Article
Computer Science, Artificial Intelligence
Yechen Wang, Bin Song, Zhiyong Zhang
Summary: This paper proposes an image inpainting method based on GAN inversion and autoencoder. The method learns the mapping from noise to low-dimensional feature maps using a generator in an autoencoder-based GAN, and then converts the feature maps into high-resolution images. Experimental results show that the proposed method is more suitable for high-resolution image inpainting and performs better in inpainting large-range damaged images.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Huitong Yang, Liang Lei, Haiwei Sang
Summary: In this paper, a novel stereo network GAMNet is proposed to enhance the accuracy of stereo-based disparity estimation. The network consists of a lightweight attention module, an MPF module, and a stacked encoder-decoder with the DCA module and 3D convolutions. Experimental results demonstrate that GAMNet outperforms previous methods and a new 3D scene reconstruction evaluation strategy is proposed.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Jesus Gutierrez, Sergio Martin, Victor Rodriguez
Summary: Fall detection systems use various technologies, including artificial vision with artificial neural networks, to determine whether a fall has occurred. This study proposes a new approach by introducing human dynamic stability descriptors to overcome the limitations of using kinematic descriptors. The results show promising potential for better performance in detecting falls among elderly people.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Juan Wang, Zetao Zhang, Minghu Wu, Yonggang Ye, Sheng Wang, Ye Cao, Hao Yang
Summary: In this work, an improved BlendMask nuclei instance segmentation framework is proposed, which incorporates dilated convolution aggregation module and context information aggregation module to enhance the performance of detecting and segmenting dense small objects and adhering nuclei. A distributional ranking loss function is also introduced to alleviate the imbalance between the target and the background. The proposed method outperforms several recent classic open-source nuclei instance segmentation methods on the DSB2018 dataset, achieving a 3.6% improvement on AP segmentation metric compared to BlendMask.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Shaofan Wang, Yukun Liu, Yanfeng Sun, Baocai Yin
Summary: This article proposes a new method called Shuffling Atrous Convolutional U-Net (SACNet) to address two issues in medical image segmentation: disrupted distribution of essential feature of objects and blurred object boundaries. SACNet solves these problems by using the Shuffling Atrous Convolution (SAC) module to merge different atrous convolutional layers, and outperforms other methods in three medical image segmentation tasks.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Shuo Zhu, Yu Wang, Zongyang Wang
Summary: This paper proposes an improved lightweight YOLOv5s model for license plate detection, using various strategies to improve the accuracy and speed of recognition. Experimental results on the CCPD dataset demonstrate that the model achieves excellent performance and is an effective approach for license plate detection.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Yaxin Ji, Lan Di
Summary: This paper presents a textile defect detection method that utilizes a multi-proportion spatial attention mechanism and channel memory feature fusion network. It addresses the difficulties presented by complicated shapes and large size variations, resulting in improved detection accuracy.
IET IMAGE PROCESSING
(2023)
Review
Computer Science, Artificial Intelligence
Qiulang Ji, Jihong Wang, Caifu Ding, Yuhang Wang, Wen Zhou, Zijie Liu, Chen Yang
Summary: This paper proposes a Dual-path Multi-scale Attention Guided network (DMAGNet) for medical image segmentation. By introducing Dual-path Multi-scale Attention Fusion Module (DMAF) and Multi-scale Normalized Channel Attention Module (MNCA), accurate segmentation of pathological regions is achieved. Experimental results demonstrate that DMAGNet outperforms the original U-Net method and other advanced methods in brain, lung, and liver segmentation tasks.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Yi-Peng Liu, Dongxu Zeng, Zhanqing Li, Peng Chen, Ronghua Liang
Summary: Retinal vessel segmentation is a crucial task for eye retinopathy diagnosis in computer vision. The distribution deviation between the source and target datasets often leads to inaccurate segmentation results and hampers the generalization ability to unseen domains, impacting subsequent disease diagnosis. To address this, the authors propose the spectral-spatial normalization (SS-Norm) module, which decomposes features into frequency components using discrete cosine transform (DCT) and analyzes their semantic contribution. By reweighting the frequency components, the authors normalize the distribution in the spectral domain, enhancing the model's generalization ability. Extensive experiments on six datasets validate the effectiveness of the proposed method.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Cheng Yang, Guanming Lu
Summary: This study proposes an unsupervised image-blind super-resolution method called RDFL, which addresses the challenge of accurately replicating the complexity of real-world image degradation. The method utilizes degradation feature-based learning and a Transformer-based SR network to achieve high-quality image reconstruction.
IET COMPUTER VISION
(2023)
Article
Computer Science, Artificial Intelligence
Longbao Wang, Yican Shen, Jin Yang, Hui Zeng, Hongmin Gao
Summary: This paper proposes a rotated object detection method for remote sensing images based on deformable convolution, named rotated points. The method uses deformable convolution mapping to the rotated object bounding box and uses a sample assignment scheme to associate object classification and localization performance, which avoids the problems of angle loss discontinuity and parameter regression inconsistency common in rotated object detection methods.
IET IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Xun Zhang, Jingxian Liu, Yalu Zheng, Yan Zheng, Masroor Hussain
Summary: Most shape classification methods are based on a single closed contour, but practical shapes often have complex contours. This research proposes a novel method that encodes complex shapes into an angle code and a sparsity code, and then uses a 1-D CNN for feature extraction and classification. Experiments on two datasets demonstrate the superior classification accuracy of this method. The proposed method outperforms compared methods and each module in the network contributes significantly to the accuracy.
IET IMAGE PROCESSING
(2023)
Article
Engineering, Electrical & Electronic
Chaojun Shi, Shiwei Zhao, Ke Zhang, Xiaohan Feng
Summary: Face-based age estimation relies heavily on ResNets but ignores the importance of large-scale facial information and other age attributes. This paper proposes a multi-task learning framework called multi-task multi-scale attention, which incorporates attention mechanism. The framework includes a multi-scale attention module to extract local age-sensitive information and multi-scale features, and uses multi-task learning to predict gender and race. The proposed method achieves superior performance compared to other state-of-the-art methods on the MORPH Album II and Adience datasets.
IET SIGNAL PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Aizhong Mi, Mingming Gao, Zhanqiang Huo, Yingxu Qiao, Jian Chen, Haiyang Jia
Summary: This article proposes a real-time semantic segmentation method based on BiSeNet V2, which includes a lightweight semantics recalibration module and a detail enhancement module, capable of extracting global semantic contextual information and enhancing detail information effectively.
IET COMPUTER VISION
(2023)
Article
Computer Science, Artificial Intelligence
Cheng Han, Yongqing Cai, Xinpeng Pan, Ziyun Wang
Summary: This paper proposes an efficient depth estimation fusion module to balance the feature mapping of equirectangular and cubemap projections. It also introduces a novel inflated network architecture for equirectangular projection to extend the receptive field. Extensive experiments show that the proposed method predicts clear boundaries and accurate depth results, outperforming mainstream panoramic depth estimation algorithms.
IET IMAGE PROCESSING
(2023)