- Advanced Image Processing Techniques
- Advanced Vision and Imaging
- Image Enhancement Techniques
- Advanced Image and Video Retrieval Techniques
- Digital Media Forensic Detection
- Optical measurement and interference techniques
- Advanced Steganography and Watermarking Techniques
- Video Surveillance and Tracking Methods
- Image Processing Techniques and Applications
- Advanced Image Fusion Techniques
- Video Analysis and Summarization
- Image Retrieval and Classification Techniques
- Generative Adversarial Networks and Image Synthesis
- Chaos-based Image/Signal Encryption
- Sentiment Analysis and Opinion Mining
- Computer Graphics and Visualization Techniques
- Tactile and Sensory Interactions
- Ideological and Political Education
- Visual Attention and Saliency Detection
- Robotics and Sensor-Based Localization
- Image and Signal Denoising Methods
- Human Pose and Action Recognition
- Biometric Identification and Security
- 3D Shape Modeling and Analysis
- Image and Video Quality Assessment
Northwestern Polytechnical University
2019-2024
Nanyang Technological University
2023-2024
Southwest Jiaotong University
2023-2024
Guangzhou University
2024
Xi'an Technological University
2023
Wuchang University of Technology
2023
ETH Zurich
2022
Xi'an University of Science and Technology
2022
University of Nottingham Ningbo China
2022
Xi'an University of Technology
2022
To cope with the high resource (network and compute) demands of real-time video analytics pipelines, recent systems have relied on frame filtering. However, filtering has typically been done neural networks running edge/backend servers that are expensive to operate. This paper investigates on-camera filtering, which moves beginning pipeline. Unfortunately, we find commodity cameras limited compute resources only permit via differencing based low-level features. Used incorrectly, such...
Recent deep learning methods have achieved promising results in image shadow removal. However, their restored images still suffer from unsatisfactory boundary artifacts, due to the lack of degradation prior embedding and deficiency modeling capacity. Our work addresses these issues by proposing a unified diffusion framework that integrates both priors for highly effective In detail, we first propose model, which inspires us build novel unrolling dubbed ShandowDiffusion. It remarkably...
Previous raw image-based low-light image enhancement methods predominantly relied on feed-forward neural networks to learn deterministic mappings from normally-exposed images. However, they failed capture critical distribution information, leading visually undesirable results. This work addresses the issue by seamlessly integrating a diffusion model with physics-based exposure model. Different vanilla that has perform Gaussian denoising, injected model, our restoration process can directly...
Existing deep learning-based depth completion methods generally employ massive stacked layers to predict the dense map from sparse input data. Although such approaches greatly advance this task, their accompanied huge computational complexity hinders practical applications. To accomplish more efficiently, we propose a novel lightweight network framework, Long-short Range Recurrent Updating (LRRU) network. Without learning complex feature representations, LRRU first roughly fills obtain an...
To enhance low-light images to normally-exposed ones is highly ill-posed, namely that the mapping relationship between them one-to-many. Previous works based on pixel-wise reconstruction losses and deterministic processes fail capture complex conditional distribution of normally exposed images, which results in improper brightness, residual noise, artifacts. In this paper, we investigate model one-to-many via a proposed normalizing flow model. An invertible network takes images/features as...
When creating a photo album of an event, people typically select few important images to keep or share. There is some consistency in the process choosing images, and discarding unimportant ones. Modeling this selection will assist automatic summarization. In paper, we show that consistent among different viewers, related event type album. We introduce concept event-specific image importance. collected new dataset with human annotation relative importance each also propose Convolutional...
Generally, the purpose of a steganalysis algorithm is to establish presence secret messages in stego data. However, quantitative steganalyzers can reveal more information about communication by estimating exact volume embedded messages. Quantitative crucial step for breaking codes many practical scenarios. This work concerns videos. Most video steganographical algorithms embed modifying values motion vector compressed domain. We propose general framework constructing that are able detect...
Automatic organization of personal photos is a problem with many real world ap- plications, and can be divided into two main tasks: recognizing the event type photo collection, selecting interesting images from collection. In this paper, we attempt to simultaneously solve both album-wise recognition image- wise importance prediction. We collected an album dataset labels image labels, refined existing CUFED dataset. propose hybrid system consisting three parts: A siamese network-based...
With the rapid development of deep learning, generating realistic fake face videos is becoming easier. It common to make news, network pornography, extortion and other related illegal events using forgery. In order attenuate harm forgery video, researchers proposed many detection methods based on tampering traces introduced by However, these generally have poor cross-database performance. Therefore, this paper proposes a multi-feature fusion method improve generalization ability detector....
LiDAR depth-only completion is a challenging task to estimate dense depth maps only from sparse measurement points obtained by LiDAR. Even though the methods have been widely developed, there still significant performance gap with RGB-guided that utilize extra color images. We find existing can obtain satisfactory results in areas where are almost accurate and evenly distributed (denoted as normal areas), while limited foreground background overlapped due occlusion overlap areas) no around...
Recent deep learning methods have achieved superior results in shadow removal. However, most of these supervised rely on training over a huge amount and shadow-free image pairs, which require laborious annotations may end up with poor model generalization. Shadows, fact, only form partial degradation images, while their non-shadow regions provide rich structural information potentially for unsupervised learning. In this paper, we propose novel diffusion-based solution removal, separately...
Purpose: Oral drug administration is the most common and convenient route, offering good patient compliance but solubility limits oral applications. Celecoxib, an insoluble drug, requires continuous high-dose administration, which may increase cardiovascular risk. The nanostructured lipid carriers prepared from drugs excipients can effectively improve bioavailability, reduce dosage, lower risk of adverse reactions. Methods: In this study, we hyaluronic acid-modified celecoxib (HA-NLCs) to...
Using unmanned aerial vehicles (UAVs) as devices for traffic data collection exhibits many advantages in collecting information. This paper presents an efficient method based on the deep learning and handcrafted features to classify taken from drone imagery. Experimental results show that compared classification algorithms pre-trained CNN or hand-crafted features, proposed algorithm higher accuracy vehicle recognition at different UAV altitudes with view scopes, which can be used future...
Abstract This paper is focused on the nonlinear state estimation problem with finite-step correlated noises and packet loss. Firstly, by using projection theorem repeatedly, mean covariance of process noise measurement in condition measurements before current epoch are calculated. Then, based Gaussian approximation recursive filter (GASF) prediction compensation mechanism, one-step predictor dropouts derived, respectively. Based these, a proposed. Subsequently, numerical implementation...
Tracking with kernelized correlation filters (KCF) is an excellent object tracking algorithm, which widely concerned. In KCF, each candidate patch in tracked region corresponds to a confidence ratio indicating the probability of containing target, and maximum output. traditional its performance decreases complex scenes model liable be contaminated due updated every frame. To overcome these limitations, we combine all available ratios form map, then by analyzing infer scene adopt different...
While super-resolution (SR) methods based on diffusion models exhibit promising results, their practical application is hindered by the substantial number of required inference steps. Recent utilize degraded images in initial state, thereby shortening Markov chain. Nevertheless, these solutions either rely a precise formulation degradation process or still necessitate relatively lengthy generation path (e.g., 15 iterations). To enhance speed, we propose simple yet effective method for...