- Advanced Image and Video Retrieval Techniques
- Video Analysis and Summarization
- Video Surveillance and Tracking Methods
- Advanced Vision and Imaging
- AI in cancer detection
- Anomaly Detection Techniques and Applications
- Image Retrieval and Classification Techniques
- Face recognition and analysis
- Face and Expression Recognition
- Medical Image Segmentation Techniques
- Radiomics and Machine Learning in Medical Imaging
- Biometric Identification and Security
- Human Pose and Action Recognition
- Adipokines, Inflammation, and Metabolic Diseases
- COVID-19 diagnosis using AI
- Colorectal Cancer Screening and Detection
- Gastric Cancer Management and Outcomes
- Visual Attention and Saliency Detection
- Regulation of Appetite and Obesity
- Digital Imaging for Blood Diseases
- Domain Adaptation and Few-Shot Learning
- Vehicle Dynamics and Control Systems
- Image Enhancement Techniques
- Image Processing Techniques and Applications
- Advanced Neural Network Applications
National Yang Ming Chiao Tung University
2007-2025
National Chung Hsing University
2015-2025
China Medical University
2025
National Cheng Kung University
2004-2024
Hefei University of Technology
2022-2024
Anqing City Hospital
2024
Ministry of Ecology and Environment
2024
Sun Yat-sen University
2015-2022
Sun Yat-sen Memorial Hospital
2022
Jinan University
2011-2022
Video summarization aims to generate a compact summary of the original video for efficient browsing. To provide summaries which are consistent with human perception and contain important content, supervised learning-based methods proposed. These aim learn content based on continuous frame information human-created summaries. However, simultaneously considering both inter-frame correlations among non-adjacent frames intra-frame attention attracts humans importance representations rarely...
To reduce human efforts in browsing long surveillance videos, synopsis videos are proposed. Traditional video generation applying optimization on tubes is very time consuming and infeasible for real-time online generation. This dilemma significantly reduces the feasibility of practical situations. solve this problem, problem formulated as a maximum posteriori probability (MAP) estimation paper, where positions appearing frames objects chronologically rearranged real without need to know...
Phishing is a form of online identity theft associated with both social engineering and technical subterfuge major threat to information security personal privacy. Here, the authors present an effective image-based antiphishing scheme based on discriminative keypoint features in Web pages. Their invariant content descriptor, Contrast Context Histogram (CCH), computes similarity degree between suspicious authentic The results show that proposed achieves high accuracy low error rates.
Many driver monitoring systems (DMSs) have been proposed to reduce the risk of human-caused accidents. Traditional DMSs focus on detecting specific predefined abnormal driving behaviors, such as drowsiness or distracted driving, using generic models trained with data collected during driving. However, it is difficult collect sufficient representative training construct detection models, which are applicable all drivers. Consequently, this paper proposes a new personal-based hierarchical DMS...
Ghrelin has proven to be protective against sepsis-induced acute lung injury (ALI) via anti-inflammatory effects. However, its mechanisms remain poorly understood. Alveolar macrophages (AMs) play a key role in mediating inflammatory responses during ALI by secretion of cytokines and chemokines. This study was undertaken investigate whether ghrelin suppresses effects AMs therefore may help attenuate ALI. A sepsis model rats achieved using cecal ligation puncture. treatment markedly improved...
To provide semantic image style transfer results which are consistent with human perception, transferring styles of regions the to their corresponding content is necessary. However, when object categories between and images not same, it difficult match two for transfer. solve matching problem guide based on matched regions, we propose a novel context-aware method by performing context followed hierarchical local-to-global network architecture. The aims obtain using correlations different...
We present a trajectory-based approach to detect salient regions in videos by dominant camera motion removal. Our is designed general way so that it can be applied taken either stationary or moving cameras without any prior information. Moreover, multiple of different temporal lengths also detected. To this end, we extract set spatially and temporally coherent trajectories keypoints video. Then, velocity acceleration entropies are proposed represent the trajectories. In way, long-term object...
We investigated whether analysis of endoscopic images using a refined feature selection with neural network (RFSNN) technique could predict Helicobacter pylori-related gastric histological features.A total 104 dyspeptic patients were prospectively enrolled for panendoscopy and biopsy evaluation the updated Sydney system. The each patient analyzed to obtain 84 image parameters. significant parameters from 30 randomly selected (15 15 without H. pylori infection) associated features used...
Recently, most background modeling approaches represent distributions of changes by using parametric models such as Gaussian mixture models. Because significant illumination and dynamic moving backgrounds with time, variations are hard to be modeled Moreover, how efficiently effectively update parameters reflect remains a problem. In this paper, we propose novel coarse-to-fine detection theory algorithm extract foreground objects on the basis nonparametric represented binary descriptors. We...
Shot change detection is an essential step in video content analysis. However, automatic shot often suffers from high false rates due to camera or object movements. To solve this problem, we propose approach based on local keypoint matching of frames. This aims detect both abrupt and gradual transitions between shots without modeling different kinds transitions. Our experiment results show that the proposed algorithm effective for most changes.
This study presents a computer-aided diagnosis system using sequential forward floating selection (SFFS) with support vector machine (SVM) to diagnose gastric histology of <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Helicobacter</i> xmlns:xlink="http://www.w3.org/1999/xlink">pylori</i> ( xmlns:xlink="http://www.w3.org/1999/xlink">H.</i> ) from endoscopic images. To achieve this goal, candidate image features associated clinical symptoms are...
In this paper, a video re-coloring algorithm for dichromats is presented. Different from image schemes, reproducing requires maintaining temporal color consistency between frames, i.e., the same in different frames should be re-colored to identical new color. To achieve goal, we extract key colors shots after motion estimation at first. Based on importance of colors, process order defined perform efficient remapping and solve contrast problem. Then, remapped frame pixel values are...
A new computer-aided diagnosis method is proposed to diagnose the gastroesophageal reflux disease (GERD) from endoscopic images of esophageal-gastric junction. To avoid interferences different endoscope devices and automatic camera white balance adjustment, heterogeneous descriptors computed color models are used represent images. Instead concatenating these a super vector, hierarchical descriptor fusion support vector machine (HHDF-SVM) framework simultaneously apply for GERD overcome curse...
Histopathological images provide the medical evidences to help disease diagnosis. However, pathologists are not always available or overloaded by work. Moreover, variations of pathological with respect different organs, cell sizes and magnification factors lead difficulty developing a general method solve histopathological image classification problems. To address these issues, we propose novel cross-scale fusion (CSF) transformer which consists multiple field-of-view patch embedding module,...
Rough face alignments lead to suboptimal performance of identification systems. In this study, we present a novel approach for identifying genders from facial images without proper alignments. Instead using only one input test, generate an image set by randomly cropping out patches neighborhood the detection region. Each is represented as subspace and compared with other sets measuring canonical correlation between two associated subspaces. By finding optimal discriminative transformation...
In order to continuously keep sight of environments, taking long-time surveillance videos is required. Thus, enormous storage and video browsing become important problems. To solve these problems, we propose an online synopsis method rearrange positions appearing frames foreground objects chronologically in real-time. Comparing with traditional approaches, our can directly be applied streaming generate without pre-screening entire during the processing.