计算机与数字工程2025，Vol.53Issue(2)：523-527,550,6.DOI:10.3969/j.issn.1672-9722.2025.02.039
SLKT:一种高精度图像语义预分割方法
SLKT:A High Precision Image Semantic Pre-segmentation Method
摘要
Abstract
The superpixel cluster segmentation method is often used as a preprocessing method for many deep learning mod-els,but the oversimplification of the algorithm structure not only reduces the accuracy of image pre-segmentation but also reduces the accuracy of the model in subsequent semantic segmentation tasks.To deal with this problem,a new region similarity measure and merging criterion are designed based on the general superpixel segmentation method SLIC.At the same time,the K-means algo-rithm,which integrates the sample density of the DPC algorithm and the idea of a three-way cluster delay decision,is used as the re-clustering method after the initial segmentation.The experimental results show that the new similarity measure and merging crite-rion can fully consider the spatial characteristics of the image,and the reclustering of the segmented pixels at the SLIC can further improve the accuracy of pre-segmentation and the actual segmentation score.关键词
超像素分割/样本密度/三支聚类/语义分割/空间特征Key words
superpixel segmentation/sample density/three-way clustering/semantic segmentation/spatial features分类
信息技术与安全科学引用本文复制引用
程柳,祁云嵩,刘杰..SLKT:一种高精度图像语义预分割方法[J].计算机与数字工程,2025,53(2):523-527,550,6.基金项目
国家自然科学基金项目(编号:61471182)资助. （编号:61471182）
