Peixuan Liu

Peixuan Liu

Auto-generated profile, based on publicly available information

Peixuan Liu

School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, China

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

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

Representative Publications

View All Publications
21 July 2025 in IEEE Journal of Translational Engineering in Health and Medicine
Verified
Journal Article
Open access
SZShaojie Zhang
Shaojie Zhang
Ying-Hui WangYing-Hui Wang
Ying-Hui Wang
PLPeixuan Liu
Peixuan Liu
WLWei Li
Wei Li
JYJinlong Yang
Jinlong Yang
IAIbragim Atadjanov
Ibragim Atadjanov
Mirror reflections and highlights are commonly observed in images captured by wireless capsule endoscopy (WCE), leading to image blurring or distortion, which increases the difficulty for doctors to observe and analyze gastrointestinal images. Existing methods fail to satisfactorily address specular highlights in the gastrointestinal tract, often resulting in loss of image details, texture blurring, or texture continuity errors. Therefore, we propose a highlight removal method for capsule endoscopy images. This approach enhances confidence term calculation in pixel prioritization by leveraging the ratio characteristics between the R channel and the B channel in the RGB color space of WCE images, enabling the evaluation of the repair possibility for each pixel. Furthermore, by dynamically adjusting the size of the sample block window based on variance and optimizing the selection of matching patches in the sample block window using RGB color channel distance and pixel distance, the accuracy of the best matching patch search is improved. Our extensive experimental results demonstrate that compared to the best-performing Criminisi method among existing traditional approaches, our method achieves reductions of 3.10% and 2.75% in standard deviation and coefficient of variation, respectively. Additionally, compared to the DeepGin method in deep learning, reductions of 4.22% and 10.50% are observed in these metrics, respectively. The regions after highlight removal and the surrounding regions near the highlights have more similar colors and continuous textures.