A5101903127- Image and Signal Denoising Methods
- Remote-Sensing Image Classification
- Advanced Image and Video Retrieval Techniques
- Video Surveillance and Tracking Methods
- Human Pose and Action Recognition
- Advanced Image Processing Techniques
- Sparse and Compressive Sensing Techniques
- Advanced Image Fusion Techniques
- Medical Image Segmentation Techniques
- Advanced Neural Network Applications
- Frequency Control in Power Systems
- Numerical methods in inverse problems
- Human Motion and Animation
- Infrared Target Detection Methodologies
- Face recognition and analysis
- Noise Effects and Management
- Remote Sensing and Land Use
- Multimodal Machine Learning Applications
- Advanced Measurement and Detection Methods
- Statistical and numerical algorithms
- Industrial Vision Systems and Defect Detection
- Gait Recognition and Analysis
- Image Retrieval and Classification Techniques
- Advanced Vision and Imaging
- Biometric Identification and Security
Beijing Normal University
2018-2024
Hoya (Japan)
2023
In this paper, the traditional model based variational methods and deep learning algorithms are naturally integrated to address mixed noise removal, specially for Gaussian mixture Gaussian-impulse removal problem. To be different from single type (e.g. Gaussian) it is a challenge problem accurately discriminate types levels each pixel. We propose method iteratively estimate parameters, then algorithm can automatically classify according statistical parameters. The proposed separated into...
Image hashing has been widely used in image retrieval tasks. Many existing methods generate codes based on feature representations. They rarely consider the rich information such as clustering contained set well uncertain relationships between images and tags simultaneously. In this paper, we develop a Weighted Generative Adversarial Networks (WeGAN) to transfer of construct code. WeGAN consists three modules: 1) learning process for transferring knowledge single images; 2) by means codes,...
Non-Gaussian residual error and noise are common in the real applications, they can be efficiently addressed by some non-quadratic fidelity terms classic variational method. However, have not been well integrated to architectures design convolution neural networks (CNN) based image denoising In this paper, we propose a deep learning approach handle non-Gaussian error. Our method is developed on an universal approximation property for probability density functions of error/noise. By...
Human motion transfer aims to motions from a target dynamic person source static one for synthesis. An accurate matching between the and in both large subtle changes is vital improving transferred quality. In this paper, we propose MotionFormer, hierarchical ViT framework that leverages global local perceptions capture matching, respectively. It consists of two encoders extract input features (i.e., image human image) decoder with several cascaded blocks feature transfer. each block, set as...
Image segmentation is a fundamental research topic in image processing and computer vision. In recent decades, researchers developed large number of algorithms for various applications. Among these algorithms, the normalized cut (Ncut) method widely applied due to its good performance. The Ncut model an optimization problem whose energy defined on specifically designed graph. Thus, results existing are largely dependent preconstructed similarity measure graph since this usually given...
Due to the large intraclass variances and complicated object distribution, recognizing objects with complex appearances arbitrary orientations has been an active research topic a challenging task in remote sensing fields. In this article, we formulate recognition as high-level feature-learning problem, novel supervised method is proposed learn feature representations from high-resolution images for recognition. Our simultaneously coherently achieves learning classifier training, which...
Distance metric learning aims to learn from the given training data a valid distance metric, with which similarity between samples can be more effectively evaluated for classification. Metric is often formulated as convex or nonconvex optimization problem, while many existing algorithms become inefficient large scale problems. In this paper, we formulate kernel classification and solve it by iterated of support vector machines (SVM). The new formulation easy implement, efficient in training,...
Morphological methods play a crucial role in remote sensing image processing, due to their ability capture and preserve small structural details. However, most of the existing deep learning models for semantic segmentation are based on encoder-decoder architecture including U-net Segment Anything Model (SAM), where downsampling process tends discard fine In this paper, we propose new approach that integrates learnable morphological skeleton prior into neural networks using variational...
Abstract In the inverse problem of image processing, we have witnessed that non-convex and non-smooth regularizers can produce clearer edges than convex ones such as total variation (TV). This fact be explained by uniform lower bound theory local gradient in regularization. recent years, although it has been numerically shown nonlocal various patches based methods recover textures well, still desire a theoretical interpretation. To this end, propose block regularization model on patches. By...
Ear recognition task is known as predicting whether two ear images belong to the same person or not. In this paper, we present a novel metric learning method for recognition. This formulated pairwise constrained optimization problem. each training cycle, selects nearest similar and dissimilar neighbors of sample construct constraints, then solve problem by iterated Bregman projections. Experiments are conducted on AMI, USTB II WPUT databases. The results show that proposed approach can...
Image segmentation is a fundamental research topic in image processing and computer vision. In the last decades, researchers developed large number of algorithms for various applications. Amongst these algorithms, Normalized cut (Ncut) method widely applied due to its good performance. The Ncut model an optimization problem whose energy defined on specifically designed graph. Thus, results existing are largely dependent pre-constructed similarity measure graph since this usually given...
We present XFormer, a novel human mesh and motion capture method that achieves real-time performance on consumer CPUs given only monocular images as input. The proposed network architecture contains two branches: keypoint branch estimates 3D vertices 2D keypoints, an image makes predictions directly from the RGB features. At core of our is cross-modal transformer block allows information to flow across these branches by modeling attention between coordinates spatial Our smartly designed,...
We present XFormer, a novel human mesh and motion capture method that achieves real-time performance on consumer CPUs given only monocular images as input. The proposed network architecture contains two branches: keypoint branch estimates 3D vertices 2D keypoints, an image makes prediction directly from the RGB features. At core of our is cross-modal transformer block allows information flow across these branches by modeling attention between coordinates spatial Our smartly designed, which...
In this paper, the traditional model based variational method and learning algorithms are naturally integrated to address mixed noise removal problem. To be different from single type (e.g. Gaussian) removal, it is a challenge problem accurately discriminate types levels for each pixel. We propose iteratively estimate parameters, then algorithm can automatically classify according statistical parameters. The proposed separated into regularization, synthesis, parameter estimation...