Xiu He

Dr. Xiu He

Macau University of Science and Technology, Macau, Macao
orcid

Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
20 June 2023 in IEEE Access
Verified
Journal Article
Open access
XHXiu He
Xiu He
YCYUYUN CHEN
YUYUN CHEN
YFYong Fan
Yong Fan
XKXiaoxi Kong
Xiaoxi Kong
Zhanchuan CaiZhanchuan Cai
Zhanchuan Cai
The single type of multimedia data such as text, image, video, audio, etc., has been studied extensively in the past decades. With the advances in multimedia applications and services, the exponential growth of multimedia data has emerged. How to effectively represent them and capture the relationships among multimedia data are of crucial importance in the study of multimedia applications. In this paper, we propose a novel reversible multimedia representation method based on complex base. In the proposed multimedia representation method, we first process different modalities data such as text, image, and audio separately and convert them into different Gaussian integers so that they are on the same representation plane, resulting in a joint multimedia representation model. Therefore, the relationships of multimedia data are established by using the geometric operations on the same representation plane. Experimental results show that the proposed multimedia representation method is lossless and reversible. In addition, we explore the potential applications of the proposed representation method in multimedia processing, multimedia camouflage and multimedia secret sharing schemes verify that the proposed representation method has good practical applicability in multimedia security.

2 November 2023 in IEEE Access
Verified
Journal Article
Open access
YFYong Fan
Yong Fan
XHXiu He
Xiu He
YLYiyi Lin
Yiyi Lin
Zhanchuan CaiZhanchuan Cai
Zhanchuan Cai
With the development of industrial informatization, video processing technology is receiving more and more attention. Extracting background is a prerequisite for many video processing techniques, so video background modeling technology is becoming highly sought-after. Currently, there are a variety of approaches to estimating background; however, many of these methods have the fault of not being able to accurately distinguish between foreground and background, especially when objects move slowly or remain still for a period of time. In this paper, a novel background modeling scheme is proposed for surveillance video, based on keyframe and particle shape properties. The model consists of three parts: the first part is to reduce running time and eliminate the ghost phenomenon caused by adjacent redundant frames by extracting keyframe and dividing the extracted frames into several groups; the second part includes three steps, computing binarized difference, characterizing the binarized difference and screening the difference where a quadruple, composed of particle shape properties, is designed to quantitatively describe binarized differences; the third part involves generating the temporary background and updating the temporary background according to the similarity of data obtained at the newly proposed locations. Experiment results on SBMnet and SBI datasets and comparisons with some emerging algorithms show that the performance of the proposed model is superior or comparable to the other state-of-the-art methods particularly when dealing with stationary objects. Furthermore, the proposed method ranks as the second for intermittent category video, compared to the other 31 state-of-art methods. Moreover, the speed of the proposed method, 5.38 Frame Per Second ( FPS ) for SBMnet dataset and 10.94 FPS for SBI dataset, is faster than most public methods.

4 September 2024 in IEEE Access
Verified
Journal Article
Open access
GCGang Chen
Gang Chen
QFQing Feng
Qing Feng
XHXiu He
Xiu He
Jian YaoJian Yao
Jian Yao
Inpainting images that have sizable missing blocks presents a considerable challenge in terms of preserving visual consistency and attaining a convincing result. In this study, the multi-scale generative adversarial network with multi-head external attention for image inpainting (denoted as MGDAN) is proposed. First, an adaptive weight style loss is designed into the generator of multi-scale generative adversarial networks (GAN) to guide the inpainting of the style and structure in the image inpainting, which improves the inpainting effect of image contour and emoticon. Second, the Adaptive Mix model is introduced to address the imbalance in generator and discriminator training by reducing the distance between the difficult and easy samples. This approach is intended to improve the overall performance of the network and drive high-quality image generation. Third, the generator and local discriminator in the image inpainting process are enhanced by the introduction of a multi-head external attention mechanism. This addition aimed to capture the long-distance and multi-level dependent relationships between different areas of the image. It is proved beneficial for the generating clear geometric contours in the restored images and improving the overall global consistency of the inpainted results. Using the experimental methods, we find the optimal number of heads. The effectiveness of the proposed approach also is demonstrated through the extensive experiments conducted on public datasets.

16 May 2024 in IEEE Transactions on Industrial Informatics
Verified
Journal Article
Closed access
XHXiu He
Xiu He
Zhanchuan CaiZhanchuan Cai
Zhanchuan Cai
Audio-visual products are an important carrier of ideological communication, which meets the diverse, multiaspect spiritual and cultural needs. However, although current audio representation methods have shown superior performance, they have some limitations and cannot fully take advantage of the diverse, multilevel characteristics of the audio-visual products. The article proposes a complex system-based audio representation method, which diverse audio signals are represented uniformly as Gaussian integers and marked into the same 2-D complex plane. Especially, the relationships of audio signals can be redefined based on the geometric properties. The experiments demonstrate that the proposed audio signal representation method is feasible and lossless. Furthermore, we explore potential industrial information security applications under the proposed representation method, and the applications verify that the representation method has the practical value.

19 January 2024
Verified
Conference Paper
Open access
FDFeng Ding
Feng Ding
HCHuahua Cao
Huahua Cao
XHXiu He
Xiu He
Re Identification (ReID) is an important research direction in the field of computer vision, aimed at identifying the same pedestrian from different camera perspectives. To address the issue of pedestrian re identification, this paper proposes a pedestrian re identification method based on deep features and transfer learning. This article proposes the use of convolutional neural networks to learn the deep features of pedestrians, which can retain more valuable information and calculate the similarity of each image pair using a simple and effective cosine distance method. Our method can be applied to different scenarios, and experiments have been conducted on different datasets. This article provides detailed validation and comparison of the proposed method. The experimental results show that the pedestrian re recognition method based on deep features and transfer learning can effectively improve the accuracy and robustness of pedestrian re recognition. Specifically, this method has achieved good performance on multiple publicly available datasets, significantly outperforming traditional methods.