Camera motion introduces blur, affecting many computer vision tasks. Dark Channel Prior (DCP) helps the blind deblurring on scenes including natural, face, text, and low-illumination images. However, it has limitations is less likely to support kernel estimation while bright pixels dominate input image. We observe that in clear images are not be after blur process. Based this observation, we first illustrate phenomenon mathematically define as Bright (BCP). Then, propose a technique for such...

In the past few years, semi-supervised node classification in attributed network has been developed rapidly. Inspired by success of deep learning, researchers adopt convolutional neural to develop Graph Convolutional Networks (GCN), and they have achieved surprising accuracy considering topological information employing fully connected (FCN). However, given topology may also induce a performance degradation if it is directly employed classification, because possess high sparsity certain...

Existing topic modeling approaches possess several issues, including the overfitting issue of Probablistic Latent Semantic Indexing (pLSI), failure capturing rich topical correlations among topics in Dirichlet Allocation (LDA), and high inference complexity. In this paper, we provide a new method to overcome pLSI by using amortized with word embedding as input, instead prior LDA. For generative model, large number free latent variables is root overfitting. To reduce parameters, replaces...

Graph Neural Networks (GNNs) have achieved tremendous success in graph representation learning. Unfortunately, current GNNs usually rely on loading the entire attributed into network for processing. This implicit assumption may not be satisfied with limited memory resources, especially when is large. In this paper, we pioneer to propose a Binary Convolutional Network (Bi-GCN), which binarizes both parameters and input node features. Besides, original matrix multiplications are revised binary...

With the malicious use and dissemination of multi-modal deepfake videos, researchers start to investigate detection. Unfortunately, most existing methods tune all parameters deep network with limited speech video datasets are trained under coarse-grained consistency supervision, which hinders their generalization ability in practical scenarios. To solve these problems, this paper, we propose first multi-task audio-visual prompt learning method for detection, by exploiting multiple foundation...

In this paper, we address the problem of image bit-depth expansion and present a novel method to generate high (HBD) images from single low (LBD) image. We expand by reconstructing least significant bits (LSBs) for LBD after it is rescaled bit-depth. For regions whose intensities are neither locally maximum nor minimum, neighborhood flooding applied convert 2D interpolation into 1D interpolation, local maxima/minima (LMM) where not applicable, virtual skeleton marking algorithm proposed...

Image forensics aims to detect the manipulation of digital images. Currently, splicing detection, copy-move and image retouching detection are attracting significant attention from researchers. However, editing techniques develop over time. An emerging technique is colorization, in which grayscale images colorized with realistic colors. Unfortunately, this may also be intentionally applied certain confound object recognition algorithms. To best our knowledge, no forensic has yet been...

Recently, Human Attribute Recognition (HAR) has become a hot topic due to its scientific challenges and application potentials, where localizing attributes is crucial stage but not well handled. In this paper, we propose novel deep learning approach HAR, namely Distraction-aware HAR (Da-HAR). It enhances CNN feature by improving attribute localization through coarse-to-fine attention mechanism. At the coarse step, self-mask block built roughly discriminate reduce distractions, while at fine...

In this paper, we propose a new reversible watermarking algorithm based on additive prediction-error expansion which can recover original image after extracting the hidden data. Embedding capacity of such algorithms depend prediction accuracy predictor. We observed that performance predictor full context is preciser as compared to partial prediction. view observation, an efficient adaptive (EAP) method context, exploits local characteristics neighboring pixels much effectively than other...

Semi-supervised classification is a fundamental technology to process the structured and unstructured data in machine learning field. The traditional attribute-graph based semi-supervised methods propagate labels over graph which usually constructed from features, while convolutional neural networks smooth node attributes, i.e., real topology. In this paper, they are interpreted perspective of propagation, accordingly categorized into symmetric asymmetric propagation methods. From both...

Motivated by the capability of Generative Adversarial Network on exploring latent semantic space and capturing variations in data distribution, adversarial learning has been adopted network embedding to improve robustness. However, this important ability is lost existing adversarially regularized methods, because their results are directly compared samples drawn from perturbation (Gaussian) distribution without any rectification real data. To overcome vital issue, a novel Joint Embedding...

Most attempts on extending Graph Neural Networks (GNNs) to Heterogeneous Information (HINs) implicitly take the direct assumption that multiple homogeneous attributed networks induced by different meta-paths are complementary. The doubts about hypothesis of complementary motivate an alternative consensus. That is, aggregated node attributes shared essential for representations, while specific ones in each network should be discarded. In this paper, a novel Bottleneck (HGIB) is proposed...

Image colorization is the task to color a grayscale image with limited cues. In this work, we present novel method perform using sparse representation. Our first trains an over-complete dictionary in YUV space. Then taking and small subset of pixels as inputs, our colorizes overlapping patches via representation; it achieved by seeking representations that are consistent both pixels. After that, aggregate colorized weights get intermediate result. This process iterates until properly...

In color image compression, the chroma components are often sub-sampled before compression and up-sampled after compression. Although sub-sampling saves bit-cost for it induces extra distortions in compressed images. this paper, we propose two approaches to tackle problem. First, a method transform domain apply both components. Then, based on sub-sampling, novel modify luma component. our proposed modification algorithm, that occurred can be coupled together utilized With algorithms, achieve...

The success of graph convolutional neural networks (GCNNs) based semi-supervised node classification is credited to the attribute smoothing (propagating) over topology. However, attributes may be interfered by utilization topology information. This distortion will induce a certain amount misclassifications nodes, which can correctly predicted with only attributes. By analyzing impact edges in propagations, simple edges, connect two nodes similar attributes, should given priority during...

Deepfake video detection has drawn significant attention from researchers due to the security issues induced by deepfake videos. Unfortunately, most of existing approaches have not competently modeled natural structures and movements human faces. In this paper, we formulate problem into a graph classification task, propose novel paradigm named Facial Action Dependencies Estimation (FADE) for detection. We Multi-Dependency Graph Module (MDGM) capture abundant dependencies among facial action...

Digital image forensics seeks to detect statistical traces left by acquisition or post-processing in order establish an images source and authenticity. cameras acquire with one sensor overlayed a color filter array (CFA), capturing at each spatial location sample from the three necessary channels. The missing pixels must be interpolated process known as demosaicking. This is highly nonlinear can vary greatly between different camera brands models. Most practical algorithms, however,...

In network analysis, community detection and embedding are two important topics. Community tends to obtain the most noticeable partition, while aims at seeking node representations which contains as many diverse properties possible. We observe that current problems being resolved by a general solution, i.e., "maximizing consistency between similar nodes maximizing distance dissimilar nodes." This solution only exploits structure (facet) of network, effectively satisfies demands detection....

In this paper, we carry out a performance analysis from probabilistic perspective to introduce the error diffusion-based halftone visual watermarking (EDHVW) methods' expected performances and limitations. Then, propose new general EDHVW method, content aware double-sided embedding diffusion (CaDEED), via considering watermark decoding with specific of cover images watermark, different noise tolerance abilities various image content, importance levels every pixel (when being perceived) in...

Although many graph convolutional neural networks (GCNNs) have achieved superior performances in semisupervised node classification, they are designed from either the spatial or spectral perspective, yet without a general theoretical basis. Besides, most of existing GCNNs methods tend to ignore ubiquitous noises network topology and content thus unable model these uncertainties. These drawbacks certainly reduce their effectiveness integrating content. To provide probabilistic perspective...