Stereo image pairs can be used to improve the performance of super-resolution (SR) since additional information is provided from a second viewpoint. However, it challenging incorporate this for SR disparities between stereo images vary significantly. In paper, we propose parallax-attention superresolution network (PASSRnet) integrate pair SR. Specifically, introduce mechanism with global receptive field along epipolar line handle different large disparity variations. We also new and largest...

Current CNN-based super-resolution (SR) methods process all locations equally with computational resources being uniformly assigned in space. However, since missing details low-resolution (LR) images mainly exist regions of edges and textures, less are required for those flat regions. Therefore, existing involve redundant computation regions, which increases their cost limits applications on mobile devices. In this paper, we explore the sparsity image SR to improve inference efficiency...

Video super-resolution (SR) aims at generating a sequence of high-resolution (HR) frames with plausible and temporally consistent details from their low-resolution (LR) counterparts. The key challenge for video SR lies in the effective exploitation temporal dependency between consecutive frames. Existing deep learning based methods commonly estimate optical flows LR to provide dependency. However, resolution conflict HR outputs hinders recovery fine details. In this paper, we propose an...

Recently, the performance of single image super-resolution (SR) has been significantly improved with powerful networks. However, these networks are developed for SR specific integer scale factors (e.g., ×2/3/4), and cannot handle non-integer asymmetric SR. In this paper, we propose to learn a scale-arbitrary network from scale-specific Specifically, develop plug-in module existing perform SR, which consists multiple scale-aware feature adaption blocks upsampling layer. Moreover, conditional...

Stereo image pairs encode 3D scene cues into stereo correspondences between the left and right images. To exploit within images, recent CNN based methods commonly use cost volume techniques to capture correspondence over large disparities. However, since disparities can vary significantly for cameras with different baselines, focal lengths resolutions, fixed maximum disparity used in hinders them handle variations. In this paper, we propose a generic parallax-attention mechanism (PAM)...

In this paper, we propose a novel sparsity-based algorithm for anomaly detection in hyperspectral imagery. The is based on the concept that background pixel can be approximately represented as sparse linear combination of its spatial neighbors while an cannot if anomalies are removed from neighborhood. To physically meaningful, sum-to-one and nonnegativity constraints imposed to abundance vector mixture model, upper bound constraint sparsity level better recovery test pixel. First, proposed...

Satellite video cameras can provide continuous observation for a large-scale area, which is important many remote sensing applications. However, achieving moving object detection and tracking in satellite videos remains challenging due to the insufficient appearance information of objects lack high-quality datasets. In this article, we first build dataset with rich annotations task tracking. This collected by Jilin-1 constellation composed 47 1 646 038 instances interest 3711 trajectories We...

Training a convolutional neural network (CNN) to detect infrared small targets in fully supervised manner has gained remarkable research interests recent years, but is highly labor expensive since large number of per-pixel annotations are required. To handle this problem, paper, we make the first attempt achieve target detection with point-level supervision. Interestingly, during training phase by point labels, discover that CNNs learn segment cluster pixels near targets, and then gradually...

In this paper, we introduce a new approach to address the challenge of generalization in hyperspectral anomaly detection (AD). Our method eliminates need for adjusting parameters or retraining on test scenes as required by most existing methods. Employing an image-level training paradigm, achieve general enhancement network AD that only needs be trained once. Trained set anomaly-free images with random masks, our can learn spatial context characteristics between anomalies and background...

Single-frame infrared small target (SIRST) detection aims at separating targets from clutter backgrounds on images. Recently, deep learning based methods have achieved promising performance SIRST detection, but the cost of a large amount training data with expensive pixel-level annotations. To reduce annotation burden, we propose first method to achieve single-point supervision. The core idea this work is recover per-pixel mask each given single point label by using clustering approaches,...

Deep subspace clustering network has shown its effectiveness in hyperspectral image (HSI) clustering. However, there are two major challenges that need to be addressed: 1) lack of effective supervision for feature learning; and 2) negative effect caused by the high redundancy global dictionary atoms. In this article, we propose an end-to-end trainable HSI Specifically, ensure extracted features well-suited subsequent clustering, cluster assignments with confidence employed as pseudo-labels...

The purpose of non-uniformity and blind pixel correction is to provide a more reliable foundation for subsequent image processing target detection. Existing methods generally struggle balance the contradiction between over-smoothing residual noise. Particularly, can easily filter out texture details dim small targets. Based on multi-frame response model infrared focal plane array detector, we propose two-stage 3-D fully convolutional network factor estimation, integrated with an suppression...

As a long-standing problem, infrared small target detection is challenging due to the dimness of targets and complexity background. Considering limitation traditional approaches, we propose an accurate robust method for using multiscale gray variance difference measures. A adaptive measure first used enhance improve accuracy. Then, proposed alleviate impact background fluctuation robustness our method. By integrating these two can be extracted accurately threshold-adaptive segmentation....

Infrared small target detection is challenging due to the various background and low signal-to-clutter ratios. Considering information deficiency faced by single spatial or temporal information, we construct a false alarm filter for infrared detection. A multiscale patch-based contrast measure first used suppress remove cloud edges at coarse level. Then, variance broken regions noise fine By integrating these two methods, targets can be extracted accurately robustly using an adaptive...

Hyperspectral anomaly detection (HAD) aims to distinguish anomalies from background-by-background modeling. Deep learning has been applied HAD and achieves promising results. However, there exist several issues that need be addressed: 1) unrealistic Gaussian assumption on the latent representations may limit its application; 2) deep features are not well-suited due separation between feature detection; 3) lack of adequate exploitation spectral-spatial features; 4) negative effect caused by...

Hyperspectral anomaly detection (HAD) aims at distinguishing anomalies from background in an unsupervised manner. Autoencoder (AE) and its variant-based methods have achieved promising performance HAD. However, most existing neglect to exploit the local structure information of hyperspectral images (HSIs) that reflects underlying relationships between each pixel surroundings. Hence, representation capabilities networks are restricted. Moreover, reconstruction during training compels learn...

Compared to hyperspectral trackers that adopt the “pre-training then fine-tuning” training paradigm, those using prompt-tuning” paradigm can inherit expressive capabilities of pre-trained model with fewer parameters. Existing utilizing prompt learning lack an adequate template design, thus failing bridge domain gap between data and models. Consequently, their tracking performance suffers. Additionally, these networks have a poor generalization ability require re-training for different...

Correlation filter-based tracking methods have been intensively investigated for their high efficiency and robustness. However, a single feature-based tracker cannot adapt to challenging situations, such as severe deformation, rotation, illumination variations. Besides, simple linear interpolation-based model updating mechanism is prone degradation, consequently drifting. In this paper, 2-D location filter combined with 1-D scale jointly estimate the state of object under tracking, three...

In this paper, we propose a novel constrained sparse-representation-based binary hypothesis model for target detection in hyperspectral imagery. This is based on the concept that pixel can only be linearly represented by union dictionary combined background and dictionary, while both dictionary. To physically meaningful, non-negativity constraint imposed to weight vector. suppress signals upper bound also These bounds are adaptively estimated similarities between atoms target. Then, vectors...

With great significance in military and civilian applications, subpixel target detection is of interest hyperspectral remote sensing. The targets usually also need to be unmixed identify their components. Traditionally, these are first detected then obtain corresponding abundances. Therefore, unmixing independently performed. However, there potential relations between two processes that investigated. In this article, we integrate using iterative constrained sparse representation. main idea...

Infrared dim small target detection is regarded as a critical technology for the interpretation of space-based remote sensing images. In recent years, driven by deep learning and surge data, remarkable effects have been achieved in infrared Nevertheless, intrinsic feature scarcity low signal-to-clutter ratio (SCR) characteristics pose tremendous challenges to learning-based methods. this letter, we present novel sub-pixel sampling cuneate network (SPSCNet) detect targets The overall model...

Small target detection is a crucial and challenging task in infrared search track system. Background estimation-based methods an effective important approach for small detection. Affected by the pixels, existing background estimation may reconstruct inaccurate background. Based on image inpainting technique, we propose novel two-stage to predict more accurate backgrounds. At first stage, inner outer window-based (IOWII) used obtain rough estimation. Then mask of candidate region...