Capturing contextual information across adjacent slices is critical for accurate medical image segmentation. While 2D methods are computationally efficient, they often fail to model inter-slice continuity. In contrast, 3D methods capture volumetric context but require substantial computational resources. To overcome these limitations, we propose a 2.5D segmentation framework that incorporates inter-slice context while maintaining relatively high efficiency. We introduce a Dual Cross Attention (DCA) module that captures both global spatial and channel dependencies across adjacent slices. Furthermore, to enhance inter-slice consistency, we employ class-wise prototype learning, which aligns pixel embeddings of the same class across adjacent slices. Additionally, we introduce a semantic correlation loss to align DCA-refined features with decoder predictions via cosine similarity, guiding each channel to correspond more semantically with its associated class. Experiments conducted on the FLARE22 and MM-WHS datasets demonstrate that our method outperforms both 2D and 2.5D baselines, validating the effectiveness of modeling inter-slice dependencies and promoting class-level representation alignment in medical image segmentation.

Facial image editing via the latent space of generative models has recently gained significant attention, particularly for its efficiency in eliminating the need for model training. Among these methods, support vector machines (SVMs) are widely used to define semantic edit directions. However, existing methods lack clear guidelines on the selection and quantity of training data for the SVM, making the preparation process time-consuming, especially in the case of diffusion model-based approaches. In this paper, we progressively reduce the number of images and evaluate the results in terms of quality, identity preservation, and attribute consistency. Based on our findings, we propose a practical lower bound for the number of images required for effective SVM training along with criteria to ensure attribute-specific editing, thus improving editing efficiency.

Existing video action recognition methods often fail to bridge the semantic gap between fine-grained, frame-level labels and coarse, high-level actions. To address this, we propose a hierarchical reasoning framework that predicts coarse action labels from sequences of fine-grained labels. Using the Breakfast dataset, we apply rule-based preprocessing to align and pair fine and coarse labels, resolving temporal misalignment and background segments. Our model-agnostic framework supports integration with outputs from Temporal Action Segmentation (TAS) models. We evaluate sequence models—LSTM, TCN, Transformer, and Mamba—under causal and non-causal settings using multiple loss functions, including cross-entropy and cosine similarity. Results show that causal models, particularly LSTM and Mamba, outperform others in accuracy, edit distance, and F1-score, especially with hybrid losses. Our method is robust to noisy fine labels and preserves interpretability through explicit fine-to-coarse mapping. This work offers a scalable and modular solution for multi-level action understanding across diverse video domains.

Recent advances in diffusion models have positioned them as powerful generative frameworks for high-resolution image synthesis across diverse domains. The emerging "h-space" within these models, defined by bottleneck activations in the denoiser, offers promising pathways for semantic image editing similar to GAN latent spaces. However, as demand grows for content erasure and concept removal, privacy concerns highlight the need for identity disentanglement in the latent space of diffusion models. The high-dimensional latent space poses challenges for identity removal, as traversing with random or orthogonal directions often leads to semantically unvalidated regions, resulting in unrealistic outputs. To address these issues, we propose Black Hole-Driven Identity Absorption (BIA), a novel approach for identity erasure within the latent space of diffusion models. BIA uses a "black hole" metaphor, where the latent region representing a specified identity acts as an attractor, drawing in nearby latent points of surrounding identities to "wrap" the black hole. Instead of relying on random traversals for optimization, BIA employs an identity absorption mechanism by attracting and wrapping nearby validated latent points associated with other identities to achieve a vanishing effect for specified identity. Our method effectively prevents the generation of a specified identity while preserving other attributes, as validated by improved scores on identity similarity (SID), FID metrics, qualitative evaluations, and user studies as compared to SOTA.

Vertebrae localization and identification from Com-puted Tomography (CT) scans playa crucial role in the diagnosis of spine-related disorders. However, localization and labeling of vertebrae are laborious and challenging due to the complex anatomical structure of the spine, low contrast, and fuzzy bound-aries in CT scans. This study introduces an encoder-decoder-based multi-view learning approach by training the model using distinct representations (views) for vertebra identification in digitally reconstructed radiographs (DRR). Multi-view learning aims to enhance model robustness, accuracy, and generalization capabilities by leveraging information from multiple digitally acquired DRR images. To generate the DRR images, we developed a simulation environment that produces multiple DRR views from a given CT scan. We employed a contrastive learning strategy for training the backbone network to enhance the learning of global representations across these multi-views. Subsequently, we trained a localization network to detect vertebrae centroids, followed by an identification network to classify each vertebra accordingly. Moreover, we validated our model on the VerSe 2019 dataset and outperformed other state-of-the-art (SOTA) methods.

We introduce a new method for face de-identification using a frozen diffusion model. In contrast to previous methods that carefully design and train a generative model, we reformulate face de-identification as an identity editing task and employ a pre-trained unconditional diffusion model. Also, unlike previous facial image editing approaches that try to preserve the identity and change only the demanded attributes, we aim to shift the identity while preserving the rest. This approach is significantly efficient because there is no need to construct or train any part of the diffusion model for identity shift. To the best of our knowledge, this is the first work to perform face de-identification with image editing. Ultimately, our findings, supported by both qualitative and quantitative results, show that image editing can effectively achieve de-identification.

Computed tomography (CT) scans of the abdomen have become the gold standard for assessing body composition (BC). Accurate estimation of skeletal muscle and adipose tissues from CT scan slices is crucial for diagnosis and prognosis. Much research in abdominal image analysis focuses on the third lumbar vertebra (L3) due to its stability and ease of labeling compared to other lumbar vertebrae. This study leverages labeled L3 slices (source domain) to predict unlabeled slices from thoracic T1 to sacrum S5 region (target domain). We proposed a Twin Encoder–Decoder Network (TED-Net) with distinct weight initialization employing Cross-domain Consistency Learning (CDCL) for joint training across the domains. This strategy extends the network’s knowledge by enforcing consistency between predictions from two segmentation networks. The training objective includes supervised loss terms for the source domain and unsupervised loss terms for the target domain. This results in increases of 6.68%, 3.31%, and 4.40% in Precision, Dice Similarity Coefficient, and Intersection over Union, respectively, indicating significant improvement in performance on the target domain, suggesting that domain-invariant feature learning through cross-domain consistency learning enhances a network’s adaptability over unlabeled domains.

Recently, autoencoder-based hyperspectral anomaly detection methods have demonstrated excellent performance on hyperspectral images (HSIs). The autoencoder (AE) can simultaneously reconstruct both the anomaly targets and background, but the lack of prior information limits ability to detect anomalies. This study proposes a novel hyperspectral anomaly detection method based on a guided AE to reduce the feature representation for anomaly targets. First, a multi-layer AE network with skip connections is proposed to fully extract the abundant latent features from HSIs and enhance the expressive ability of the network. The reconstructed HSI can be obtained by the proposed AE network. Second, to suppress anomaly targets in the obtained reconstructed HSI and better represent background features, a guided module based on a guided image is added to the network to reduce the feature representation of anomaly targets by providing feedback information. Moreover, the guided image is calculated using a proposed spectral similarity method that uses the local spatial features of the HSI. Finally, we use the reconstruction error as a performance metric and compare the results of our proposed method with other state-of-the-art methods on six real-world HSIs. The results demonstrate the effectiveness and superiority of the proposed method.

Moving object detection (MOD) is a fundamental step in many high-level vision-based applications, such as human activity analysis, visual object tracking, autonomous vehicles, surveillance, and security. Most of the existing MOD algorithms observe performance degradation in the presence of complex scenes containing camouflage objects, shadows, dynamic backgrounds, and varying illumination conditions, and captured by static cameras. To appropriately handle these challenges, we propose a Generative Adversarial Network (GAN) based on a moving object detection algorithm, called MOD_GAN. In the proposed algorithm, scene-specific GANs are trained in an unsupervised MOD setting, thereby enabling the algorithm to learn generating background sequences using input from uniformly distributed random noise samples. In addition to adversarial loss, during training, norm-based loss in the image space and discriminator feature-space is also minimized between the generated images and the training data. The additional losses enable the generator to learn subtle background details, resulting in a more realistic complex scene generation. During testing, a novel back-propagation based algorithm is used to generate images with statistics similar to the test images. More appropriate random noise samples are searched by directly minimizing the loss function between the test and generated images both in the image and discriminator feature-spaces. The network is not updated in this step; only the input noise samples are iteratively modified to minimize the loss function. Moreover, motion information is used to ensure that this loss is only computed on small-motion pixels. A novel dataset containing outdoor time-lapsed images from dawn to dusk with a full illumination variation cycle is also proposed to better compare the MOD algorithms in outdoor scenes. Accordingly, extensive experiments on five benchmark datasets and comparison with 30 existing methods demonstrate the strength of the proposed algorithm.

Dynamic Background Subtraction (BS) is a fundamental problem in many vision-based applications. BS in real complex environments has several challenging conditions like illumination variations, shadows, camera jitters, and bad weather. In this study, we aim to address the challenges of BS in complex scenes by exploiting conditional least squares adversarial networks. During training, a scene-specific conditional least squares adversarial network with two additional regularizations including L1-Loss and Perceptual-Loss is employed to learn the dynamic background variations. The given input to the model is video frames conditioned on corresponding ground truth to learn the dynamic changes in complex scenes. Afterwards, testing is performed on unseen test video frames so that the generator would conduct dynamic background subtraction. The proposed method consisting of three loss-terms including least squares adversarial loss, L1-Loss and Perceptual-Loss is evaluated on two benchmark datasets CDnet2014 and BMC. The results of our proposed method show improved performance on both datasets compared with 10 existing state-of-the-art methods.

Moving object detection is a fundamental step in various computer vision applications. Robust Principal Component Analysis (RPCA) based methods have often been employed for this task. However, the performance of these methods deteriorates in the presence of dynamic background scenes, camera jitter, camouflaged moving objects, and/or variations in illumination. It is because of an underlying assumption that the elements in the sparse component are mutually independent, and thus the spatiotemporal structure of the moving objects is lost. To address this issue, we propose a spatiotemporal structured sparse RPCA algorithm for moving objects detection, where we impose spatial and temporal regularization on the sparse component in the form of graph Laplacians. Each Laplacian corresponds to a multi-feature graph constructed over superpixels in the input matrix. We enforce the sparse component to act as eigenvectors of the spatial and temporal graph Laplacians while minimizing the RPCA objective function. These constraints incorporate a spatiotemporal subspace structure within the sparse component. Thus, we obtain a novel objective function for separating moving objects in the presence of complex backgrounds. The proposed objective function is solved using a linearized alternating direction method of multipliers based batch optimization. Moreover, we also propose an online optimization algorithm for real-time applications. We evaluated both the batch and online solutions using six publicly available datasets that included most of the aforementioned challenges. Our experiments demonstrated the superior performance of the proposed algorithms compared with the current state-of-the-art methods.

Background modeling constitutes the building block of many computer-vision tasks. Traditional schemes model the background as a low rank matrix with corrupted entries. These schemes operate in batch mode and do not scale well with the data size. Moreover, without enforcing spatiotemporal information in the low-rank component, and because of occlusions by foreground objects and redundancy in video data, the design of a background initialization method robust against outliers is very challenging. To overcome these limitations, this paper presents a spatiotemporal low-rank modeling method on dynamic video clips for estimating the robust background model. The proposed method encodes spatiotemporal constraints by regularizing spectral graphs. Initially, a motion-compensated binary matrix is generated using optical flow information to remove redundant data and to create a set of dynamic frames from the input video sequence. Then two graphs are constructed, one between frames for temporal consistency and the other between features for spatial consistency, to encode the local structure for continuously promoting the intrinsic behavior of the low-rank model against outliers. These two terms are then incorporated in the iterative Matrix Completion framework for improved segmentation of background. Rigorous evaluation on severely occluded and dynamic background sequences demonstrates the superior performance of the proposed method over state-of-the-art approaches.

Background estimation and foreground segmentation are important steps in many high-level vision tasks. Many existing methods estimate background as a low-rank component and foreground as a sparse matrix without incorporating the structural information. Therefore, these algorithms exhibit degraded performance in the presence of dynamic backgrounds, photometric variations, jitter, shadows, and large occlusions. We observe that these backgrounds often span multiple manifolds. Therefore, constraints that ensure continuity on those manifolds will result in better background estimation. Hence, we propose to incorporate the spatial and temporal sparse subspace clustering into the robust principal component analysis (RPCA) framework. To that end, we compute a spatial and temporal graph for a given sequence using motion-aware correlation coefficient. The information captured by both graphs is utilized by estimating the proximity matrices using both the normalized Euclidean and geodesic distances. The low-rank component must be able to efficiently partition the spatiotemporal graphs using these Laplacian matrices. Embedded with the RPCA objective function, these Laplacian matrices constrain the background model to be spatially and temporally consistent, both on linear and nonlinear manifolds. The solution of the proposed objective function is computed by using the linearized alternating direction method with adaptive penalty optimization scheme. Experiments are performed on challenging sequences from five publicly available datasets and are compared with the 23 existing state-of-the-art methods. The results demonstrate excellent performance of the proposed algorithm for both the background estimation and foreground segmentation.

Computing a background model from a given sequence of video frames is a prerequisite for many computer vision applications. Recently, this problem has been posed as learning a low-dimensional subspace from high dimensional data. Many contemporary subspace segmentation methods have been proposed to overcome the limitations of the methods developed for simple background scenes. Unfortunately, because of the absence of motion information and without preserving intrinsic geometric structure of video data, most existing algorithms do not provide promising nature of the low-rank component for complex scenes. Such as largely occluded background by foreground objects, superfluity in video frames in order to cope with intermittent motion of foreground objects, sudden lighting condition variation, and camera jitter sequences. To overcome these difficulties, we propose a motion-aware regularization of graphs on low-rank component for video background modeling. We compute optical flow and use this information to make a motion-aware matrix. In order to learn the locality and similarity information within a video we compute inter-frame and intra-frame graphs which we use to preserve geometric information in the low-rank component. Finally, we use linearized alternating direction method with parallel splitting and adaptive penalty to incorporate the preceding steps to recover the model of the background. Experimental evaluations on challenging sequences demonstrate promising results over state-of-the-art methods.

According to the ways to see the real environments, mirror metaphor augmented reality systems can be classified into video see-through virtual mirror displays and reflective half-mirror displays. The two systems have distinctive characteristics and application fields with different types of complexity. In this paper, we introduce a system configuration to implement a prototype of a reflective half-mirror display-based augmented reality system. We also present a two-phase calibration method using an extra camera for the system. Finally, we describe three error sources in the proposed system and show the result of analysis of these errors with several experiments.

In this paper, we present visual hand and arm signals recognition and rifle firing positions estimation for virtual military training systems which perform various military tasks in the virtual environments. Among them, we deal with the platoon leader training and rifle firing position training for various military situations. In the platoon leader training, the trainee whose role is a platoon leader keeps command and control over his/her members by visual signals. The proposed visual signals and rifle firing positions recognition system exploits the optical motion capture system. For the real-time performance, our recognition method localizes the head and hands in the user-centered partitions of 3D space and analyzes their temporal transitions, and finally makes a simple decision to classify the visual hand and arm signals. Experimental results show the evaluation of the proposed method.

Visual odometry is important to obtain the vehicle motion as well as the relative motion of the surroundings such as moving/stationary objects to the vehicle. This paper presents a vehicle motion estimation approach which uses 3D vector correspondences. We design an objective function using 3D vector registration error and optimize it. The proposed method estimates more accurate rotation matrix compared to the previous method.

Abstract Image‐based rendering for walkthrough in the virtual environment has many advantages should over the geometry‐based approach, due to the fast construction of the environment and photo‐realistic rendered results. In image‐based rendering technique, rays from a set of input images are collected and a novel view image is rendered by the resampling of the stored rays. Current such techniques, however, are limited to a closed capture space. In this paper, we propose a multiple path‐based capture configuration that can handle a large‐scale scene and a disparity‐based warping method for novel view generation. To acquire the disparity image, we segment the input image into vertical slit segments using a robust and inexpensive way of detecting vertical depth discontinuity. The depth slit segments, instead of depth pixels, reduce the processing time for novel view generation. We also discuss a dynamic cache strategy that supports real‐time walkthroughs in large and complex street environments. The efficiency of the proposed method is demonstrated with several experiments. Copyright © 2005 John Wiley & Sons, Ltd.

In a traditional streaming system, the field of view (FOV) of an image is limited based on the FOV of the camera. One solution is the use of a pan/tilt camera, yet this is difficult for multiple clients to handle, plus there is a delay between the client and the server due to camera motion. Therefore, to solve these problems, a new streaming system is proposed that uses panorama image. However, the server does not compress the whole panorama image, which would involve sending huge amounts of data through the Internet. Instead, only parts of the panorama image are compressed and transmitted based on the navigation requests of the client, thereby reducing the compression burden on the server.

Abstract Modelling from images is a cost‐effective means of obtaining virtual cultural heritage models. These models can be effectively constructed from classical Structure from Motion algorithm. However, it's too difficult to reconstruct whole scenes using SFM method since general oriental historic sites contain a very complex shapes and brilliant colours. To overcome this difficulty, the current paper proposes a new reconstruction method based on a moving planar mirror. We devise the mirror posture instead of scene itself as a cue for reconstructing the geometry. That implies that the geometric cues are inserted into the scene by compulsion. With this method, we can obtain the geometrical details regardless of the scene complexity. For this purpose, we first capture image sequences through the moving mirror containing the interested scene, and then calibrate the camera through the mirror's posture. Since the calibration results are still inaccurate due to the detection error, the camera pose is revised using frame‐correspondence of the corner points that are easily obtained using the initial camera posture. Finally, 3D information is computed from a set of calibrated image sequences. We validate our approach with a set of experiments on some cultural heritage objects. Copyright © 2004 John Wiley & Sons, Ltd.

Image-based rendering for walkthroughs in a virtual environment has many advantages over a geometry-based approach, due to the fast construction of the environment and, photo-realistic rendered results, etc. However, since light field approaches are limited to a closed space, they cannot be directly applied to large and complex street environments. Accordingly, the current paper, presents an image-based rendering method that can handle a large-scale scene, thereby, allowing street walkthroughs with light field style rendering. First, the problem of capture configuration is described based on the previous work. Then an efficient sampling configuration of image sequences is presented, and the efficiency of the proposed method demonstrated with several experiments.

3D models of urban sites with geometry and facade textures are needed for many planning and visualization applications. Approximate 3D wireframe model can be derived from aerial images but detailed textures must be obtained from ground level images. Integrating such views with the 3D models is difficult as only small parts of buildings may be visible in a single view. We describe a method that uses two or three vanishing points, and three 3D to 2D line correspondences to estimate the rotational and translational parameters of the ground level cameras. The valid set of multiple combinations of 3D to 2D line pairs is chosen by a hypotheses generation and evaluation Some experimental results are presented.

Visualization of city scenes is important for many applications including entertainment and urban mission planning. Models covering wide areas can be efficiently constructed from aerial images. However, only roof details are visible from aerial views; ground views are needed to provide details of the building facades for high quality 'fly-through' visualization or simulation applications. We present an automatic method of integrating facade textures from ground view images into 3D building models for urban site modeling. We first segment the input image into building facade regions using a hybrid feature extraction method, which combines global feature extraction with Hough transform on an adaptively tessellated Gaussian Sphere and local region segmentation. We estimate the external camera parameters by using the corner points of the extracted facade regions to integrate the facade textures into the 3D building models. We validate our approach with a set of experiments on some urban sites. Categories and Subject Descriptors (according to ACM CCS): I.3.3 [Computer Graphics]: Modeling packages

We propose a new computational model for mimicking the behavior of a human eye movement during saccades. The different characteristics of two types of saccades, such as a reflexive saccade and an intentional saccade, are reflected on the proposed model. We divided the visual pathway for generating a saccadic eye movement into three parts, of which each part was modeled using different neural networks. The visual pathway from the visual receptors to the visual cortex including the frontal eye field was modeled by the self-organizing feature map, and the visual pathway from the visual cortex to the superior colliculus was modeled by a modified learning vector quantization network. The visual pathway front the superior colliculus to the motoneuron is modeled by a multilayer neural network with backpropagation learning algorithm. Experimental results from computer simulation show that the proposed computational model is able to mimic well the behavior of the human eye movement for two different saccades.

Article A multiresolution control method using view directional feature Share on Authors: HyungSeok Kim Department of Computer Science, KAIST YooSungGoo, Taejon, Korea Department of Computer Science, KAIST YooSungGoo, Taejon, KoreaView Profile , Soon Ki Jung Department of Computer Engineering, Kyungpook National University PookGoo, Taegu, Korea Department of Computer Engineering, Kyungpook National University PookGoo, Taegu, KoreaView Profile , KwangYun Wohn Department of Computer Science, KAIST, YooSungGoo, Taejon Department of Computer Science, KAIST, YooSungGoo, TaejonView Profile Authors Info & Claims VRST '98: Proceedings of the ACM symposium on Virtual reality software and technologyNovember 1998 Pages 163–169https://doi.org/10.1145/293701.293723Online:02 November 1998Publication History 3citation863DownloadsMetricsTotal Citations3Total Downloads863Last 12 Months2Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

The paper is concerned with the efficient generation of graphic images by exploiting the frame coherence in visibility to minimize polygon flow. The basic idea is to manage the visible polygon set (VPS) somewhat loosely, thereby reducing the computation effort considerably, while maintaining the quality of the rendered image reasonably high. Our method executes the culling-prediction-interpolation-rendering loop repeatedly, in an intelligent fashion. It calculates the exact visible polygons for two frames; one for the current view and the other for the predicted view after some frames. The in-between frames can be rendered from those polygons that belong to the two extreme frames. Furthermore, our algorithm regulates the frame rate to the desired target frame rate by controlling the number of the in-between frames. As our algorithm needs to calculate the visible polygons for the key frames, any other culling algorithms can be easily used with our algorithm in this part.