MaLissa (Matched Lissajous CLE)
收藏资源简介:
MaLissa是由浦项工科大学与VPIX Medical联合创建的首个开放基准数据集,专注于高帧率Lissajous共聚焦激光内窥镜(CLE)的图像修复研究。该数据集包含2,060组配对的低质量视频帧与高质量参考图像,数据来源于猪胃组织样本,采用1024×1024分辨率采集,其中低质量帧以10Hz频率采集导致70%以上像素缺失,而高质量帧以2Hz频率采集并通过双线性插值补全。数据集通过相位相关配准算法将稀疏帧与拼接后的宽视场高质量马赛克图像对齐,旨在解决临床内窥镜场景下运动伪影和像素稀疏性问题,为实时活体光学活检提供关键训练资源。
MaLissa is the first open-access benchmark dataset jointly developed by Pohang University of Science and Technology (POSTECH) and VPIX Medical, focusing on image restoration research for high-frame-rate Lissajous confocal laser endomicroscopy (CLE). The dataset comprises 2,060 paired low-quality video frames and corresponding high-quality reference images, acquired from porcine gastric tissue samples at a resolution of 1024×1024. Specifically, the low-quality frames are captured at 10 Hz, leading to over 70% pixel loss, while the high-quality frames are collected at 2 Hz and completed via bilinear interpolation. The dataset aligns the sparse frames with the stitched wide-field high-quality mosaic images using a phase-correlation registration algorithm, aiming to address the issues of motion artifacts and pixel sparsity in clinical endoscopy scenarios, and serve as a critical training resource for real-time in vivo optical biopsy.



