A5003623628- Image Enhancement Techniques
- Advanced Vision and Imaging
- Advanced Image Processing Techniques
- Image and Signal Denoising Methods
- Advanced Image Fusion Techniques
- Image and Video Quality Assessment
- Image Processing Techniques and Applications
- Video Coding and Compression Technologies
- Advanced Image and Video Retrieval Techniques
- Color Science and Applications
- Image Retrieval and Classification Techniques
- Video Surveillance and Tracking Methods
- Visual Attention and Saliency Detection
- Video Analysis and Summarization
- Photoacoustic and Ultrasonic Imaging
- Optical measurement and interference techniques
- Medical Image Segmentation Techniques
- Advanced Neural Network Applications
- Handwritten Text Recognition Techniques
- Generative Adversarial Networks and Image Synthesis
- Human Pose and Action Recognition
- Face recognition and analysis
- Face and Expression Recognition
- Multimedia Communication and Technology
- Atmospheric chemistry and aerosols
Xidian University
2016-2025
Seoul National University of Science and Technology
2024
ETH Zurich
2019
Karlsruhe Institute of Technology
2015-2016
Kyung-in Women's University
2004-2015
Huawei Technologies (China)
2013
Sungkyunkwan University
2003-2009
Samsung (South Korea)
2005-2006
We propose automatic contrast-limited adaptive histogram equalization (CLAHE) for image contrast enhancement. automatically set the clip point CLAHE based on textureness of a block. Also, we introduce dual gamma correction into to achieve enhancement while preserving naturalness. First, redistribute block in dynamic range each Second, perform enhance luminance, especially dark regions reducing over-enhancement artifacts. Since adaptively enhances boosting it is very effective enhancing...
Gradient-boosted decision trees (GBDTs) are widely used in machine learning, and the output of current GBDT implementations is a single variable. When there multiple outputs, constructs corresponding to variables. The correlations between variables ignored by such strategy causing redundancy learned tree structures. In this article, we propose general method learn for called GBDT-MO. Each leaf GBDT-MO predictions all or subset automatically selected This achieved considering summation...
We provide a position-patch based face hallucination method using convex optimization. Recently, novel has been proposed to save computational time and achieve high-quality hallucinated results. This employed least square estimation obtain the optimal weights for hallucination. However, approach can biased solutions when number of training position-patches is much larger than dimension patch. To overcome this problem, letter proposes new which on Experimental results demonstrate that our...
Dilated convolutions support expanding receptive field without parameter exploration or resolution loss, which turn out to be suitable for pixel-level prediction problems. In this paper, we propose multiscale single image super-resolution (SR) based on dilated convolutions. We adopt expand the size incurring additional computational complexity. mix standard and in each layer, called mixed convolutions, i.e., convolutional feature extracted by are concatenated. theoretically analyze intensity...
In this paper, we propose fully convolutional siamese fusion networks for object tracking. We adopt the strategy of layers tracking to achieve good feature representation based on neural networks. Specifically, fuse three VGGNet normalized cross correlation (NCC). First, use as basis fusion, and reduce size a convolution kernel. Then, resize be same deconvolutional layer fusion. Next, NCC between target search region, produce response map. Finally, get result from map by maximum response....
Abstract. During the 2006 Tropical Warm Pool International Cloud Experiment (TWP-ICE) in tropics, 2008 Indirect and Semi-Direct Aerosol Campaign (ISDAC) Arctic, 2010 Small PARTicles In CirrUS (SPARTICUS) campaign at mid-latitudes, high-resolution images of ice crystals were recorded by a Particle Imager temperatures (T) between −87 0 °C. The projected maximum dimension (D'), length (L'), width (W') pristine columns, plates, component bullets bullet rosettes measured using newly developed...
Face super-resolution (SR) has become an indispensable function in security solutions such as video surveillance and identification system, but the distortion facial components is a great challenge it. Most state-of-the-art methods have utilized priors with deep neural networks. These require extra labels, longer training time, larger computation memory. In this paper, we propose novel Edge Identity Preserving Network for SR Network, named EIPNet, to minimize by utilizing lightweight edge...
In this paper, we propose retinex-based perceptual contrast enhancement in images using luminance adaptation. Strong illumination causes the loss of local details an image. We adopt adaptation and multi-scale retinex (MSR) to successfully remove effect image while enhancing details. First, estimate component by adaptive smoothing get just-noticeable difference (JND) from it Then, calculate weakening factor JND conduct MSR based on enhance Finally, perform gamma correction with weighted...
In this paper, we propose interactive image segmentation using adaptive constraint propagation (ACP), called ACP Cut. segmentation, the inputs provided by users play an important role in guiding segmentation. However, these simple often cause bias that leads to failure preserving object boundaries. To effectively use limited information, employ for semisupervised kernel matrix learning which adaptively propagates information into whole image, while successfully keeping original data...
This paper reviews the first NTIRE challenge on perceptual image enhancement with focus proposed solutions and results. The participating teams were solving a real-world photo problem, where goal was to map low-quality photos from iPhone 3GS device same captured Canon 70D DSLR camera. considered problem embraced number of computer vision subtasks, such as denoising, resolution sharpness enhancement, color/contrast/exposure adjustment, etc. target metric used in this combined PSNR SSIM scores...
In this paper, we propose deep video compression with hyperprior-based entropy coding, named HDVC. The proposed method is based on the (DVC) framework that replaces traditional block-based end-to-end learning, aiming to improve efficiency and reduce computational complexity while maintaining visual quality. Based DVC framework, introduce coding into motion optimize vector estimation (i.e. optical flow estimation) using window attention fast residual channel attention. Moreover, intermediate...
When people take a picture through glass, the scene behind glass is often interfered by specular reflection. Due to relatively easy implementation, most studies have tried recover transmitted from multiple images rather than single image. However, use of not practical for common users in real situations due critical shooting conditions. In this paper, we propose image reflection removal using convolutional neural networks. We provide ghosting model that causes effects captured images. First,...
Low-light images are seriously corrupted by noise due to the low signal-to-noise ratio. In intensity, just-noticeable-difference (JND) is high, and thus not perceived well human eyes. However, after contrast enhancement, becomes obvious severe, because JND decreases as intensity increases. Thus, enhancement without considering visual perception causes serious amplification in low-light images. this paper, we propose perceptual of based on two-step suppression. We adopt suppression...
Near infrared (NIR) images have clear textures but do not contain color. In this paper, we propose NIR to RGB domain translation using asymmetric cycle generative adversarial networks (ACGANs). The image (3 channels) has richer information than the (1 channel), which makes NIR-RGB in information. We adopt GANs that different network capacities according direction. combine UNet and ResNet generator use feature pyramid (FPNs) discriminator. With help of a 128 × large receptive field, capture...
Existing contrast enhancement techniques, such as histogram equalization (HE) and optimal contrast-tone mapping (OCTM), construct a pixel function based on probability. However, they often allocate large dynamic range to smooth areas, thus preventing the allocation of resources in some regions that people are more interested in. To deal with this problem, we propose optimized perceptual tone (OPTM), which performs images considering human visual attention. First, saliency attention...
In this paper, we propose semi-supervised bi-dictionary learning for image classification with smooth representation-based label propagation (SRLP). Natural images contain complex contents of multiple objects complicated background, clutter, and occlusions, which prevents features from belonging to a specific category. Therefore, employ reconstruction-based implement discriminative dictionary in probabilistic manner. We jointly learn called anchor the feature space its corresponding soft...