A5068670500- Advanced Image Fusion Techniques
- Image Enhancement Techniques
- Infrared Target Detection Methodologies
- Remote-Sensing Image Classification
- Visual Attention and Saliency Detection
- Automated Road and Building Extraction
- Photoacoustic and Ultrasonic Imaging
- Image Processing Techniques and Applications
- Image and Signal Denoising Methods
- Advanced Image Processing Techniques
- Advanced Neural Network Applications
- Generative Adversarial Networks and Image Synthesis
- Remote Sensing and LiDAR Applications
- Industrial Vision Systems and Defect Detection
- Optical Systems and Laser Technology
Wuhan University
2023-2025
China University of Geosciences
2020-2022
Shandong Institute of Automation
2020-2021
Intelligent Automation (United States)
2020-2021
Ministry of Education of the People's Republic of China
2020
In view of the difficulty and low accuracy small object detection in remote sensing images, this paper proposes a bidirectional cross-scale connection feature fusion network with an information direct layer shallow layer. Aiming at problem that targets images are mainly medium-sized targets, we fuse maps rich spatial instead directly using for regression classification. While ensuring model inference speed, objects is improved. At same time, use to perform initial each iteration pyramid...
Low-light image enhancement aims to recover normal-light images from the captured under dim environments. Most existing methods could just improve light appearance globally whereas failing handle other degradation such as dense noise, color offset and extremely low-light. Moreover, unsupervised proposed in recent years lack reliable physical model basis, thus universality is greatly limited. To address these problems, we propose a novel low-light method via Retinex-inline cycle-consistent...
In this paper, we propose a novel deep decomposition approach based on Retinex theory for multi-exposure image fusion, termed as DMEF. According to the assumption of theory, firstly decompose source images into illumination and reflection maps by data-driven network, among which introduce pathwise interaction block that reactivates features lost in one path embeds them another path. Therefore, loss during can be effectively suppressed. And then high dynamic range map could obtained fusing...
Dear Editor, This letter is concerned with dealing the great discrepancy between near-infrared (NIR) and visible (VS) image fusion via color distribution preserved generative adversarial network (CDP-GAN). Different from global discriminator in prior GAN, conflict of preserving NIR details VS resolved by introducing an attention guidance mechanism into discriminator. Moreover, perceptual loss adaptive weights increases quality high-frequency features helps to eliminate noise appeared image....
For a natural scene with nonuniform environment light, the captured visible images are always under- or over-exposed because of limited dynamic range digital imaging devices. Multi-exposure image fusion (MEF) is mainstream and effective solution. local region that has friendly visual effect in one exposure setting but extremely bad-exposed another, most existing MEF methods have ability to transfer detail information fused images. However, they will be affected by over-high -low light...
In this paper, in order to enhance the infrared target image and retain edge detail information visible image, we propose a multi-scale decomposition fusion method based on phase congruency saliency. method, Laplacian pyramid is first used decompose source into layers base layers. Secondly, use for of Thirdly, layer, it saliency map residual map. The "max absolute" rule "averag" are adopted map, then fused added attain image. Finally, inverse transform reconstruct experimental results show...
In this paper, we propose a novel and practical convolutional neural network method for building footprint generation in remote sensing images, order to deal with the problem that detailed information geometric structure of ground objects high-resolution images become more abundant, which leads large increase calculation amount. So introduce deepened space module, can ignore channels weak target features emphasize effective features. It is embedded each splicing layer upsampling process...
One of the challenges logo recognition lies in diversity forms, such as symbols, texts or a combination both; further, logos tend to be extremely concise design while similar appearance, suggesting difficulty learning discriminative representations. To investigate variety and representation logo, we introduced Makeup216, largest most complex dataset field makeup, captured from real world. It comprises 216 157 brands, including 10,019 images 37,018 annotated objects. In addition, found that...