A5020917892- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Visual Attention and Saliency Detection
- Image Enhancement Techniques
- Video Analysis and Summarization
- Generative Adversarial Networks and Image Synthesis
- Human Motion and Animation
- Face recognition and analysis
- Computer Graphics and Visualization Techniques
- Medical Image Segmentation Techniques
- Human Pose and Action Recognition
- Image and Signal Denoising Methods
- Advanced Vision and Imaging
- Advanced Image Processing Techniques
- Advanced Neural Network Applications
- Face Recognition and Perception
- Advanced Image Fusion Techniques
- Simulation and Modeling Applications
- Algorithms and Data Compression
- Distributed and Parallel Computing Systems
- Evacuation and Crowd Dynamics
- Video Surveillance and Tracking Methods
- Distributed systems and fault tolerance
- Hand Gesture Recognition Systems
- Ocular and Laser Science Research
Dalian Minzu University
2011-2024
Minzu University of China
2021-2024
Lappeenranta-Lahti University of Technology
2024
Bournemouth University
2024
Dalian University of Technology
2021
Zhejiang University
2009-2013
In this paper, we propose a novel form of weak supervision for salient object detection (SOD) based on saliency bounding boxes, which are minimum rectangular boxes enclosing the objects. Based idea, weakly-supervised SOD method, by predicting pixel-level pseudo ground truth maps from just boxes. Our method first takes advantage unsupervised methods to generate initial and addresses over/under prediction problems, obtain maps. We then iteratively refine learning multi-task map refinement...
Most existing salient object detection (SOD) methods are designed for RGB images and do not take advantage of the abundant information provided by light fields. Hence, they may fail to detect objects complex structures delineate their boundaries. Although some have explored multi-view field saliency detection, require tedious pixel-level manual annotations ground truths. In this paper, we propose a novel weakly-supervised learning framework on based bounding box annotations. Our method has...
Abstract In this article, we propose a method to jointly control face image generation through semantic segmentation maps and text. Existing lack detailed attributes such as beards, it is difficult explicitly represent the gender of target person by virtue maps. State‐of‐the‐art methods guided mostly solved introducing original for supervision, which cannot accurately face. At same time, text‐guided perform poorly in controlling front side pose. Therefore, an idea that map controls coarse...
Abstract Makeup in real life varies widely and is personalized, presenting a key challenge makeup transfer. Most previous transfer techniques divide the face into distinct regions for color transfer, frequently neglecting details like eyeshadow facial contours. Given successful advancements of Transformers various visual tasks, we believe that this technology holds large potential addressing pose, expression, occlusion differences. To explore this, propose novel pipeline which combines...
Due to the popularity of motion capture data in many applications, such as games, movies and virtual environments, huge collections are now available. It is becoming important store these compressed form while being able retrieve them without much overhead. However, there little work that addresses both issues together. In this paper, we address two by proposing a novel database architecture. First, propose lossless compression algorithm compress clips, which based on Alpha Parallelogram...
Abstract Most existing makeup transfer techniques focus on light styles, which limits the task of to color manipulation issues such as eye shadow and lip gloss. However, in real life is diverse personalized, not only most basic foundation, makeup, but also painted patterns face, jewelry decoration other personalized makeup. Inspired by painting steps drawing outline first then coloring, we propose a network for realizes face learning correspondence. Specifically, feature extraction module...
Currently, most image retrieval systems use either purely visual features or textual metadata associated with images. They have advantages and disadvantages respectively. To overcome their drawbacks improve the performance without sacrificing efficiency, we propose stepwise refinement multimodal scheme based on annotation keywords content, which can benefit from strength of text- content-based retrieval. The system starts query triggered by some keywords, further refines result blobs regions...
Image retrieval based on region is one of the most promising and active research directions in recent year's CBIR, while segmentation , feature selection extraction are key issue s . However, existing approaches always ad o pt a uniform approach for all images same system. In this paper, we propose adaptive image according to different category To improve performance, image. Textured segmented by Gaussian Mixture Models (GMM), non-textured our proposed block-based normalized cut. accurately...
Automatic image annotation is a promising solution to narrow the semantic gap between low-level content and high-level concept, which has been an active research area in fields of retrieval, pattern recognition, machine learning. However, even most dedicated algorithms are often unsatisfactory. Image refinement attracted much more attention recently. In this paper, novel algorithm using dynamic voting based on mutual information proposed. Unlike traditional algorithm, proposed adopts...
Abstract In this paper, we propose a few-shot method for pose transfer of anime characters—given source image an character and target pose, the to character. Despite recent advances in on real people images, these methods typically require large numbers training images different person under poses achieve reasonable results. However, are expensive obtain they created with lot artistic authoring. To address this, meta-learning framework transfer, which can well generalize unseen given just...
Abstract In this paper, we address RGB‐D salient object detection task by jointly leveraging semantics and contour details of objects. We propose a novel semantics‐and‐details complementary fusion network to adaptively integrate cross‐model multilevel features. Specifically, employ two kinds modules in our model, which are designed for fusing high‐level semantic features integrating detail the scene components, respectively. The module aggregates interdependent relationships nonlinear...
Good trimap is essential for high-quality alpha matte. However, making hardwork, especially complex images. In this paper, an active learning framework proposed to make high quality trimap. There are two methods which employed: minimization of uncertainty sampling (MUS) and maximization expected model output change (EMOC). MUS finds the informative area in image can decrease uncertain EMOC important areas give maximum Two combined define map. Active map shows image. It help users The...