A5101581669- Human Pose and Action Recognition
- Multimodal Machine Learning Applications
- Advanced Image and Video Retrieval Techniques
- Advanced Image Processing Techniques
- Music and Audio Processing
- Advanced Image Fusion Techniques
- Image Enhancement Techniques
- Anomaly Detection Techniques and Applications
- Video Analysis and Summarization
- Video Surveillance and Tracking Methods
- Digital Media Forensic Detection
- Domain Adaptation and Few-Shot Learning
- Speech and Audio Processing
- Advanced Steganography and Watermarking Techniques
- Image and Signal Denoising Methods
- Image Processing and 3D Reconstruction
- Advanced Vision and Imaging
- Advanced Measurement and Detection Methods
- Infrared Target Detection Methodologies
- Face and Expression Recognition
- Recommender Systems and Techniques
- Topic Modeling
- Remote-Sensing Image Classification
- Advanced Graph Neural Networks
- Image and Object Detection Techniques
Beijing University of Posts and Telecommunications
2021-2024
Nanyang Institute of Technology
2024
Army Medical University
2021
Troy University
2020
Donghua University
2018
Huazhong University of Science and Technology
2018
Embedded Systems (United States)
2017
Wuhan University
2015
Ordnance Engineering College
2009
Weatherford College
2008
Images captured in low-light conditions usually suffer from very low contrast, which increases the difficulty of subsequent computer vision tasks a great extent. In this paper, image enhancement model based on convolutional neural network and Retinex theory is proposed. Firstly, we show that multi-scale equivalent to feedforward with different Gaussian convolution kernels. Motivated by fact, consider Convolutional Neural Network(MSR-net) directly learns an end-to-end mapping between dark...
Rain removal from a single image is challenge which has been studied for long time. In this paper, novel convolutional neural network based on wavelet and dark channel proposed. On one hand, we think that rain streaks correspond to high frequency component of the image. Therefore, haar transform good choice separate background some extent. More specifically, LL subband more inclined express information, while HL, LH tend represent edges respectively. other accumulation distance makes look...
Audio-visual event (AVE) localization aims to localize the temporal boundaries of events that contains visual and audio contents, identify categories in unconstrained videos. Existing work usually utilizes successive video segments for modeling. However, ambient sounds or irrelevant targets some often cause problem audio-visual semantics inconsistency, resulting inaccurate global To tackle this issue, we present a consistent segment selection network (CSS-Net) paper. First, propose novel...
People counting and pedestrian flow statistics are challenging tasks because of the perspective distortions, appearance changes occlusion. In this paper, we address two tasks: people in images highly dense crowds a place over period time. Our first contribution is to propose new convolution neural network (CNN) model which composed deep shallow fully fulfill task counting. We extract different layer features from last network, concatenate them together. After that add deconvolution layers...
Visual object tracking is a challenging task when the appearance changes caused by scale variation and occlusion. In this paper, an algorithm proposed which capable of dealing with case that occlusion occur simultaneously. A kernelized correlation filter (KCF) first learned to obtain response, whose maximum value denotes optimal location. order represent sample better, convolutional features are extracted from pre-trained neural networks (CNNs). Then, strategy adaption used estimate during...
Image captioning is a challenging task that generates natural language description based on the visual understanding of given image. Significant region representation milestone in image captioning. Despite great success existing region-based works, they only focus salient objects and encode these independently, still plagued by lack global contextual information relationships. In fact, structured relationships are exactly merits traditional grid features emerging scene graph features. this...
In this paper, a scalable, robust and recovery-driven authentication scheme targeting at verifying the authenticity of JPEG2000 images is proposed. It achieved by truncating bit planes wavelet coefficients into two portions in codec based on lowest compression rate (CBR). The invariant features, which are generated from upper portion, signed sender's private key to generate crypto signature. By embedding signature has ability for content as long final transmitted image not less than CBR....
Rain removal from a single image is challenge which has been studied for long time. In this paper, novel convolutional neural network based on wavelet and dark channel proposed. On one hand, we think that rain streaks correspond to high frequency component of the image. Therefore, haar transform good choice separate background some extent. More specifically, LL subband more inclined express information, while LH, HL, HH tend represent edges. other accumulation distance makes look like haze...
Cloud contamination is the most common defect leading to quality degradation in remote sensing images. Numerous cloud-cover assessment (CCA) methods have been developed literature. The traditional Landsat 7 CCA algorithm attempted detect clouds by taking advantages of different spectral properties from five bands. However, it suffers weakness omitting thin cirrus and requirement thermal In this paper, we derived an automated (ACCA) model that measures statistical deviations spatial domain...
Conversational Recommender Systems (CRSs) in E-commerce platforms aim to recommend items users via multiple conversational interactions. Click-through rate (CTR) prediction models are commonly used for ranking candidate items. However, most CRSs suffer from the problem of data scarcity and sparseness. To address this issue, we propose a novel knowledge-enhanced deep cross network (K-DCN), two-step (pretrain fine-tune) CTR model We first construct billion-scale conversation knowledge graph...
The technique of generating a caption for blurred image described by AI still exists the hurdle recognition. In image, figuring out semantic subject is severe challenge. this paper, we implement classifier as an auxiliary oriented filter that combines with standard dense based architecture. This refined architecture used to categorize main from media and transform it into specific predicting range field. proposed framework can describe outcomes semantical relations are hidden in image; they...