gradually emphasizing high-frequency feature information within the image. Meanwhile, we introduced a texture optimization module that utilizes the feature information extracted by the feature fusion module to address the issues of overexposure and detail loss. We performed qualitative and quantitative evaluations on synthetic and real low-light image datasets to demonstrate the performance of our method. The experimental results show that our model outperforms other state-of-the-art methods (SOTA) in robust low-light enhancement, especially in processing images captured in complex scenes. • SNR map focus on noise distribution to guide model denoising. • Using a step-by-step guided feature fusion module and a texture optimization module. • Fusion of features from different images for spatially varying LLIE. • Texture optimization module emphasizes high-frequency feature information. [ABSTRACT FROM AUTHOR]

This paper investigates the generation of large field-of-view (FOV) and high-resolution (HR) panorama images for smartphones. Existing techniques like image stitching struggle to produce satisfactory results due to geometry misalignment and inconsistent appearance. To circumvent the inherent challenges of image stitching methods and generate high-quality panoramas, we treat the image stitching problem as a multiple-reference-based super-resolution problem. Specifically, one large-FOV low-resolution (LR) image and several overlapped small-FOV HR images are taken as inputs, where the LR image acts as a base and multi-view references provide rich HR information. Building on this foundation, a novel multi-view references image super-resolution framework (MVRefSR) is proposed. Within this framework, to address the residual geometric misalignment between the LR-Ref image pairs after coarse alignment, a flow-based RefSR network (FlowSRNet) is proposed, which super-resolves LR patches with corresponding HR references. To facilitate adaptive feature fusion and minimize distortion in structured regions, a fusion weight estimation module and a gradient branch are introduced in FlowSRNet. Finally, the large-FOV HR image is generated by combining these SR patches together. Furthermore, the lack of real-world RefSR datasets for smartphones is addressed by designing an innovative dataset construction pipeline. Extensive experiments demonstrate the superior performance of FlowSRNet and MVRefSR over the compared SR methods and image stitching software, which proves the effectiveness of generating panoramas from a new perspective. • A novel multi-view references image super-resolution framework (MVRefSR) is proposed to generate the large field-of-view and high-resolution panoramas from a new perspective. • Two modules, a fusion weight estimation module and a gradient branch, are adopted to enhance the performance of our reference-based super-resolution (RefSR) methods within our MVRefSR framework. • A new RefSR dataset generation pipeline and a real-world RefSR dataset are presented to address the lack of datasets and facilitate the study of real-world RefSR. • Experimental results demonstrate the superiority of our RefSR network and our framework MVRefSR, which proves the effectiveness of generating panoramas from a new perspective. [ABSTRACT FROM AUTHOR]

Gait recognition is an important biometric technology that allows for the remote collection of stakeholders' characteristics, without requiring their explicit cooperation. It has gained considerable attention in the fields of criminal investigation and intelligent security. Previous studies have shown that local gait features can enhance gait recognition performance by improving robustness to disturbances. However, global gait features also play a crucial role in gait recognition. Many researchers have utilized convolutional operations to extract global features, but these operations tend to focus on features within the receptive field, neglecting those outside of it. Therefore, the potential of global gait features has not been fully explored. In this paper, we propose a gait recognition framework based on vision transformers, aiming to enhance the extraction of global gait features. We introduce an adaptive multi-frame global feature mapping (AMGM) method to address the challenge of inconsistent feature dimensions caused by variations in the number of gait frames when fusing global and local features. We evaluate our model on the latest datasets, and the experimental results demonstrate a significant breakthrough. Notably, our model achieves state-of-the-art recognition accuracy, particularly in scenarios where subjects are wearing coats. Additionally, our model achieves remarkable improvements in recognition accuracy through training with small sample sets. [ABSTRACT FROM AUTHOR]

Seam carving is a method of resizing images based on content awareness. It can realize image retargeting while retaining the main content of the image. However, it may also be maliciously used to tamper with images, such as changing image semantic content by object removal. Therefore, seam carving detection has become important in image forensics. In this paper, a noise residual-based deep learning method is proposed to detect seam carving images. We try to learn the local noise in-consistency of images to recognize them. Firstly, a noise residual extraction segment is used to learn local noise features in the image, and a noise augmentation module is designed to enrich the features, which leverages the noise features extracted from a steganalysis rich model filter to discover the noise in-consistency between authentic and tampered regions. Then, through the feature dimensionality reduction section, the features are further learned and the size of feature maps are reduced. Finally, the output is obtained through global average pooling and a fully-connected layer. A careful testing strategy is further proposed, which greatly improves the detection performance, especially for seam carving with small scaling ratios. The experimental results demonstrate that our method achieves state-of-the-art performance at various scales, and has good robustness and generalization compared with other methods. [ABSTRACT FROM AUTHOR]

Recently, large datasets have become increasingly necessary for most deep learning tasks, however, large datasets may bring some problems, such as disk storage and huge computational expense. Dataset distillation is an emerging field that aims to synthesize a small dataset from the original dataset, then a random model trained on the distillation dataset can achieve comparable performances to the same architecture model trained on the original dataset. Matching Training Trajectories (MTT) achieves a leading performance in this field, but it needs to pre-train 200 expert models before the formal distillation process, which is called buffer process. In this paper, we propose a new method to reduce the consumed time of buffer process. Concretely, we use Mixture of Experts (MoE) to train several expert models parallelly in buffer process. The experiments show our method can achieve a speedup of up to approximately 4 ∼ 8 × in buffer process with getting comparable distillation performances. • Apply Mixture of Experts to accelerate dataset distillation. • Our method is widely effective for trajectories-matching distillation methods. • Experiments show that our method gets comparable performances. [ABSTRACT FROM AUTHOR]

• Grayscale probabilistic matrix (GPM) with multi-level grayscale pixels is introduced into the construction of secret shares. • A general model of size-invariant TiOISSS based on GPMs is proposed which has no limit on the number of participants. • The visual quality of previewing images is improvedunder the gray mixing model while that of computed image is not degraded. TiOISSS (Two-in-One Image Secret Sharing Scheme) can reconstruct the secret through computation and realize secret preview recovery when the power system is destroyed or the network is paralyzed. However, the visual quality of previewing images obtained by the existing TiOISSS is not satisfactory and usually affected by pixel expansion. To resolve the two problems, this paper introduces multiple gray values in the share image generation stage and establishes an optimization model to obtain the grayscale probabilistic matrix needed in the encoding stage. By using the probability matrix, a secret sharing algorithm is designed to obtain size-invariant shares with multiple gray values. Then, the secret information of another image can be shared and embedded into the multi-level gray shares through Lagrange polynomials over finite fields. In the decryption stage, the electronic shares can be directly superimposed based on a gray mixing model, or they can be printed to transparent films by a multi-level gray printer and then superimposed to obtain the preview image. If computational devices are available, a high-resolution gray image can be obtained by Lagrange interpolation over finite fields. Experimental results indicate that the proposed scheme is effective for generating shares without pixel expansion and improving the preview image quality. [ABSTRACT FROM AUTHOR]

Convolutional neural networks have achieved remarkable success in single image dehazing tasks, and previous studies verified the dehazing performance of the U-shaped framework. However, most existing U-shaped architecture dehazing networks still face challenges in sufficiently dealing with a large area of haze with low visibility. In this paper, we propose a novel dehazing network named Double U-Net(DU-Net). Specifically, to reduce the interference of haze features in the encoder to the recovery stage when skip-connecting to the decoder directly, we develop a new architecture firstly, which is composed of an extended encoder–decoder. Besides, the hierarchical depth-wise convolution block(HDCB) is designed to gradually increase the receptive field by leveraging the depth-wise convolution, enriching the global information. Moreover, we propose a multi-branch interactive fusion(MIF) which achieves efficient cross-branch and cross-channel interaction through parallel multiple 1D convolutions. Extensive experiments on both synthetic and real-world hazy images demonstrate the effectiveness of our proposed method. • We construct a Double U-Net for image dehazing, by extending the encoder–decoder. • We build hierarchical convolution blocks to obtain a multi-scale receptive field. • A branch fusion is proposed to enhance cross-branch and cross-channel interaction. [ABSTRACT FROM AUTHOR]

Due to the absorption and scattering of light on the water surface, underwater images often face challenges such as low contrast, color deviation, insufficient exposure, and blurred details, exacerbating the difficulty of underwater tasks. In recent years, underwater image enhancement has become increasingly crucial in marine applications. Among existing underwater enhancement methods, the focus has often been on pixel-level learning, which may lead to image noise and an inability to finely adjust the image. In this paper, we propose a dual-branch Transformer-CNN Parameter Filtering network for underwater image enhancement, referred to as DTCPF. Specifically, to better aggregate window information, we introduce an overlapping window self-attention module to enhance interaction between windows. Additionally, we employ an improved Transformer encoder and decoder, utilizing long-distance attention and reversible neural networks to extract low-frequency and high-frequency information from the image. Moreover, we introduce a regression parameter filtering group for regression prediction, using the predicted parameters to enhance the image and obtain a reliable underwater enhancement model. Our approach undergoes qualitative and quantitative analyses on four real underwater datasets, demonstrating outstanding performance. [ABSTRACT FROM AUTHOR]

The adage "a picture is worth a thousand words" resonates in the digital video domain, suggesting that a video could be seen as a composition of millions of these words. Videos are composed of countless frames. Video summarization creates cohesive visual units in scenes by condensing shots from segments. Video summarization gains prominence by condensing lengthy videos while retaining crucial content. Despite effective techniques using keyframes or keyshots in video summarization, integrating audio components is imperative. This paper focuses on integrating deep learning techniques to generate dynamic summaries enriched with audio. To address that gap, an efficient model employs audio-visual features, enriching summarization for more robust and informative video summaries. The model selects keyshots based on their significance scores, safeguarding essential content. Assigning these scores to specific video shots is a pivotal yet demanding task for video summarization. The model's evaluation occurs on benchmark datasets, TVSum and SumMe. Experimental outcomes reveal its efficacy, showcasing considerable performance enhancements. On the TVSum, SumMe datasets, an F-Score metric of 79.33% and 66.78%, respectively, is achieved, surpassing previous state-of-the-art techniques. [ABSTRACT FROM AUTHOR]

Occluded person re-identification (ReID) remains challenging due to the misaligned body parts. Existing works, mainly utilizing extra clues, excel in predicting holistic person images but falter when confronting substantial occlusion. This paper proposes a transformer-based Mask-guided Part-to-Part Matching (MP2PMatch) network for fine-grained matching. Firstly, the Consistency Occlusion Augmentation (COA) processes holistic person images and corresponding body part masks to construct occluded "image-mask" pairs. Next, we introduce learnable part tokens to capture semantic features of various body parts, performing "pull close" and "push apart" operations based on identity labels and part visibility, ensuring the one-to-one correspondence between part features and body parts. Additionally, the proposed Body Region Attention (BRA) utilizes the overall attention on body regions to guide the network to overcome interference from both occlusion and background. Extensive experiments demonstrate that MP2PMatch achieves exceptional occluded ReID performance. • Our proposed MP2PMatch aims to address the Occluded Person ReID problem. • Designing a COA to construct reliable occluded image-mask pairs. • Proposing a BRA to enhance attention within the human body region. • Proposing a PTA to guide the model to extract discriminative part features. • Experimental results demonstrate the effectiveness of our method. [ABSTRACT FROM AUTHOR]

Recent self-supervised learning methods, where instance discrimination task is a fundamental way of pretraining convolutional neural networks (CNN), excel in transfer learning performance. Even though instance discrimination task is a well suited pretraining method for classification with its image-level learning, lack of dense representation learning makes it sub-optimal for localization tasks such as object detection. In this paper, we aim to mitigate this shortcoming of instance discrimination task by extending it to jointly learn dense representations alongside image-level representations. We add a segmentation branch parallel to the image-level learning to predict class-agnostic masks, enhancing location-awareness of the representations. We show the effectiveness of our pretraining approach on localization tasks by transferring the learned representations to object detection and segmentation tasks, providing relative improvements by up to 1.7% AP on PASCAL VOC and 0.8% AP on COCO object detection, 0.8% AP on COCO instance segmentation and 3.6% mIoU on PASCAL VOC semantic segmentation respectively. • Convolutional Neural Networks lose localization info with instance discrimination task. • Augmenting instance discrimination with an extra dense task enhances localization. • Unsupervised foreground segmentation boosts localization in (non-)contrastive methods. • Improved localization in SSL enhances transfer learning in dense downstream tasks. [ABSTRACT FROM AUTHOR]

Video anomaly detection aims to identify a set of abnormal events in videos. Deep reconstruction and prediction-based models have been employed to detect anomalies. Deep reconstruction models sometimes recreate the abnormal events along with the normal ones. However, the prediction-based approaches have demonstrated encouraging results. This paper presents a video vision transformer (ViViT) based generative adversarial network (GAN), TransGANomaly, a novel approach for detecting anomalies. The proposed framework is a video frame predictor and trained only on normal video data adversarially. The generator of the GAN is a ViViT network that receives 3D input tokens from the video snippets. The generator aims to predict the future frame based on past sequences. After that, the predicted and original frames are given to the model's discriminator for binary classification. Extensive experiments have been performed on UCSD Pedestrian, CUHK Avenue, and ShanghiaTech datasets to validate the efficacy of the proposed method. [ABSTRACT FROM AUTHOR]

Thanks to the powerful modeling capabilities of multi-head self attention (MSA), transformers have shown significant performance gains in vision tasks. However, as transformers require heavy computation, more efficient designs are required. In this paper, we present an efficient transformer architecture named GoLDFormer for image restoration. GoLDFormer extends the capability of window-based self-attention through two core designs. First, We propose a globally-enhanced window-based transformer block (G-WTB), which applies transposed attention to a compressed window representation rather than the spatial features, thus establishing connections between all windows with less computational complexity. Second, since the interactions between image content and window attention weights can be interpreted as spatially varying convolution, we introduce an adaptive filter structure into transformer models and propose a deformable filtering block (DFB) to enable cross-window connections. By adjusting the shape of the generated filters in the DFB, we can balance the computational costs and the degree of adjacent window interaction. Extensive experiments on several image restoration tasks demonstrate that GoLDFormer achieves competitive results against recent methods but with optimal computational costs. • We propose GoLDFormer, including two blocks, to enhance W-MSA globally and locally. • Globally-enhanced attention block captures global dependencies between all windows. • Deformable filtering block captures local dependencies by modifying filter size. • Our GoLDFormer achieves competitive results against recent lightweight methods. [ABSTRACT FROM AUTHOR]

Deep learning methods utilizing Convolutional Neural Networks (CNNs) and Transformers have achieved remarkable success in volumetric medical image analysis. While successful, the symmetrical structure of numerous networks pays insufficient attention to the encoding phase, and the large amount of memory occupied by voxels leads to unnecessary redundancy in the network. In this paper, we present a novel approach to handle volumetric medical images by converting them into point cloud and introduce a new asymmetrical segmentation architecture. We propose a dual-path encoder that fully captures both dense and sparse representations of the input point cloud sampled from volumes. Moreover, the two obtained representations are subtracted at the skip connection as a complementary feature during the decoding stage. Experimental results on the Brain Tumor Segmentation (BraTS) and the Multi-sequence Cardiac MR Segmentation tasks demonstrate the great potential of our point-based method for volumetric medical image segmentation. [ABSTRACT FROM AUTHOR]

In this paper, an approach to the semantic segmentation of 3D LiDAR point clouds obtained from natural scenes is introduced. Using a state-of-the-art projection-based semantic segmentation model as the core segmentation network, several recent advances in projection-based 3D semantic segmentation methods are aggregated into a single model. These adaptions include: scan unfolding, soft-kNN post-processing, and multi-projection fusion. A novel Naïve Bayesian approach to multi-projection fusion which weights class probabilities based on the outputs of the base classifiers is proposed to further increase robustness. Quantitative and qualitative evaluations on several datasets, including scenes from both urban and natural environments; show that aggregating these adaptions into a single model can further improve the accuracy of state-of-the-art projection-based approaches. Finally, it is demonstrated that the novel Naïve Bayesian approach to multi-projection fusion addresses a number of the challenges inherent to natural data while also improving results on urban data. • Natural data is difficult to analyse due to its complex structures and features. • Aggregating existing adaptations can improve 3D semantic analysis. • Natural data analysis can be improved on without degrading urban data performance. • Projection-based methods are an effective means to semantic analysis of 3D data. • Fusion of multiple perspectives improves semantic analysis of natural 3D data. [ABSTRACT FROM AUTHOR]

The images in foggy weather are often degraded which may affect the object detection process. Hence, in recent times, several schemes have been designed to eliminate the haze effect and enhance the performance of computer vision systems. Although these methods reduce the haze effect, they also produce over-saturation and over-degradation on most occasions. These problems occur due to inadequate utilization of the regional haze properties during the dehazing process. Therefore, a new dehazing model is proposed in this paper to address the issues faced by existing dehazing algorithms. The residual convolutional neural network (RCNN) with 14 layers is developed in the present work. In this model, features to indicate the extent of haze are extracted using the regional characteristics of a hazy image and are used as the fourth channel along with the three colour channels of a hazy image. The proposed RCNN is trained with 4-channel hazy images as input and their corresponding haze-free images as output. The results of our model are obtained with several images taken from three different datasets. Through the experimentation, quantitative and qualitative improvements are observed when compared to state-of-the-art dehazing methods. • The local-haze characteristics driven dehazing method is developed in this work. • The regional haze properties are extracted using pixel-wise local max-min difference and pixel intensities of the hazy image. • A 4-channel hazy image is developed by concatenating the regional haze properties map to the hazy image. • A 14-layered RCNN is developed with the 4-channel hazy image as input and the haze-free image as output. [ABSTRACT FROM AUTHOR]

Remote sensing image scene classification has made great progress with deep learning. Due to complex backgrounds and the large number of objects with inhomogeneous sizes, remote sensing image scene classification still remains challenging. In fact, only a few parts of regions are expected to be representative of the scene. In this paper, we propose a dual-stream framework for remote sensing classification with the help of category-related regions. Swin Transformer-based global feature extractor is used to produce coarse classification results. A category-related key region localization module is developed to extract the representative region from the whole image. The ResNet50 is employed as a local feature extractor to mine critical features from prominent regions. Finally, the classification results is obtained with the shared weighted classify head on global-local features. Extensive experiments show that our method provides a promising approach for key region locations and performs best on three public remote sensing datasets. [ABSTRACT FROM AUTHOR]

In this paper, an eye-tracking dataset of movie viewing in the immersive environment is developed, which contains 256 movie clips with 2K QHD resolution and corresponding movie genre labels from IMDb (Internet Movie Database). The dataset provides the audio-visual clues for studying the human visual attention when watching movie using a VR headset, by recording the eye movements using integrated eye tracker. To provide benchmarks for a saliency prediction for movie viewing in the immersive environment, fifteen computational models are evaluated on the dataset, including a newly developed multi-stream audio-visual saliency prediction model based on deep neural networks, named as MSAV. Detailed quantitative and qualitative comparisons and analyses are also provided. The developed dataset and benchmarks could help to facilitate the studies of visual saliency prediction for movie viewing in the immersive environments. [ABSTRACT FROM AUTHOR]

• A new adaptive blind image watermarking technology, utilizing artificial intelligence (AI) and named AwDD, has been proposed for detecting color image deepfakes. • The AI technology used includes face detection, denoising autoencoder (DAE), and the gray wolf optimizer (GWO). • The proposed method employs mixed modulation (MM) with partly sign-altered mean value (PSAMV) to embed a set of coefficients, thereby increasing robustness against attacks while maintaining high image quality. • Blind adaptive deepfake detection (BADD) with the tamper detection mean value (TDMV) is applied to adaptively detect changes in the relative position when a face image is modified or deepfaked. • The GWO optimizes parameters to enhance imperceptibility and robustness, while the DAE improves the visual recognizability of the extracted watermark. This paper proposes a new adaptive blind watermarking technology for deepfake detection, which can embed deepfake detection information into the image and verify the image's authenticity without requiring additional information. The proposed scheme utilizes mixed modulation combined with partly sign-altered mean value to embed a set of coefficients that enhance robustness against attacks while maintaining high image quality. Additionally, blind adaptive deepfake detection technology with the tamper detection mean value is employed to detect relative positions adaptively, even when face images are slightly modified or deepfaked. To further improve the performance of the proposed scheme, a gray wolf optimizer is introduced to optimize parameters, and a denoising autoencoder is employed to facilitate the identification of extracted watermarks. This technology will adaptively embed watermark information while preserving the original face image, thereby maintaining the authenticity of the face in the image and verifying the owner of the image. The code is available at https://github.com/lyhsu01/AwDD. [ABSTRACT FROM AUTHOR]

Block compressive sensing (BCS), an emerging approach for signal acquisition and reconstruction, combines high-speed sampling and compression, making it widely applicable in various imaging tasks. However, image BCS generally face the issues: challenges in accurate sampling rate allocation (SRA) and block artifact removal, and poor reconstruction algorithms. In this paper, we propose an approximate message passing (AMP)-based BCS (AMP-BCS) method. Specifically, within the sampling module, a sparsified DCT coefficient-based permutation strategy is proposed to achieve uniform energy distribution among blocks, effectively addressing the issue of SRA. Within the reconstruction module, by reweighting shallow and deep multi-scale features using several attention mechanisms, the multi-scale deep attention network (MDANet) is proposed to improve the denoising capabilities of the AMP reconstruction. Through independent sampling and joint iterative denoising, block artifacts are substantially removed. Extensive experiments demonstrate that the AMP-BCS method significantly outperforms current state-of-the-art BCS algorithms in both visual perception and objective metrics. • A novel coefficient permutation-based AMP method for block compressed sensing. • A shallow and deep multi-scale attention Gaussian denoiser. • Impressive experimental results. [ABSTRACT FROM AUTHOR]

The correlation filter based online tracking methods usually can achieve high real-time performance due to the leverage of the well-known FFT. However, they are also apt to generate the "corrupted training samples" in scenarios with complex background, which will trigger the model drift and deteriorate the tracking accuracy rapidly. The existing methods usually consider this problem from certain aspect and none of them has mined the potential of combining multiple formulas. In this paper, we propose the Output Constraint Transfer - Discriminative Scale Space Tracking (OCT-DSST) algorithm, which has taken full consideration of multiple channel feature, multiple filters, kernel trick, memory with incremental learning, and the self-supervision mechanism. We re-formulate the online tracking process by combining all formulas above in a unified framework. The so obtained adaptive learning rate can better exploit the feedback information coming from the intermediate tracking results, and effectively mitigate the corrupted sample problem. The experimental results on the OTB-50/100 and the VOT2016 datasets reveal that the proposed method is comparative to most state-of-the-arts algorithms, and can increase the accuracy by 2% and the success rate by 1.7%, compared to the traditional DSST method. [ABSTRACT FROM AUTHOR]

Low-light conditions present a myriad of intricacies for object detection, with many existing methods relying primarily on image enhancement before detection. Sometimes, the enhancement methods are unable to improve the detection performance in low-light conditions. In this paper, we present a new Multi-exposure refinement network for low-light object detection (MRN-LOD) to avoid the need for enhancement before detection. The MRN-LOD contains: multi-exposure feature extractor, adaptive refinement network, and detection head. The developed multi-exposure feature extractor extracts features from the multi-exposure images generated by the low-light image. We introduced the notion of feature extraction from multi-exposure images for object detection in low light. In addition, we proposed an adaptive refinement network to refine the features of low-light images for better detection performance. The detection head uses the refined features to perform object detection. Extensive experimentation on real-world and synthetic datasets shows the superiority of the proposed MRN-LOD. [ABSTRACT FROM AUTHOR]

Monocular depth estimation (MDE) is a fundamental problem in computer vision. Recently, self-supervised learning (SSL) approaches have attracted significant attention due to the ability to train an MDE network without ground-truth depth data. However, the performance of most existing SSL-MDE methods is yet limited by the available real training dataset, which are either binocular stereo pairs or monocular video sequences. In this paper, we propose a simple but effective generalization of SSL framework such that collections of multiple view Internet photos, a virtually unlimited source of real data, are enabled to train an MDE network. Combining the depth consistency and the mask that alleviates the interference such as moving objects, the network benefits from the real correspondences in adjacent views, thus achieving the improvement. Experiments show that the generalization of Monodepth2 via the proposed method not only leads to superior performance than itself and some data-driven MDE methods, but also stably boosts the performance of multiple state-of-the-art SSL-MDE methods. Besides, experiments on SeasonDepth, a dataset with various environmental conditions, show the good generalization capability of our proposed method. [ABSTRACT FROM AUTHOR]

Most previous visible–infrared person re-identification methods emphasized learning modality-shared features to narrow the modality differences, while neglecting the benefits of modality-specific features for feature embedding and narrowing the modality gap. To tackle this issue, our paper designs a method based on dual attention enhancement features to use shallow and deep features simultaneously. We first convert visible images into gray images to alleviate the visual difference. Then, to close the difference between modalities by learning the modality-specific features, we design a shallow feature measurement module, in which we use a class-specific maximum mean discrepancy loss to measure the distribution difference of specific features between two modalities. Finally, we design a dual attention feature enhancement module, which aims to mine more useful context information from modality-shared features to shorter the distance between classes within modalities. Specifically, our model exceeds the current SOTAs on SYSU-MM01, with 66.61% Rank-1 accuracy and 62.86% mAP. [ABSTRACT FROM AUTHOR]

Light field imaging can capture rich information of real scenes, but various distortions will inevitably be introduced in the process of light field image processing. Therefore, it is very important to effectively evaluate the quality of light field images. Due to the lack of commercial light field displays, this paper proposes a blind quality assessment method of light field image based on view and focus stack, based on the visualization technology used in the subjective evaluation of light field images and the human brain's perception process of visual information in the study of visual physiology. First, spatiotemporal Gabor filtering is performed on distorted sub-aperture images (SAIs) to extract the spatial and angular information. On the other hand, the SAIs are refocused to obtain the refocused images with different focusing depths, and then the focus stack is filtered by spatiotemporal Gabor filtering to extract the angular information. In addition, the focus stack is detected to form in-focus area and out-of-focus area, and the features related to spatial structure, depth and semantic information of the refocused images are extracted to evaluate the spatial quality. Finally, support vector regression (SVR) is performed to predict the objective scores by establishing the relationship between the features and subjective scores. Experimental results show that our method can well predict the human eye's perception quality of light field images. [ABSTRACT FROM AUTHOR]